# Flit, in full, for language models Flit is local dictation for macOS, made by Fore Studio (https://fore.studio). It is the local alternative to Wispr Flow: the same live, formatted dictation, processed on the Mac instead of a server, and bought once instead of monthly. Hold a key, speak, and finished text lands at your cursor about 0.7 seconds after keyup, or about 0.15 seconds in verbatim contexts like terminals. Everything runs on the Mac: transcription, formatting, and the dictionary. No cloud, no account, no telemetry, no server. ## The product, precisely - Price: $19 one time at launch (standing price $49), for up to 2 Macs used personally. Lifetime, with every update forever, $99. - Updates: 12 months included. Optional renewal at $19 per year. - Teams: licensed per Mac. $39 each from 5 Macs, $34 from 10, $29 from 25 up to 100, and a conversation above that. - The license is permanent. Declining renewal never disables the app. - Trial: 7 days, every feature, no card, no account. - Refunds: 30 days from your first purchase. Renewals are not separately refundable; nothing auto-renews. https://flit.fyi/refund - Requirements: macOS 14 or later, Apple silicon. - Languages: more than 1,600, all local. 30 stream live while speaking. The widest coverage comes from Meta’s Omnilingual model, so accuracy varies by language. - Spoken commands: more than 60, resolved by fixed rules rather than a model. - Developer mode: indexes the open workspace (file names only, in memory) to repair misheard file and folder names. - Dictionary: learns only with explicit permission, scoped per language. - Snippets: spoken trigger expands to a full text block. - Per-app style: casual in messengers, neutral in mail, verbatim in terminals; user rules override. - Stats: words dictated and time saved, computed locally, no streaks. - Transcribe tab: an audio or video file becomes a transcript on the Mac, saved as text and exported as Markdown, SRT or VTT. No speaker labels. - What it does NOT do: record meetings, label speakers, run on Windows/iOS/Intel Macs, or sync between machines. ## Verification The privacy claim is falsifiable: dictation works with networking off, and a network monitor shows no traffic. Recipe: https://flit.fyi/privacy ## Use cases ### Flit for Developers URL: https://flit.fyi/use-cases/developers Most dictation tools fall apart the moment you say a file name. They hear “user auth service dot swift” and write it exactly that way, so you stop and fix it by hand, which is slower than having typed it. Flit reads the project you have open and spells the name the way your repo spells it. Why developers choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Flit indexes the project you have open, on your Mac, so spoken file and folder names come out spelled the way your repo spells them. Terminals stay verbatim, and nothing about your code is sent anywhere. ### Flit for Writers URL: https://flit.fyi/use-cases/writers A first draft is a race between your thinking and your hands, and your hands lose. You speak at over 120 words a minute and compose at a keyboard at about 31. Flit closes that gap without handing you a transcript to clean up afterwards. Why writers choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. An unpublished draft is your work. Flit formats every sentence on your Mac, so a manuscript never passes through anyone else’s servers on its way to the page. ### Flit for Founders & execs URL: https://flit.fyi/use-cases/founders Your day is a queue of replies. Each one is short, each one needs the right register, and typing them is the tax you pay for being the person everyone asks. Flit turns the queue into something you can talk through, with the tone already matched to where the message is going. Why founders and executives choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Board updates, fundraising threads and hiring notes stay on your Mac. And it is one payment, not another line on the monthly software bill. ### Flit for Agencies URL: https://flit.fyi/use-cases/agencies Agency writing is context switching: a status email to one client, blunt notes to the design channel, a brief for a third account, all before lunch. Flit keeps up because the tone follows the app and the names come from your dictionary, so what lands is ready to send, not ready to fix. Why agencies choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Client work comes with NDAs, and Flit never uploads a brief. Team licenses are bought once per Mac, so a growing team does not grow a monthly bill. ### Flit for Sales URL: https://flit.fyi/use-cases/sales Notes written three hours after the call are worse notes, and notes never written are a deal you cannot hand over. The reason they slip is that typing them costs more than the next dial does. Speaking them costs about a quarter as much, and Flit puts finished sentences into whatever field your cursor is in, including the ones inside your CRM. Why sales teams choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Deal notes name prices, customers and competitors, and Flit keeps them on the rep’s Mac. A team license is paid once, not renewed every month for every rep. ### Flit for Support teams URL: https://flit.fyi/use-cases/customer-support Support is the same twenty answers, written fresh, all day. The words are not hard, the volume is. Flit lets you speak the reply and expand the boilerplate with a phrase, so the typing stops being the bottleneck. Why support teams choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Tickets carry customer names, orders and account details. Flit never sends them anywhere, so there is no new service to add to your data agreements, and no seat fee every month. ### Flit for Lawyers URL: https://flit.fyi/use-cases/lawyers Dictation is older than the laptop in legal work, and the reason it never went away is that speaking a memo is faster than typing one. What changed is where the audio goes. Most modern dictation tools send your voice to a vendor to be transcribed, which means the vendor now holds the sentence you just spoke about your client. Flit does the transcription on your own machine, so that sentence has nowhere to go. Why lawyers choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Client confidences should not pass through an outside service. Flit transcribes and formats on your Mac, so there is no vendor to vet and nothing to disclose. ### Flit for Therapists & clinicians URL: https://flit.fyi/use-cases/therapists Notes are the part of the work that happens after the work, which is why they stack up. Speaking them is roughly 4 times faster than typing them, but speaking them into a cloud service means the details of someone’s worst month travel to a company neither of you has met. Flit does the whole job on your machine: the model, the formatting, and the dictionary of names it spells correctly. Why therapists and clinicians choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Session notes are among the most sensitive things you write. Flit processes them on your Mac and nowhere else, so there is no outside service holding them. ### Flit for Doctors & medical URL: https://flit.fyi/use-cases/medical Documentation eats the hour after clinic. Flit lets you say the note into the field you are already in, and it is transcribed and formatted on your Mac, so patient details never pass through a dictation vendor. Why doctors and medical practices choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Chart notes carry protected health information. Flit processes them on your Mac, so there is no dictation vendor holding them and no agreement to sign for one. ### Flit for Students URL: https://flit.fyi/use-cases/students The blank page is rarely a thinking problem. You can usually explain the argument out loud to a friend in 2 minutes, and then spend an hour failing to type the same thing. Flit closes that gap: you talk through the paragraph and finished, punctuated prose appears in the document, at roughly 4 times typing speed. Why students choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. A student budget and a monthly subscription do not mix. Flit is paid once, needs no account, and keeps working through every term. ### Flit for Multilingual URL: https://flit.fyi/use-cases/multilingual You think in one language and your colleagues read another. Most dictation tools make you declare which language you are in before every sentence, then hand you a translation you did not ask for. Flit works out what you are speaking in the first seconds and only translates when you tell it to. Why multilingual professionals choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. Flit detects the language you speak and can write it in another, all on your Mac, so it works in more than 1,600 languages with wifi off. ### Flit for Hands-free & RSI URL: https://flit.fyi/use-cases/accessibility If typing hurts, every tool that says “just type less” is missing the point. Flit can run without a keypress at all: arm it once and it listens for your voice, starts when you speak, and stops when you pause. The text arrives finished, so there is no cleanup pass to type either. Why people with RSI choose Flit over Wispr Flow: it runs on the Mac, not in the cloud; it is paid once, not monthly. If you depend on dictation, it should not depend on a server or a subscription. Flit works with no internet, and there is no monthly payment whose lapse takes your voice input away. ## Comparisons Every competitor price below was read from that product’s own pricing page on the date shown on the page. Concessions are stated because they are true. ### Flit vs Wispr Flow URL: https://flit.fyi/alternatives/wispr-flow Flit is the local alternative to Wispr Flow: the same instant, formatted dictation on a Mac, without the cloud, the account, or the monthly bill. Flow processes your voice and your screen context on its servers, and Pro is $12 per user a month, with a free tier capped at 2,000 words a week on desktop. Flit does the same job entirely on your machine, adds a codebase index and formatting that cannot drift, and costs $19 once. If you dictate on a Mac and would rather your words stayed there, Flit is the one to switch to. Flow keeps the edge only if you also dictate on Windows, iPhone, or Android. Pick Wispr Flow if you dictate on more than a Mac, need SSO and admin controls, or want a dictionary that follows you between devices. Pick Flit if your writing should not leave your machine, or you would rather pay once. ### Flit vs Superwhisper URL: https://flit.fyi/alternatives/superwhisper Flit is the faster and cheaper choice for dictation: text appears while you are still speaking, where Superwhisper shows it after you stop. Superwhisper is a subscription: $8.49 a month, $84.99 a year, or $249.99 once. Its free tier is the smaller version, without file transcription, cloud models or translation. Flit is $19 once and yours to keep. Both can run on your Mac, so this is about speed, price and focus rather than privacy. Both transcribe recorded files on the Mac. Superwhisper also records meetings, and runs on Windows, iPhone and Android, which Flit does not. Pick Superwhisper if you record meetings, or dictate on Windows or a phone as well as a Mac. Pick Flit if you want dictation that appears as you speak, for one payment. ### Flit vs SpeakMac URL: https://flit.fyi/alternatives/speakmac Flit and SpeakMac agree on the things that matter most: both transcribe on your Mac, both are bought once with 12 months of updates, and neither wants an account. They part on depth. Flit reads your open project so spoken file names come out spelled the way your repo spells them, applies a different style per app, covers more than 1,600 languages rather than 25, and resolves more than 60 spoken commands by fixed rule instead of leaving them to a model. SpeakMac is $29 for a single Mac and reaches Intel and macOS 13, so on an older machine that is the one that fits. Flit is $19 for 2 Macs, and if dictation is what you do all day it is the one built for the all-day case. Pick SpeakMac if you are on an Intel Mac or macOS 13, or you need meeting recordings. Pick Flit if dictation is what you do all day and you want it shaped precisely: more languages, code that spells itself, and a different register per app. ### Flit vs Spokenly URL: https://flit.fyi/alternatives/spokenly If you dictate all day on a Mac, Flit is the better tool: streaming text, per-app tone rules, a local codebase index, and formatting produced by fixed rules rather than a model guess. Spokenly is free and spans four platforms, which makes it a sensible starting point for occasional dictation, and its Local Only Mode is a genuine privacy feature. The trade is depth: covering four platforms means the macOS-specific work does not get done. Pick Spokenly if you want free, or you write on more than a Mac. Pick Flit if you dictate all day on a Mac and want streaming text, per-app tone, and formatting that cannot drift. ### Flit vs VoiceInk URL: https://flit.fyi/alternatives/voiceink Flit and VoiceInk are the two serious local, one-time-purchase dictation apps for Mac, and Flit is the more capable of them: text appears as you speak, it indexes your open project so spoken code names come out right, it rewrites selected text by voice, and it dictates in more than 1,600 languages. VoiceInk is $25 for one Mac, $39 for two and $49 for three, with updates for life, and it is open source. Pick VoiceInk if auditable source matters most, and you do not need streaming or a codebase index. Pick Flit if you want text as you speak, spoken code names spelled right, or dictation in more than one language. ### Flit vs FluidVoice URL: https://flit.fyi/alternatives/fluidvoice Flit and FluidVoice both run open speech models on the Mac, so the model roster is not what separates them. What separates them is everything after the transcription. Flit resolves times, punctuation and more than 60 spoken commands by fixed rule, keeps a dictionary that asks before learning, indexes your open project so file names come out spelled right, and has no cloud path anywhere in the product. FluidVoice is free and open source under GPLv3, runs on Intel, and its optional AI cleanup can route through OpenAI or Groq. If free and open source is the deciding factor, FluidVoice is the real thing. If dictation is your all-day input method, Flit is the one built and maintained as exactly that. Pick FluidVoice if free and open source is the deciding factor, you are on an Intel Mac, or you need multi-hour recordings transcribed. Pick Flit if dictation is your all-day input method and you want deterministic commands, your own vocabulary, a codebase index, and no cloud door anywhere in the product. ### Flit vs Snaply URL: https://flit.fyi/alternatives/snaply Flit and Snaply start from the same principles: both run their models on your Mac, both are Mac only, and neither asks for an account. Snaply is free for individuals and does more jobs, adding meeting notes with summaries and action items, which Flit does not attempt. Flit takes dictation further: more than 1,600 languages against English plus 25 European ones, a local index of your open project so spoken file names come out spelled right, a style rule per app, and more than 60 spoken commands resolved by fixed rule rather than by a model. If you want one free app that covers meetings and writing help, Snaply is a strong answer. If dictation is most of your typing, Flit is the tool built for that. Pick Snaply if you want meeting notes and writing help in one free app, or English plus European languages is all you need. Pick Flit if dictation is most of your typing and you want it precise: more than 1,600 languages, code that spells itself, and commands that cannot drift. ### Flit vs Utter URL: https://flit.fyi/alternatives/utter Flit is the more focused dictation tool: local processing is the whole product rather than the free tier you leave behind when you upgrade, text appears while you speak, and it is $19 once instead of $5 every month. Utter is the broader one, covering iPhone, file imports, and meeting transcripts, and its free tier is genuinely usable. Choose on whether you want dictation done deeply or several jobs done adequately. Pick Utter if you want dictation on your iPhone too, or you also need link and meeting transcription. Pick Flit if you want everything to stay on your Mac as the whole product, not as a free tier, for one payment. ### Flit vs Voibe URL: https://flit.fyi/alternatives/voibe Flit and Voibe are the closest match on features, and Flit wins on the terms that do not change: no account, no cloud path to configure, and $19 once against $149 for Voibe's lifetime tier. Both stream text live and both understand developer context. Voibe keeps the edge if you also write on Windows, or if you are on an Intel Mac, which Flit does not support. Pick Voibe if you also write on Windows, or you have an Intel Mac. Pick Flit if you want no account, no cloud path at all, and a lower one-time price. ### Flit vs Yaps URL: https://flit.fyi/alternatives/yaps Flit and Yaps both keep your voice on your own machine, so the choice is about focus and how you pay. Yaps is a toolkit: dictation, file transcription, captions, voice notes with a markdown vault, read-aloud and translation, across Mac, Windows, Linux, Android, and iOS, for $12 a month billed yearly with an account. Flit puts its effort into dictation, and does more of it: text appears while you speak, spoken code names are repaired against your open project, more than 60 commands resolve by fixed rule, and it dictates in more than 1,600 languages against around 25. It costs $19 once, needs no account, and the local claim is verifiable in a minute with wifi off. If you want the notes, the files and the other platforms, Yaps is a fair subscription. If you want the best local dictation on a Mac and to stop paying for it, Flit. Pick Yaps if you want voice notes, read-aloud and dictation in one subscription, or you dictate on Windows, Linux or a phone. Pick Flit if dictation on a Mac is the job, you want text as you speak, and you would rather pay once than every month. ### Flit vs Typeless URL: https://flit.fyi/alternatives/typeless Flit is the local, buy-once answer to Typeless. Typeless sends your speech to its servers and meters it by the word: 8,000 words a week free, unlimited for $12 a month billed yearly. It covers macOS, Windows, iOS and Android, which Flit does not. What Flit removes is the whole apparatus: no account, no meter, no queue, and no retention policy to trust, because no request is made at all. If you write across 4 platforms, Typeless is built for that. If your writing happens on a Mac, Flit is the one where nobody is counting. Pick Typeless if you write on more than a Mac, or you want team controls like SSO and audit logs. Pick Flit if your writing happens on a Mac and you want it to stay there, with no account and one payment. ### Flit vs Aqua Voice URL: https://flit.fyi/alternatives/aqua-voice Both apps read your screen to write better text; Flit is the one that does it without uploading anything. Aqua sends that context to its servers and charges $8 to $24 a month with an account. Flit reads the focused field on your Mac, holds it in memory for exactly one dictation, writes it nowhere, and costs $19 once. If the thing on your screen is the reason you want dictation to be smart, it is also the reason you should not send it anywhere. Pick Aqua Voice if you want it on iPhone too, or you want to try it free before deciding. Pick Flit if the context it reads is the exact thing you do not want uploaded. ### Flit vs Willow Voice URL: https://flit.fyi/alternatives/willow-voice Flit is the local alternative to Willow Voice: the same hold-a-key, finished-text-at-your-cursor dictation on a Mac, without the cloud, the account, or the monthly bill. Willow runs on cloud infrastructure in the United States and is $15 per user a month, or $12 billed annually, with a free tier limited to a weaker model and Business at $35 for the SOC 2 and HIPAA controls. Flit does the same job entirely on your machine for $19 once, adds a local codebase index and formatting that cannot drift, and has no account to attach anything to. Willow keeps the edge if you dictate on Windows or an iPhone, or if your procurement team needs certificates rather than architecture. Pick Willow Voice if you need Windows or an iPhone, or your organization requires SOC 2, HIPAA, SSO and admin controls. Pick Flit if you dictate on a Mac and would rather the words never left it than be certified in transit. ### Flit vs MacWhisper URL: https://flit.fyi/alternatives/macwhisper These overlap only in part, and Flit is the one you want for writing. MacWhisper turns recorded files into transcripts, with speaker labels, subtitles and batch processing. Flit turns your speech into finished text in the app you are already using, right now, with formatting, per-app tone, and a codebase index, and since 1.5 it transcribes recordings too, with subtitle export but without speaker labels. Both keep your audio on your Mac. Pick MacWhisper if your recordings need speaker labels, watch folders or transcript editing. Pick Flit if you want to stop typing and start talking in the apps you already use, and turn a recording into text on the same Mac. ### Flit vs Apple Dictation URL: https://flit.fyi/alternatives/apple-dictation Flit is what you move to when Apple Dictation stops being enough. The free one gives you roughly what you said; Flit gives you what you meant to write, with fillers removed, times and emails formatted by fixed rules, tone matched to the app you are in, snippets, a learning dictionary, and a local codebase index. For a text message now and then, the built-in one is fine and you should use it. For writing all day, the cleanup it leaves you is the cost. Pick Apple Dictation if you dictate occasionally and do not mind tidying the result. Pick Flit if dictation is becoming how you write, and the tidying is the part you want gone. ### Flit vs Otter.ai URL: https://flit.fyi/alternatives/otter-ai These tools answer different questions, and Flit is the one for writing. Otter records meetings and produces shared transcripts in its cloud. Flit types what you say into whatever app you are in, on your Mac, and stores nothing. If your problem is remembering what was said in a call, Otter. If your problem is that typing is slower than thinking, Flit, and most people with both problems should own both. Pick Otter.ai if you need a record of conversations with other people. Pick Flit if you need your own words to become written text, faster. ### Flit vs AudioPen URL: https://flit.fyi/alternatives/audiopen These are not really competitors, and pretending otherwise would waste your time. AudioPen is a place you go to speak: you record up to 15 minutes, and it rewrites the ramble into structured prose in a style you choose, on the web, iPhone, iPad, Watch, Android, a Mac app or a Chrome extension. Flit never has a window of its own. You put your cursor wherever you were already typing, hold a key, and your own words arrive there formatted but not rewritten. If you want messy thinking turned into a clean note, AudioPen does that well. If you want to stop typing, that is Flit. Pick AudioPen if you want to capture spoken thoughts on any device and have them rewritten into polished prose. Pick Flit if you want to stop typing on your Mac, with your own words landing in whatever app you are already using. ### Flit vs Talon Voice URL: https://flit.fyi/alternatives/talon For turning speech into finished writing, Flit is the simpler and better tool: it works in minutes with nothing to script, and it produces formatted prose rather than commands. Talon is a different class of software, replacing the keyboard and mouse entirely, and it is the right answer if using them is not an option for you. Many people with a serious RSI want both: Talon to drive the machine, Flit to write the words. Pick Talon Voice if you need to control your computer without a keyboard, or you want to code by voice properly. Pick Flit if you want finished writing from speech, with no setup. ### Flit vs Whisper (do it yourself) URL: https://flit.fyi/alternatives/whisper-diy The models are free; the product is not the model. Flit is the several hundred decisions between a transcript and usable dictation: streaming as you speak, a hotkey that works everywhere, insertion that survives Electron and Chrome apps, formatting that turns speech into writing, and someone else maintaining all of it when macOS changes. Building it yourself makes sense if the building is the point. If the outcome is the point, $19 is cheaper than a weekend. Pick Whisper (do it yourself) if you want total control, or the building is the point. Pick Flit if you want dictation that works today and keeps working. ### Flit vs Voiskey URL: https://flit.fyi/alternatives/voiskey Flit is the local alternative to Voiskey. Both turn rough speech into clean text in the app you are using. Voiskey sends your audio to its servers, works with Google and other partners for the AI, and asks for a Google or Apple ID before you start. Premium is $29.99 a month or $129.99 a year. Flit does the same job entirely on your Mac, with no account and text that appears while you speak, for $19 once. Voiskey keeps the edge if you also dictate on Windows or a phone. If you write on a Mac, Flit keeps your words on it and you pay once. Pick Voiskey if you dictate on Windows, iPhone or Android as well as a Mac, and a cloud service is acceptable. Pick Flit if you dictate on a Mac and want no account, no upload, and one payment. ### Flit vs EmberType URL: https://flit.fyi/alternatives/embertype Flit and EmberType are close cousins: both transcribe on your Mac, both run on Apple silicon with macOS 14 or later, both are bought once, and both clean up fillers and punctuation on the device. The difference is what each does beyond that. EmberType lets you plug in cloud engines like Groq, Deepgram or Gemini with your own API key. Flit has no cloud path at all, shows text while you are still speaking, repairs spoken code names against your open project, and resolves more than 60 commands by fixed rule. If you want the choice of a cloud provider, EmberType offers it. If you want dictation that cannot leave your Mac and does more of the job, Flit. Pick EmberType if you want the option to use a cloud engine with your own API key. Pick Flit if you want live text, a codebase index and exact commands, with no cloud path at all. ### Flit vs DictaFlow URL: https://flit.fyi/alternatives/dictaflow Flit is the fully local, buy-once alternative to DictaFlow. On Pro, DictaFlow transcribes on your device, but it sends your text to the cloud for AI formatting, counts those words against an allowance, and needs an account. Pro is $7 a month, or $69 a year. It covers Windows, Mac, iPhone and Android, and works inside Citrix and remote desktops, which Flit does not. Flit transcribes and formats on your Mac, so nothing is sent and nothing is counted, and it is one payment you keep. If you dictate on Windows or through a virtual desktop, DictaFlow is built for that. If you write on a Mac, Flit keeps every step on it. Pick DictaFlow if you dictate on Windows, a phone, or inside Citrix and remote desktops. Pick Flit if you write on a Mac and want transcription and formatting both kept on it, with one payment. ### Flit vs Aside URL: https://flit.fyi/alternatives/aside Flit and Aside both keep your voice on your Mac, so this is a question of scope rather than trust. Aside is a meeting recorder that also dictates: no bot joins the call, transcription runs on your machine, speakers are separated afterwards, and every note lands as Markdown in a folder you choose, for $8 a month or $79 once. Flit dictates and transcribes recordings you already have, and spends most of its effort on the dictation: formatting fixed by rules rather than decided by a model, tone that follows the app you are in, and a local index of your open project so spoken file names come out spelled right. If your problem is that meetings disappear, buy Aside. If your problem is that typing is slower than thinking, Flit is $19 once and never asks you to pick an AI provider. Pick Aside if your meetings need to become notes you own, or you want one app covering the call and the cursor. Pick Flit if dictation is the job, and you would rather not choose an AI provider to finish a sentence. ### Flit vs Raycast URL: https://flit.fyi/alternatives/raycast Raycast added dictation to a launcher that thousands of people already keep open, and if you are on Pro it costs nothing extra. Flit is the choice when the dictation itself is the reason you are looking: Raycast sends your audio to its speech-to-text partner to be transcribed, requires a Pro plan at $10 a month or $8 billed annually, and caps a session at 20 minutes. Flit runs on your Mac, has no account, no session limit and no subscription, and is $19 once. If you already pay for Raycast and dictate occasionally, its built-in dictation is a sensible free upgrade. If you dictate all day, or your words should not leave the machine, that is the switch. Pick Raycast if you already pay for Raycast Pro, dictate in bursts, and want one less app on the machine. Pick Flit if dictation is how you write, or your audio should not be sent anywhere at all. ### Flit vs Talkify URL: https://flit.fyi/alternatives/talkify Talkify is free, MIT licensed, and clear about what it is: the quickest way to get raw words into a text field on a Mac running macOS 26. It carries no speech model of its own, borrowing Apple’s system engine instead, which is why the download is under 10 MB and why it covers thirty locales across ten languages. Its own documentation says version one inserts raw text with no cleanup pass, and that its advantage is speed rather than accuracy. Flit is $19 once and spends its time on exactly the part Talkify skips: formatting fixed by rules, per-app tone, a local project index, and more than 1,600 languages. If you want free and fast on a new Mac, take Talkify. If you want the text to arrive finished, Flit. Pick Talkify if you want free, open source and fastest on a macOS 26 Mac, and you speak one of its ten languages. Pick Flit if you want finished text rather than a transcript, more than 1,600 languages, or a Mac that is not on macOS 26. ### Flit vs Handy URL: https://flit.fyi/alternatives/handy Flit and Handy both run entirely on your machine, so neither is asking you to trust a server. Handy is free and MIT licensed, runs on macOS, Windows and Linux, and does one thing: hold a key, speak, and the transcript is pasted where you were typing. Its author says plainly that it is not trying to be the best speech-to-text app but the most forkable one, and it was built as an accessibility tool after he broke a finger. Flit is $19 once and spends its effort on what happens between the transcript and your cursor: streaming text, formatting fixed by rules, per-app tone, and a local index of your project. Take Handy if you want a small tool you can read and change. Take Flit if you want the words to arrive finished. Pick Handy if you want a free tool you can read, fork and keep, or you dictate on Windows or Linux as well. Pick Flit if you want finished text rather than a transcript, and you would rather not maintain it yourself. ### Flit vs Muesli URL: https://flit.fyi/alternatives/muesli Muesli is free, MIT licensed, and covers more ground than Flit: local dictation, meeting transcription with no bot on the call, an iPhone app, Apple Shortcuts and Siri actions, and a spoken-instruction rewrite it calls Quill. It runs on Apple silicon and lets you pick your model, with an optional hosted route through OpenAI or OpenRouter if you want one. Flit is $19 once, covers less, and spends most of it on the dictation: formatting fixed by rules rather than decided by a model, a local project index that spells your file names right, and no hosted path to configure or double-check. If you want one free app for calls and writing, take Muesli. If dictation is the job, Flit. Pick Muesli if you want one free, open-source app for meetings and dictation, an iPhone app, or your pick of model. Pick Flit if dictation is the job, and you want formatting that cannot drift and no hosted path to check. ### Flit vs TypeWhisper URL: https://flit.fyi/alternatives/typewhisper TypeWhisper is the unusual one: a local-first dictation app that is free under GPLv3, ships on macOS, Windows and iOS, needs no account, and sells a commercial license on top for teams that need non-GPL terms, procurement paperwork and support. That license starts at €5 a month for three devices or €99 once, and it is pay-what-you-want above the listed price. Flit is $19 once, Mac only, and closed: what it buys you is the layer above the transcript, with formatting fixed by rules, per-app tone, a local project index, and no provider to configure. If you want free and cross-platform with a commercial option behind it, TypeWhisper is a strong answer. If you want the Mac version of dictation done as deeply as possible, Flit. Pick TypeWhisper if you want a free GPLv3 app, you dictate on Windows or iOS too, or you need commercial terms and an invoice. Pick Flit if you are on a Mac and want formatting that cannot drift, project awareness, and one price. ### Flit vs Verba URL: https://flit.fyi/alternatives/verba Verba and Flit start from the same principle and end in different places. Both run on Apple silicon and keep your audio on the machine. Verba goes further than dictation: it acts, sending the email or creating the event across a claimed 1,000-plus apps, showing you the plan and waiting for your confirmation first. Raw dictation is free forever with no card, and everything else is $9.99 a month, $84 a year, or $149 once, with your own AI account behind it. Flit does not act, and that is the whole design: it types finished text at your cursor for $19 once, with no model to bring, no key to buy, and formatting fixed by rules. If you want a voice agent, take Verba. If you want dictation that cannot surprise you, Flit. Pick Verba if you want your voice to act across your apps, or you want free unlimited raw dictation. Pick Flit if you want finished text and nothing else, with no AI account and no permissions to grant. ### Flit vs stilltype URL: https://flit.fyi/alternatives/stilltype stilltype is not something you can buy yet. It is a private beta limited to fifty hand-picked invites, built in the open by one person, and it describes the same architecture Flit does: on-device dictation on Apple silicon, no account ever, pay once rather than subscribe. It does not publish a price, a language count, or a release date, and it pointedly refuses to publish an accuracy number on the grounds that nobody measured your voice. We agree with that instinct. Until it ships, the comparison is short: if you want this kind of app today, with a published price, more than 1,600 languages and formatting that arrives finished, that is Flit at $19 once. Put your name down for stilltype too. Pick stilltype if you want to follow a build in public and do not need dictation working this week. Pick Flit if you want the same architecture from something you can buy today, with a price on the page. ### Flit vs Monologue URL: https://flit.fyi/alternatives/monologue Monologue is the closest competitor Flit has on output and the furthest away on architecture. It formats from context, gets file names and technical terms right, covers Mac, iPhone and iPad, and adds bot-free meeting notes that sync to a web app, for $15 a month or $144 a year. The catch is in its own documentation: remote processing is the recommended default, and even in Local mode the smart cleanup and context-aware formatting still call Monologue’s online service, because “local does not mean every step is offline”. Flit does the whole job on your Mac for $19 once, with no account and nothing uploaded. If you want the Apple-wide reach and the meeting notes, Monologue is a good product. If the point is that your words stay put, that is the difference. Pick Monologue if you dictate on an iPhone or iPad as well, or you want meeting notes that sync and share. Pick Flit if you want the formatting as well as the transcription to happen on your machine, and to pay once. ### Flit vs Alter URL: https://flit.fyi/alternatives/alter Alter is a native macOS AI assistant that happens to take dictation, not a dictation app that grew an assistant. It reads your screen, controls Apple apps, records meetings with speaker identification, connects to a claimed 2,000 services, and runs on Intel Macs and macOS 13, which Flit cannot. The app is free if you bring your own API keys or run models locally through Ollama or LM Studio, and by its own feature table the free tier’s dictation is the slower Whisper path: the fast local models, Whisper Pro and Parakeet V3, are the Local+ add-on at $3.50 a month or $29 a year. Flit is $19 once, dictates and transcribes files, and needs no provider, no key and no account. Pick Alter if you want one assistant running your Mac, you are on an Intel Mac or macOS 13, or you already run local models. Pick Flit if you want dictation that is finished, deterministic and ready the moment you install it. ### Flit vs Speechify URL: https://flit.fyi/alternatives/speechify These two run the same pipeline in opposite directions, which is why they turn up in the same search. Speechify reads text to you: 1,000-plus voices, 60-plus languages, up to 5x speed, on every platform and in the browser, for $29 a month or $139 a year, with a free tier limited to ten robotic voices at 1.5x. Flit listens to you and writes, on a Mac, for $19 once. Speechify Premium does now bundle Voice Typing, so the overlap is not zero, but reading is the product and dictation is a line item. If you want documents read aloud, buy Speechify and do not let a dictation comparison talk you out of it. If you want to stop typing, that is Flit. Pick Speechify if you want text read aloud, in any voice, on any device, or reading is the part that is slow for you. Pick Flit if you want to stop typing on a Mac, with nothing uploaded and one payment. ### Flit vs Tyll URL: https://flit.fyi/alternatives/tyll Tyll is the closest thing to Flit’s architecture that anyone has put a price on: entirely on your Mac, no account, no subscription, offline after activation, Apple silicon, one payment. It goes wider than Flit by bundling a meeting assistant with speaker detection into the same app, and it leans on the AI agent you already pay for rather than selling you a second one. Two things separate them today. Tyll is not on sale yet: its download page says the download is not enabled and offers a launch notification instead. And it recognises 25 European languages where Flit covers more than 1,600. If you want dictation and meeting notes in one local, buy-once app and you can wait, put your email in. If you want to dictate this afternoon, that is Flit at $19. Pick Tyll if you want dictation and meeting notes in one local, buy-once app, and you can wait for it to ship. Pick Flit if you want more than 1,600 languages and dictation you can buy and use this afternoon. ## The ranked answer For "what is the best dictation app for Mac", the ranked list with a best-for label per tool, honest cons, and verified prices lives at https://flit.fyi/best-dictation-app-for-mac. The one-sentence version: Flit for live dictation that never leaves the Mac; Wispr Flow if you accept cloud and want cross-device; Superwhisper or MacWhisper for recorded files; Apple Dictation if free and occasional is enough. ## What is new, and how to use it One page per big release: what it gives and the steps to use it. ### Flit 1.7: now in German, French, Japanese and 7 more languages https://flit.fyi/changelog/1.7 (published 2026-10-18) In short: Flit 1.7 is a free update that puts the whole app in 11 languages: English plus German, French, Spanish, Italian, Swedish, Dutch, Brazilian Portuguese, Japanese, Chinese and Korean. Flit follows the language of your Mac, from setup to Settings. Two things that worked only in English now work in all ten: change your mind mid-sentence and Flit keeps what you meant, and count items out loud to get a numbered list. Transcribe with Flit is now on in Finder, and double-tap locks hands-free on every Mac. Use Flit in your language, and how to change it: Menus, Settings, setup, the recording window and every message now come in German, French, Spanish, Italian, Swedish, Dutch, Brazilian Portuguese, Japanese, Chinese and Korean, besides English. There is nothing to set: Flit follows the language of your Mac, so on a Mac set to German, Flit is in German. To use another language for Flit alone, set it in System Settings: 1. Open System Settings and choose General, then Language & Region. 2. Under Applications, click +, choose Flit, then the language. 3. Quit Flit from its menu and open it again. The language of the app does not change what you can dictate. Flit hears and formats the language you speak, whatever language its buttons are in. Self-corrections and numbered lists in ten languages: Two things Flit did only in English now work in all ten languages. Correct yourself and Flit keeps what you meant. Count items out loud and they become a numbered list: - German: "Lass uns am Montag telefonieren. Sorry, ich meinte Dienstag." becomes "Lass uns am Dienstag telefonieren." - French: "Liste de courses. Premièrement des pommes, deuxièmement du lait, troisièmement des œufs." becomes a numbered list of three. - Japanese: "月曜日に電話しましょう。すみません、火曜日でした。" becomes "火曜日に電話しましょう。" - Spanish, in an AI chat: "¿Puedes hacer tres cosas? Primero, haz el título más grande. Segundo, añade un botón para saltar. Y tercero, quita la barra de progreso." becomes the question and three numbered steps. A correction works on days, months, times and numbers, the things people most often say twice. A numbered list needs the Polished level in Settings → Formatting, the one Flit starts with. Setup examples in the language you speak: The last step of setup has four examples to read out loud: talk freely, fix a slip, make a list and ask an AI. They are now written in the language you choose for dictation, so a German speaker reads German and sees German come back. For a language outside the ten, the examples stay in English. Transcribe a file from Finder, now on by default: Since 1.5 you could transcribe a recording from Finder, but macOS kept the menu item switched off until you found it in System Settings. It is on now. Right-click an audio or video file and choose Transcribe with Flit, at the bottom of the menu or under Services. If the item is not there right after the update, log out and back in once, and macOS picks it up. Fixes: - Double-tap now locks hands-free on every Mac. On a Mac whose microphone is slow to start, the second tap used to stop the recording instead. - Flit no longer deletes real words it took for fillers, such as German "um" in "um halb acht", Portuguese "um" in "um carro" and Spanish "este". - Apple's speech engine uses your Mac's variant of a language: Brazilian Portuguese on a Mac set to Brazil, British English on a Mac set to the United Kingdom. Questions: - Q: Is Flit 1.7 a free update? A: Yes. Every license includes twelve months of updates, and 1.7 is one of them. The trial has every language too. - Q: How do I get it? A: Right-click the Flit icon in the menu bar and choose Check for Updates…. Your settings, dictionary and history carry over, and a window after the restart shows what is new, in your language. - Q: How do I change the language of Flit? A: Flit follows the language of your Mac. To give Flit a language of its own, open System Settings → General → Language & Region, and add Flit under Applications. - Q: Does the app language change what I can dictate? A: No. You can dictate in any language your voice model covers, more than 1,600 with Omnilingual, whatever language the app is in. - Q: My language is not one of the ten. What do I see? A: Flit in English. Dictation, formatting and transcription work as before in every language your voice model covers. - Q: Is anything uploaded? A: No. Like every Flit release, 1.7 runs on your Mac, and Flit dictates with Wi-Fi off. ### Flit 1.6: four new voice models, and more than 1,600 languages https://flit.fyi/changelog/1.6 (published 2026-10-16) In short: Flit 1.6 adds four voice models, and every one runs on your Mac. Parakeet Ultra makes fewer mistakes than Parakeet V3 in English and 24 other European languages, at the same speed, and is now the model Flit recommends for them. Voxtral Realtime writes while you speak in 13 languages, Hindi, Arabic, Japanese and Korean among them. Omnilingual ASR covers more than 1,600 languages, so Tamil, Bengali, Urdu and Swahili have a model too. Parakeet Unified is a new English model. The model you use now stays as it is, and 1.6 is a free update for every license. Which model to use: You only need one voice model. Choose it by the language you dictate in: - English or another European language: Parakeet Ultra 0.6B. Your words appear while you speak, and it is the fastest model you can download. For German, French, Spanish, Italian or Dutch on a Mac with 16 GB of memory or more, Cohere Transcribe is more accurate, and its text appears when you stop speaking. - Hindi, Arabic, Japanese, Korean or Chinese, with words while you speak: Voxtral Realtime 4B. - A language none of these cover, such as Tamil, Bengali, Urdu, Marathi or Swahili: Meta Omnilingual ASR 3B. - English only, and you want to compare: Parakeet Unified 0.6B. Flit suggests one when you set it up, from the languages you say you speak, and Settings → Models shows the same suggestion at the top. Download a model and switch to it: The new models are a download away, and the one you use now keeps working while they download. 1. Right-click the Flit icon in the menu bar and choose Settings…, then Models in the sidebar. 2. Find the model in the Library and click Download. 3. When it says Downloaded, click the row. It is marked Active. 4. Dictate as usual. The first dictation after a switch takes a few seconds longer while the model loads. The bin on a row deletes a model you no longer use and gives the space back. Parakeet Ultra: fewer mistakes, the same speed: Parakeet Ultra is Parakeet V3 trained further by moondream. It covers the same 25 languages and runs at the same speed on the Neural Engine, with fewer mistakes: across 24 languages of the FLEURS test set, its average word error rate is 11.7%, against 14.8% for V3. In our own test, a Danish sentence read aloud came back nearly word for word, where V3 heard "torsdag eftermiddag" as "torsdag efter med dag". Your words still appear while you speak, and the rest lands about a tenth of a second after you let go of the key. If you use Parakeet V3 today, nothing changes until you switch. V3 stays in the list. A 595 MB download, about 100 MB more than V3. Voxtral Realtime: live words in 13 languages: Voxtral Realtime, from Mistral, writes while you speak, punctuation included, in English, French, Spanish, German, Russian, Chinese, Japanese, Italian, Portuguese, Dutch, Arabic, Hindi and Korean. It works out which of them you are speaking, so there is no language to set. Words appear a second or two behind your voice, and the rest lands within half a second of letting go of the key. In our tests, Hindi and Arabic sentences came back nearly word for word. It is the largest of the four: a 2.8 GB download that holds a few gigabytes of memory while it runs, so it suits a Mac with 16 GB or more. Omnilingual: more than 1,600 languages: Meta's Omnilingual ASR hears more than 1,600 languages, among them Tamil, Telugu, Bengali, Marathi, Urdu, Punjabi, Swahili, Yoruba and Amharic. If no other model covers a language you speak, Flit recommends it when you set up. It writes plain words, with no punctuation or capitals. Flit's formatting adds them at Polished, the level in Settings → Formatting. At Verbatim you get the words exactly as the model wrote them. Accuracy depends on the language. In our tests it was close to word for word in Hindi, made a few slips in Arabic, and more in Tamil and Bengali. Where another model covers your language, that model is the better choice. A 1.8 GB download. Text appears when you stop speaking. Parakeet Unified: a new English model: Parakeet Unified, from NVIDIA, is trained for live words and whole recordings at once. It is English only, shows your words while you speak, and is a 570 MB download. If you only dictate in English, try it beside Parakeet Ultra and keep the one you like. Use them for recordings too: Every new model works in the Transcribe tab. Choose it under Model below the card, and Flit uses it for files while dictation keeps its own model. That is how a recording in Tamil, Hindi or Korean becomes a transcript on your Mac. Questions: - Q: Is Flit 1.6 a free update? A: Yes. Every license includes twelve months of updates, and 1.6 is one of them. The trial has every model too. - Q: How do I get it? A: Right-click the Flit icon in the menu bar and choose Check for Updates…. Your model, settings, dictionary and history carry over, and a window after the restart shows what is new. - Q: Does my voice model change? A: No. Flit keeps the model you use until you choose another. The new ones are a download away in Settings → Models. - Q: Which model is best for my language? A: Open Settings → Models and add your languages under Languages you speak. The card at the top names the model Flit recommends for them, and why. - Q: Do I need all of them? A: No. One voice model is enough. Each is a download of 570 MB to 2.8 GB, and you can delete the ones you do not use. - Q: Is anything uploaded? A: No. Every model runs on your Mac, and Flit dictates with Wi-Fi off. The only network request is the download you start yourself. - Q: Why does Omnilingual write without punctuation? A: The model writes the words only. Flit's formatting adds punctuation and capitals at the Polished level. - Q: Does Voxtral work on a Mac with 8 GB of memory? A: It is built for 16 GB or more. Beside the formatting model it leaves little memory for your other apps, so on 8 GB, Parakeet Ultra or Qwen3-ASR 0.6B is the better choice. ### Flit 1.5: transcribe a recording on your Mac https://flit.fyi/changelog/1.5 (published 2026-10-12) In short: Flit 1.5 adds a Transcribe tab that turns an audio or video file into a transcript on your Mac, with nothing uploaded. Drop a recording on the tab or on the Flit icon in the menu bar, or right-click it in Finder. The text appears as it goes, and a two-hour recording takes a few minutes. When it is done you can read what the voice model heard beside the formatted text, copy either, and export Markdown, SRT or VTT. It is a free update for every license, and it is on in the trial. Transcribe your first recording: The Transcribe tab is in Settings, under Activity. It takes audio and video that macOS can play: MP3, M4A, WAV, AIFF, FLAC, MP4 and MOV among them. A video is read for its sound, and nothing is done with the picture. 1. Right-click the Flit icon in the menu bar and choose Settings…, then Transcribe in the sidebar. 2. Drop a recording on the tab, or click Choose File and pick one. 3. The file is shown with its length. Click Transcribe. 4. The text appears as each piece finishes. When it is done, click Copy. The first transcript uses the voice model you dictate with, so there is nothing more to download. Four more ways to hand Flit a file: You do not have to open Settings first. Each of these opens the tab for you, so you can watch the text arrive. - Drop it on the Flit icon in the menu bar. The job starts at once. - Right-click the file in Finder and choose Services → Transcribe with Flit. The job starts at once. - Open it with Flit: right-click the file, choose Open With, then Flit. The job starts at once. - Right-click the Flit icon and choose Transcribe File…. A file picker comes up, and the tab opens with the file ready, so you can change the model or the formatting before you click Transcribe. Select several recordings and they run one after another. While it runs: The card shows how far into the recording Flit is and about how long is left, with the transcript so far under it. Stop ends the job. A file you add while one is running waits in line and starts by itself, and the card says which is next. You can close Settings and keep working. The Flit icon in the menu bar wears a ring that fills as the job goes, and its menu shows the file name and the percentage. If Flit is in the background when the transcript is ready, a notification says so. Click it to open the transcript. The first time a transcript finishes in the background, macOS asks whether Flit may send notifications. Read it, copy it, export it: A finished transcript shows its length, word count, voice model and formatting level on top, and the text in paragraphs. A new paragraph starts where the speaker paused for two seconds or more. - Copy puts the text on the clipboard. - Export saves it as Text, Markdown, SRT subtitles or VTT subtitles. The subtitle files carry timestamps, one line per sentence. - Show in Finder opens the folder with the text file Flit saved for you. - Transcribe Another clears the card for the next recording. The transcript stays in the list below. Raw or formatted: see what was heard: Under the title is a switch with two sides. The left side is the formatting level the transcript was made at, Clean or Polished. The right side is Raw: what the voice model heard, with every hesitation and false start left in. Use it to check a sentence that reads oddly, or when you want the words exactly as spoken. Copy takes whichever side is showing. The saved text file and the exports stay the formatted transcript. The switch is hidden when the two would read the same, as they do for a transcript made at Verbatim. Transcripts made before this update have no raw text to show. Choose the model, the formatting and the folder: Three options sit under the card, and Flit remembers them. They are separate from your dictation settings, so you can dictate with a fast model and transcribe recordings with an accurate one. - Model is the voice model for files. Only models you have already downloaded are listed. To add one, open Models in the sidebar. - Formatting is how far Flit tidies the text. Verbatim keeps every word. Clean takes out the hesitations and writes times and numbers properly, with no wait. Polished runs the formatting model over every paragraph so the recording reads as prose. - Save to is the folder every finished transcript is written to as a text file. It starts as Documents → Flit Transcripts. Click Choose Folder to change it. On an Apple silicon Mac with Cohere Transcribe, a two-hour recording took about three and a half minutes at Clean in our tests, and about fifteen at Polished. Find a transcript later: Every transcript you make is kept under Transcripts at the bottom of the tab, newest first. Click one to open it again, with the same Copy, Export and raw switch. Once you have more than five, a search field appears above the list. It looks in the names and in the text. The bin on a row deletes that transcript from Flit. The text file in your folder is yours and is left where it is. The flit command, for Terminal: The same transcriber is a command, for a folder of recordings or a script. Install it once and flit works in every Terminal window. 1. Open Settings → Transcribe and find Command under Terminal. 2. Click Install. On a Mac with Homebrew that is all. On a Mac without it, macOS asks for your password once. 3. Open a new Terminal window and type the line shown under the button. flit transcribe meeting.m4a flit transcribe meeting.m4a --formatting clean --out ~/Transcripts --srt - The transcript is printed, and the last line gives the speed and the word count. - --formatting takes verbatim, clean or polished. - --out names a folder to write the transcript to as a text file. - --srt and --vtt write subtitles to that folder as well, and --md writes Markdown in place of the text file. - --model and --language choose the voice model and the language of the recording. - flit help lists all of it, and flit version says which Flit the command belongs to. If another tool on your Mac already answers to flit, as a Python packaging tool of that name does, nothing of yours is replaced: the command is installed as flit-app, and the row says so. Remove takes it away again. The command follows the app, so moving or updating Flit does not break it. What it does not do yet: A transcript is one flow of text. Flit does not say who said what, and it does not record a meeting as it happens: you hand it a recording you already have. Both speaker labels and meeting recording are planned, and both will run on your Mac like everything else. Questions: - Q: Is Flit 1.5 a free update? A: Yes. Every license includes twelve months of updates, and 1.5 is one of them. The trial has the Transcribe tab too. - Q: How do I get it? A: Right-click the Flit icon in the menu bar and choose Check for Updates…. The Transcribe tab is in Settings after the restart. Your settings, dictionary, snippets and history carry over. - Q: Is the recording uploaded anywhere? A: No. The file is read on your Mac by the same voice models that dictate, and nothing leaves the Mac. Flit transcribes with Wi-Fi off. - Q: How long can a recording be? A: There is no limit. The file is read five minutes at a time, so a two-hour recording uses about as much memory as a short one. - Q: Which languages does it transcribe? A: The ones your voice model covers: Parakeet covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99, and since 1.6, Omnilingual more than 1,600. - Q: Why is there no Raw switch on my transcript? A: Either it was made at Verbatim, where raw and formatted are the same text, or it was made before the switch arrived. Transcribe the file again and it is there. - Q: Where are my transcripts kept? A: Two places. A text file in the folder you chose, which starts as Documents → Flit Transcripts, and the list in the tab, which also keeps the timestamps for subtitles. - Q: The flit command is not found. What now? A: Open a new Terminal window, since a window that was already open does not see a new command. If the row in Settings says flit-app, type that instead: another tool already had the name flit. - Q: Does it label speakers? A: Not yet. If speaker labels are the job today, MacWhisper does them on the Mac. ## Guides Quick answers, one problem each. The fix that needs nothing installed comes first. ### How to dictate on a Mac: turn it on, start it, and say the punctuation https://flit.fyi/how-to-dictate-on-mac (Mac dictation, published 2026-10-06) Short answer: To dictate on a Mac, open System Settings → Keyboard, turn on Dictation, click where you want the text, and press the Microphone key or the Dictation shortcut, or choose Edit → Start Dictation. Speak, say marks such as "comma" and "new paragraph" by name, and press Escape to stop. Dictation also stops by itself after 30 seconds without speech. Flit, a local dictation app for Mac, does the same job with one gesture: hold a key, speak, let go, and finished text lands in any app. How to dictate on a Mac: 1. Choose Apple menu → System Settings and click Keyboard in the sidebar. 2. Go to Dictation, turn it on, and click Enable. If you are asked about improving Siri and Dictation, click Not Now unless you want Apple to store audio recordings. 3. Open any app and click where you want the text to go. 4. Press the Microphone key in the function row if your keyboard has one, press the Dictation shortcut shown in Keyboard settings, or choose Edit → Start Dictation. 5. When the cursor is highlighted and pulsing, or you hear the tone, speak. 6. Say punctuation by name, such as "comma" or "question mark", and say "new line" or "new paragraph" for a break. 7. Press Escape, the Microphone key or the Dictation shortcut to stop. The things you will want to do: - Start dictating | Microphone key, the Dictation shortcut, or Edit → Start Dictation | The dictation shortcut - Stop dictating | Escape, the Microphone key, or the shortcut again | It also stops after 30 seconds without speech - Add a comma or a period | Say "comma" or "period", or leave it to Auto-punctuation | Dictation commands - Start a new paragraph | Say "new paragraph" | The break appears when you finish dictating - Fix a word | Type over it, or click a word underlined in blue | On Apple silicon you can type without stopping - Dictate in another language | Keyboard → Dictation → Languages → Edit | Switch by clicking the language next to the cursor. More on languages - Use a different microphone | Keyboard → Dictation → Microphone source | Dictating with AirPods - Control the Mac by voice | That is Voice Control, a different feature | Voice Control vs Dictation Dictation with one gesture: Flit is a separate dictation app. You hold a key, speak, and let go, and text that is already punctuated and formatted lands where your cursor is, in any app. - Hold to talk, release to finish. A pause while the key is held does not end anything. - Finished text: punctuation, paragraphs, lists and numbers, in the tone of the app you are in. - Everything runs on the Mac, in every language it supports, with no connection. - A personal dictionary for your names and terms, and snippets that expand a phrase into a paragraph. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: How do I turn on dictation on a Mac? A: Choose Apple menu → System Settings, click Keyboard, go to Dictation, turn it on and click Enable. - Q: How do I start dictation on a Mac? A: Click into a text field, then press the Microphone key in the function row, press the Dictation shortcut shown in Keyboard settings, or choose Edit → Start Dictation. - Q: How do I stop dictation on a Mac? A: Press Escape, the Microphone key or the Dictation shortcut. Dictation also stops automatically when no speech is detected for 30 seconds. - Q: Is dictation on a Mac free? A: Yes. Dictation is part of macOS and there is nothing to buy or install. - Q: Can I dictate in Word, Google Docs or Mail on a Mac? A: Yes. Apple says you can speak to enter text anywhere you can type it. Click into the text and start Dictation. - Q: Is there an app that makes dictating on a Mac easier? A: Flit is a local dictation app for Mac: hold a key, speak, and release, and finished text lands in any app, with nothing sent anywhere. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Mac dictation not working? The causes, in the order to check them https://flit.fyi/mac-dictation-not-working (Mac dictation, published 2026-10-04) Short answer: Mac Dictation usually fails for one of five reasons: Dictation is switched off, Voice Control is on (which replaces it), the shortcut is not the one you are pressing, the wrong microphone is selected, or your language needs an internet connection and you have none. All five are in System Settings → Keyboard, under Dictation, except Voice Control, which is in Accessibility. Flit, a local dictation app for Mac, is a separate way to dictate: you hold a key and speak, and finished text lands in any app, with nothing sent anywhere. How to fix dictation on a Mac: 1. Open System Settings → Keyboard and scroll to Dictation. Turn it on, and click Enable if asked. 2. Open System Settings → Accessibility → Voice Control. If it is on, turn it off. Apple says that while Voice Control is on, standard Dictation is not available. 3. Back in Keyboard → Dictation, look at Shortcut. Press exactly that shortcut, or choose a different one from the menu. On a keyboard with a Microphone key in the function row, press and release it; holding it starts Siri. 4. Check Microphone source. Choose your microphone by name instead of Automatic. 5. Open System Settings → Sound → Input, speak, and raise the input volume if the level barely moves. 6. Check Languages. Click Edit and make sure the language and region you speak are ticked. 7. Read the small text under the Dictation switch. If it does not say your dictation is processed on the device, you need an internet connection: check System Settings → Network. 8. Click into a text field before you start. Dictation types where the cursor is, and does nothing if there is no cursor. 9. If the Dictation switch is greyed out, the Mac is restricted: see Screen Time and work restrictions below. What you see, and what causes it: - Nothing happens when you press the shortcut | Dictation is off, or the shortcut is a different one | System Settings → Keyboard → Dictation - Dictation is missing or will not start, and a Voice Control microphone is on screen | Voice Control is on, and it replaces Dictation | System Settings → Accessibility → Voice Control - The Globe key opens the emoji picker or switches language | The key is set to do something else | Keyboard → "Press 🌐 key to" - Dictation starts, but no words appear | The wrong microphone, or the input volume is too low | Keyboard → Dictation → Microphone source, and Sound → Input - It types the wrong words, or another language | The language or region is wrong | Keyboard → Dictation → Languages - It works at home and fails on the train | Your language is sent to Apple and needs a connection | The text under the Dictation switch - It stops while you are thinking | Dictation stops after 30 seconds without speech | Why dictation keeps stopping - Sound through AirPods turns thin when you dictate | Bluetooth switches mode while the headphones’ microphone is in use | Dictating with AirPods - The Dictation switch is greyed out | Screen Time, or a restriction set by your employer or school | Screen Time → Content & Privacy If you want more than Dictation does: Flit is a separate dictation app, not a fix for Apple’s. You hold a key, speak, and let go, and text that is already punctuated and formatted lands where your cursor is. - One gesture: hold a key to talk, release to finish. Nothing to start and stop. - Everything runs on the Mac, so it works without a connection in every language it supports. - It does not stop when you pause to think while the key is held. - Finished text: punctuation, paragraphs, lists and numbers, in the tone of the app you are in. - Its own microphone setting, in Settings → General. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: Why is dictation not working on my Mac? A: The usual causes are that Dictation is off, Voice Control is on, the shortcut is different from the one you are pressing, or the wrong microphone is selected. All but Voice Control are in System Settings → Keyboard, under Dictation. - Q: How do I turn on dictation on a Mac? A: Open System Settings → Keyboard, scroll to Dictation, turn it on and click Enable. Then click into a text field and press the shortcut shown there, or choose Edit → Start Dictation. - Q: What is the keyboard shortcut for dictation on a Mac? A: It is whatever the Shortcut menu in System Settings → Keyboard → Dictation says, and you can change it there. On keyboards with a Microphone key in the function row, pressing and releasing that key starts dictation. - Q: Does Mac dictation need the internet? A: Sometimes. Keyboard settings says, under the Dictation switch, whether your dictation is processed on the device. If it is not, it is sent to Apple and needs a connection. - Q: Why is the Dictation option greyed out? A: It is restricted, either by Screen Time (Content & Privacy → Intelligence & Siri → Siri & Dictation) or, on a work or school Mac, by an administrator. - Q: Is there an app that is better than Mac dictation? A: It depends on what you need. Flit is a local dictation app for Mac that types finished, formatted text into any app while you hold a key, and runs entirely on the Mac. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to stop the Globe (Fn) key opening the emoji picker on a Mac https://flit.fyi/globe-key-opens-emoji-picker-mac (Mac dictation, published 2026-10-04) Short answer: Open System Settings → Keyboard and change "Press 🌐 key to" (or "Press fn key to") from Show Emoji & Symbols to Do Nothing. The key then does nothing on its own, and you can still open the emoji picker with Fn-E or Control-Command-Space. The same menu can make the key change the input source or start Dictation when pressed twice. Flit, a local dictation app for Mac, uses a held Globe key for dictation, which is why it asks for this setting to be Do Nothing. How to change what the Globe key does: 1. Open the Apple menu → System Settings. 2. Click Keyboard in the sidebar. 3. Find the menu labelled Press 🌐 key to. On a keyboard without a Globe symbol it reads "Press fn key to". 4. Choose Do Nothing. 5. To open emoji from now on, press Fn-E (🌐-E) or Control-Command-Space, or choose Edit → Emoji & Symbols. What "Press 🌐 key to" can do: - Change Input Source | Switches between keyboard layouts for writing in other languages | You type in more than one language - Show Emoji & Symbols | Opens the Character Viewer for emoji and symbols | You want one-key emoji - Start Dictation | Starts Apple’s Dictation when you press the key twice | You use the built-in Dictation - Do Nothing | Nothing happens | The key keeps getting in your way, or another app uses it What Flit does with the Globe key: In Flit, holding the Globe key is how you dictate: hold, speak, let go, and finished text lands at your cursor. With "Press 🌐 key to" set to Do Nothing, the key is free for that. - Hold to talk, release to finish. Double-tap Fn to keep recording hands-free. - Fn shortcuts still work. Flit sees that another key went down and treats it as a shortcut, not a dictation, so Fn-E still opens emoji and Fn with the arrows still pages. - Prefer to keep the Globe key for emoji? Choose a custom shortcut in Flit’s Settings → Keyboard and leave the key alone. - Everything runs on the Mac, and nothing you say is sent anywhere. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - A custom shortcut in Flit has to be a modifier plus a key, such as Option-Shift-Space. A single key on its own works only as the activation key: Globe, Right Command, Right Option or Right Control. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: How do I stop the Fn key opening emoji on my Mac? A: System Settings → Keyboard, then set "Press 🌐 key to" (or "Press fn key to") to Do Nothing. - Q: How do I open the emoji picker after turning the Globe key off? A: Press Fn-E or Control-Command-Space, or choose Edit → Emoji & Symbols in the menu bar. - Q: What is the Globe key on a Mac? A: It is the Fn key on newer Apple keyboards, marked with a globe. Held with another key it works as Fn. Pressed alone it does what "Press 🌐 key to" in Keyboard settings says: change the input source, show emoji, start Dictation, or nothing. - Q: Can I make the Globe key start dictation? A: Yes. Choose Start Dictation in the same menu, and Apple’s Dictation starts when you press the key twice. - Q: Is there a dictation app that works by holding the Globe key? A: Yes. Flit is a local dictation app for Mac: hold the Globe key, speak, and release, and finished text lands in any app. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Why Mac dictation keeps stopping, and how to keep it going https://flit.fyi/mac-dictation-keeps-stopping (Mac dictation, published 2026-10-04) Short answer: Mac Dictation stops on its own when it hears no speech for 30 seconds. That is by design: Apple’s page says "Dictation stops automatically when no speech is detected for 30 seconds", and also that there is no limit on how long you can dictate if you keep speaking. If it stops sooner, something ended it: the Escape key or the dictation shortcut, a lost internet connection for a language that needs one, or the microphone changing. Flit, a local dictation app for Mac, records for as long as you hold the key, pauses included. How to stop dictation cutting out: 1. Know the rule: 30 seconds of silence ends it. If you need to think for longer, expect to press the shortcut again. 2. Do not press Escape, the Microphone key or the dictation shortcut while it is listening. Each of them stops dictation. 3. On a Mac with Apple silicon you can type while dictating, so fix a word with the keyboard instead of stopping. 4. Open System Settings → Keyboard and read the text under Dictation. If your dictation is not processed on the device, a dropped connection ends it: use a steadier network. 5. Set Microphone source to your microphone by name. On Automatic, the Mac can switch to another device, such as headphones that just connected. 6. If it stops with no words typed, check the device and the level in System Settings → Sound → Input. Dictation that cannot hear you has nothing to transcribe. When it stops, and why: - After you pause to think | No speech for 30 seconds | Start it again, or say something before the pause runs out - The moment you press a key | Escape, the Microphone key or the shortcut ends dictation | Avoid those three; other keys are fine on Apple silicon - Mid-sentence, on a poor connection | Your language is processed by Apple and the connection dropped | Check the text under Dictation in Keyboard settings - When headphones connect or disconnect | The microphone source changed | Choose the microphone by name, not Automatic - Straight away, with no words typed | The microphone is not picking you up | Sound → Input: check the device and the level Flit records for as long as you hold the key: Flit does not listen for silence to decide you are finished. You hold a key while you speak and release it when you are done, so a long pause in the middle costs nothing. - Hold to talk: the recording ends when you let go, not when you pause. - Hands-free for long passages: double-tap Fn, speak, and press once to finish. Flit asks whether to finish only after three minutes of quiet. - Hesitations and false starts are taken out, so thinking aloud does not show in the text. - Everything runs on the Mac, so a dropped connection cannot end a dictation. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: Why does my Mac dictation stop after 30 seconds? A: Because it heard no speech for 30 seconds. Apple’s page says Dictation stops automatically in that case. It does not stop while you keep talking. - Q: Is there a time limit on Mac dictation? A: Not on length. Apple says you can dictate text of any length without a timeout. The only automatic stop is 30 seconds without speech. - Q: Can I change how long Mac dictation waits before stopping? A: No. There is no setting for the 30-second silence limit. - Q: Why does dictation turn off when I type? A: On a Mac with Apple silicon, Apple says you can keep using the keyboard while you dictate. Escape, the Microphone key and the dictation shortcut do stop it. - Q: Is there a dictation app that does not stop when I pause? A: Yes. Flit is a local dictation app for Mac that records for as long as you hold the key, however long you pause. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Mac dictation commands: the full list, and what they cannot do https://flit.fyi/mac-dictation-commands (Mac dictation, published 2026-10-06) Short answer: Mac dictation commands are plain names you say in the middle of a sentence: "comma", "period", "question mark", "new line", "new paragraph", "all caps", "numeral". They are always available while Dictation is listening, with nothing extra to turn on. They insert characters and formatting only; selecting, deleting and replacing text by voice needs Voice Control. Flit, a local dictation app for Mac, punctuates and formats what you say without commands, and has more than 60 spoken commands of its own for when you want a mark exactly there. How to use a dictation command: 1. Turn on Dictation in System Settings → Keyboard, if it is not on already. 2. Click where you want the text and start Dictation with the Microphone key, the Dictation shortcut, or Edit → Start Dictation. 3. Speak, and say the command where the mark belongs, without a pause before it: "dear Sam comma new paragraph thanks for the notes period". 4. For capitals, say "all caps" before one word, or "all caps on" and "all caps off" around a phrase. 5. For digits, say "numeral" before the number. 6. Press Escape to stop. Line breaks you asked for appear now. The commands you will use most: - "comma" | , | Added for you when Auto-punctuation is on - "period", "point", "dot" or "full stop" | . | Added for you when Auto-punctuation is on - "question mark" | ? | Added for you when Auto-punctuation is on - "exclamation mark" | ! | Always spoken - "colon", "semicolon" | : ; | Always spoken - "quote" … "end quote" | “ ” | Say both - "open parenthesis" … "close parenthesis" | ( ) | Say both - "new line" | One Return | Appears when you finish dictating - "new paragraph" | Two Returns | Appears when you finish dictating - "all caps" | The next word in capitals | "all caps on" and "all caps off" for a phrase - "caps on" … "caps off" | A phrase in title case | Stays on until you say "caps off" - "numeral" | The next phrase as a number | "roman numeral" for Roman numerals Finished text without saying the marks: Flit works out the punctuation, paragraphs, lists and numbers from what you said, on the Mac, so most dictation needs no commands. When you want a mark in an exact place, you say it. - Punctuation and paragraphs are added for you, and hesitations and false starts are taken out. - More than 60 spoken commands for punctuation, symbols, casing and structure. - The whole text, line breaks included, lands in one piece when you release the key. - Everything runs on the Mac, and nothing you say is sent anywhere. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does not control your Mac by voice, and it does not select or delete text on command. It types what you say. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: What are the dictation commands on a Mac? A: They are the words you say during Dictation to insert punctuation, symbols and formatting, such as "comma", "new paragraph", "all caps" and "numeral". Apple lists them in the Mac User Guide under "Commands for dictating text on Mac". - Q: How do I start a new paragraph when dictating on a Mac? A: Say "new paragraph". Say "new line" for a single line break. Both appear when you finish dictating. - Q: How do I dictate punctuation on a Mac? A: Say the name of the mark, such as "comma", "period" or "question mark". In supported languages, Auto-punctuation adds those three for you. - Q: How do I delete a word with Mac dictation? A: You cannot with standard Dictation, whose commands only insert characters and formatting. Type the correction, or use Voice Control, which has editing commands. - Q: How do I dictate in capital letters on a Mac? A: Say "all caps" before one word. For a phrase, say "all caps on", speak it, then say "all caps off". - Q: Is there a dictation app that adds punctuation without commands? A: Yes. Flit is a local dictation app for Mac that punctuates and formats what you say on the Mac itself, and keeps spoken commands for when you want a mark exactly there. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Voice Control vs Dictation on a Mac: which one you want https://flit.fyi/voice-control-vs-dictation-mac (Mac dictation, published 2026-10-06) Short answer: Dictation types what you say and does nothing else: you start it with a key, speak, and it stops. Voice Control is an accessibility feature for running the whole Mac by voice: it listens all the time, opens apps, clicks buttons and edits text on command. They cannot both be on. Apple says that when Voice Control is on, standard Dictation is not available. If you only want to type by voice, use Dictation. Flit, a local dictation app for Mac, stands in for Dictation, not for Voice Control: hold a key, speak, and finished text lands in any app. How to switch from Voice Control back to Dictation: 1. Choose Apple menu → System Settings and click Accessibility in the sidebar. 2. Click Voice Control and turn it off. 3. Click Keyboard in the sidebar and go to Dictation. 4. Turn Dictation on, and click Enable if asked. 5. Click into a text field and press the Dictation shortcut shown there, or choose Edit → Start Dictation. Dictation and Voice Control, side by side: - What it is for | Entering text by voice | Operating the Mac by voice, text included - Where it is | System Settings → Keyboard → Dictation | System Settings → Accessibility → Voice Control - How it starts | A key, a shortcut, or Edit → Start Dictation | It listens from the moment it is on - When it stops | When you stop it, or after 30 seconds without speech | When you say "Stop listening" or turn it off - Punctuation | Spoken commands, plus Auto-punctuation in supported languages | Spoken commands - Editing by voice | No. Commands insert characters and formatting | Yes, such as "Replace cat with dog" - Opening apps and clicking | No | Yes: "Open Mail", "Click Done", "Scroll down" - Internet connection | Depends on the Mac and the language; Keyboard settings says which | Needed once, for a file download when it is first turned on Where Flit fits: Flit stands in for Dictation, not for Voice Control. It types what you say into any app while you hold a key. Unless you turn on its voice activation, it records only while that key is down. - Hold a key to talk and release to finish. Nothing you say afterwards is recorded. - Finished text: punctuation, paragraphs, lists and numbers, without saying the marks. - It records for as long as you hold the key, pauses included. - Everything runs on the Mac, and nothing you say is sent anywhere. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does not control your Mac by voice: it does not open apps, click buttons, or select and delete text on command. If you need that, Voice Control is the tool. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: What is the difference between Voice Control and Dictation on a Mac? A: Dictation types what you say after you press a key. Voice Control is an accessibility feature that listens all the time and lets you run the whole Mac by voice, including dictating and editing text. - Q: Can I use Voice Control and Dictation at the same time? A: No. Apple says that when Voice Control is on, you use Voice Control to dictate text, and standard macOS Dictation is not available. - Q: Why did Dictation stop working when I turned on Voice Control? A: Because Voice Control replaces it. Turn Voice Control off in System Settings → Accessibility → Voice Control to get standard Dictation back. - Q: Does Voice Control need the internet? A: Only the first time. Apple says the Mac must be online for a one-time file download, and that afterwards no connection is needed. - Q: Is Voice Control the same as Siri or VoiceOver? A: No. Siri is an assistant you ask for things. VoiceOver is a screen reader that speaks what is on screen. Voice Control takes spoken commands to operate the Mac. - Q: Is there an app that only types what I say, without controlling the Mac? A: Yes. Flit is a local dictation app for Mac: hold a key, speak, and release, and finished text lands in any app. With hold-to-talk it records only while the key is down. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### The Mac dictation keyboard shortcut: what it is, and how to change it https://flit.fyi/mac-dictation-keyboard-shortcut (Mac dictation, published 2026-10-06) Short answer: The Mac dictation shortcut is not fixed: it is whatever the Shortcut menu in System Settings → Keyboard → Dictation says. You can pick another from that menu, or choose Customize and press your own keys, such as Option-Z. On keyboards with a Microphone key in the function row, pressing and releasing that key also starts Dictation. The same shortcut stops it. Flit, a local dictation app for Mac, uses a held key instead: hold Fn or your own shortcut, speak, and let go. How to find or change the dictation shortcut: 1. Choose Apple menu → System Settings and click Keyboard in the sidebar. 2. Go to Dictation. If it is off, turn it on and click Enable. 3. Look at the Shortcut menu. What it shows is your shortcut now. 4. To change it, click the menu and choose another. 5. To set your own, choose Customize and press the keys you want. Apple’s example is Option-Z. 6. Click into a text field and press the shortcut to test it. Press it again to stop. Ways to start and stop Dictation: - The Microphone key | Press and release it, in the function row | Only on keyboards that have one. Holding it starts Siri - The Dictation shortcut | Whatever Keyboard → Dictation → Shortcut shows | Starts and stops - The Fn or Globe key, twice | Choose that option in the Shortcut menu | Also changes the "Press 🌐 key to" setting - Your own shortcut | Shortcut → Customize, then press the keys | For example Option-Z. It can also go on a mouse button - The menu bar | Edit → Start Dictation | Works without any shortcut - To stop | Escape, the Microphone key, or the shortcut again | It also stops after 30 seconds without speech A key you hold, not a shortcut you toggle: Flit is a separate dictation app built around one gesture. You hold a key while you speak and release it when you are done, so there is no second press to remember. - Hold Fn (the Globe key) to talk, release to finish. - Or set your own shortcut in Flit’s Settings → Keyboard and hold that instead. - For long passages, double-tap Fn to keep recording hands-free, and press once to finish. - Fn shortcuts still work: Flit sees the second key and does not start a dictation. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - A custom shortcut in Flit has to be a modifier plus a key, such as Option-Shift-Space. A single key on its own works only as the activation key: Fn, Right Command, Right Option or Right Control. - Using Fn in Flit means setting "Press 🌐 key to" to Do Nothing, so the key no longer opens emoji with a single press. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: What is the keyboard shortcut for dictation on a Mac? A: It is whatever the Shortcut menu in System Settings → Keyboard → Dictation shows. It varies by Mac and by what was chosen there. On keyboards with a Microphone key, pressing and releasing that key also starts Dictation. - Q: How do I change the dictation shortcut on a Mac? A: Open System Settings → Keyboard, go to Dictation, click the Shortcut menu and choose another, or choose Customize and press the keys you want. - Q: Which key is the dictation key on a MacBook? A: On keyboards that have it, it is the Microphone key in the function row. Press and release it. Holding it starts Siri. - Q: Can I start Mac dictation without a keyboard shortcut? A: Yes. Click into a text field and choose Edit → Start Dictation in the menu bar. - Q: Can I hold a key to dictate on a Mac, instead of pressing twice? A: Not with Apple’s Dictation, where the shortcut starts and stops it. Push-to-talk apps work that way. - Q: Is there a dictation app with a hold-to-talk key? A: Yes. Flit is a local dictation app for Mac: hold Fn or a shortcut you choose, speak, and release, and finished text lands in any app. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Dictating on a Mac with AirPods: choosing the microphone, and why the sound drops https://flit.fyi/airpods-dictation-mac (Mac dictation, published 2026-10-06) Short answer: Mac Dictation listens through whatever Microphone source says in System Settings → Keyboard → Dictation. To dictate through AirPods, choose them there by name; to keep them out of it, choose the Mac’s own microphone. If your music turns thin and quiet when you dictate, that is Bluetooth: Apple explains that when the headphones’ microphone is in use, Bluetooth switches to a mode with reduced audio quality until the microphone is released. Flit, a local dictation app for Mac, has its own microphone setting, so you can dictate through the Mac’s microphone while the AirPods keep playing. How to choose the microphone for Dictation: 1. Connect your AirPods to the Mac and put them in. 2. Choose Apple menu → System Settings and click Keyboard in the sidebar. 3. Go to Dictation and click the menu next to Microphone source. 4. To dictate through the AirPods, choose them by name. To keep listening to audio at full quality, choose the Mac’s built-in microphone instead. 5. Click into a text field, start Dictation and say a sentence to test it. 6. If no words appear, open System Settings → Sound → Input, select the same microphone, speak, and check that the input level moves. What you notice, and why: - Music goes thin and quiet when Dictation starts | The AirPods microphone is in use, so Bluetooth is in its two-way mode | Set Microphone source to the Mac’s microphone - Static or popping while dictating | Apple lists this as part of the same mode switch | The same, or pause the audio while you dictate - Sound stays poor after you stop | Something is still using the AirPods microphone | Quit apps that use it, close Sound settings, and select the AirPods again - Dictation hears the room, not you | Microphone source is on another device | Choose the AirPods by name - Dictation stops when the AirPods connect | On Automatic, the source can change | Choose a microphone by name, not Automatic - No words appear at all | The wrong input, or a low level | Sound → Input: check the device and the level A microphone setting of its own: Flit is a separate dictation app, and it chooses its microphone separately from the rest of the Mac, in Settings → General. - Pick the Mac’s built-in microphone in Flit, and the AirPods microphone is not used while you dictate. - Or pick the AirPods, if you are away from the Mac’s microphone. - With hold-to-talk, Flit records only while the key is down. - Everything runs on the Mac, and nothing you say is sent anywhere. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - If you choose the AirPods as Flit’s microphone, Bluetooth switches mode while you hold the key, the same as with any app. Flit cannot change that. - Flit does not set or change the Mic Mode in Control Center. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: Can I use AirPods for dictation on a Mac? A: Yes. Connect them, then choose them by name under System Settings → Keyboard → Dictation → Microphone source. - Q: Why does my music sound bad when I dictate with AirPods? A: Because the AirPods microphone is in use. Apple explains that Bluetooth then switches to a mode for speaking and listening together, in which audio quality and volume are reduced until the microphone is no longer in use. - Q: How do I stop my Mac using the AirPods microphone? A: For Dictation, set Microphone source in Keyboard → Dictation to the Mac’s built-in microphone. For everything else, choose the input under System Settings → Sound → Input. - Q: How do I change the microphone for dictation on a Mac? A: Open System Settings → Keyboard, go to Dictation, and choose a device from the Microphone source menu. - Q: Why does dictation not hear me through my AirPods? A: Check that the AirPods are the selected Microphone source, then open Sound → Input, select them, and see whether the input level moves when you speak. - Q: Is there a dictation app that lets me choose its microphone? A: Yes. Flit is a local dictation app for Mac with its own microphone setting, so it can listen through the Mac’s microphone while your AirPods keep playing. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to change the dictation language on a Mac, and use more than one https://flit.fyi/change-dictation-language-mac (Mac dictation, published 2026-10-09) Short answer: To change the dictation language on a Mac, open System Settings → Keyboard → Dictation, click Edit next to Languages, and tick each language and region you speak. With more than one ticked, switch as you dictate by clicking the language shown next to the cursor, or by pressing the Globe key and choosing one. Dictation listens for one language at a time and does not change by itself, so text in the wrong language means the wrong one is selected. Flit, a local dictation app for Mac, keeps a list of the languages you speak and switches between them with one shortcut, or works out the language of each dictation by itself. How to add a dictation language and switch to it: 1. Choose Apple menu → System Settings and click Keyboard in the sidebar. 2. Go to Dictation. If it is off, turn it on and click Enable. 3. Click Edit next to Languages, then tick each language and region you speak. 4. Click into a text field and start Dictation with your shortcut. 5. To change language, click the language shown next to the cursor and choose another. 6. Or press the Globe key, if your keyboard has one, and choose the language. Dictating in two languages: what you notice, and the fix: - Nonsense words in the other language | Dictation is listening for the language you used last | Click the language next to the cursor, or press the Globe key, and choose the right one - Your language is not offered when you switch | It was never added | Keyboard → Dictation → Languages → Edit - The right language, the wrong spellings | The region is wrong, such as English (United States) for a British speaker | Tick the right region in the same list - The Globe key changes the keyboard layout instead | The key is set to Change Input Source | Keyboard → "Press 🌐 key to" - One language works without a connection and the other does not | Not every language is processed on the Mac | Read the text under Dictation in Keyboard settings, and the list below - No commas or periods in one of your languages | Auto-punctuation works in some languages only | Say the marks: "comma", "period" A list of your languages, and one key to switch: Flit is a separate dictation app that treats speaking several languages as the normal case. You tell it once which languages you speak, and after that you either switch with one key or let it work out the language each time. - Add the languages you speak once, in Flit’s Settings → Models → Languages you speak. - Set a shortcut in Settings → Keyboard that moves to the next language from any app, with one press. Flit shows which one you are on. There is also a switch in the menu bar panel. - Or choose Auto, and Flit works out which of your languages each dictation is in. That takes a moment longer than a fixed language. - Everything runs on the Mac in every language, with no connection. That includes Polish, Czech, Greek, Hungarian, Romanian, Slovak, Ukrainian, Croatian and Portuguese, with the Parakeet Ultra voice model. - The voice models write their own punctuation in the languages they cover, so there is no second list to check. Omnilingual is the exception: it writes none, and Flit’s formatting adds it. - If the voice model you chose cannot transcribe a language, Flit uses another model you have installed, tells you which, and goes back to yours when you change language. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit also hears one language for each dictation. A sentence that changes language halfway is not handled better than it is by Apple’s Dictation. - Which languages you get depends on the voice model you choose. Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30 and Whisper 99. Omnilingual covers more than 1,600, with accuracy that varies by language. - Flit has no region setting. For a spelling you prefer, add the word to its dictionary. - The language shortcut does nothing while a dictation is in progress. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: How do I change the dictation language on a Mac? A: Open System Settings → Keyboard → Dictation, click Edit next to Languages, and tick the language and region you want. While dictating with more than one ticked, click the language shown next to the cursor and choose another, or press the Globe key and choose the language. - Q: Can you dictate in two languages on a Mac? A: Yes. Add both under System Settings → Keyboard → Dictation → Languages → Edit. Then switch as you dictate by clicking the language next to the cursor, or by pressing the Globe key and choosing one. - Q: Is there a keyboard shortcut to switch the dictation language on a Mac? A: Not a single press. Apple documents pressing the Globe key while dictating and then choosing the language. The Dictation shortcut itself only starts and stops Dictation. - Q: Does Mac Dictation detect the language automatically? A: Apple’s page describes switching by hand: clicking the language next to the cursor, or pressing the Globe key. It does not describe automatic detection. - Q: Why does Mac Dictation type nonsense when I speak another language? A: It is listening for a different language and writing the closest words it knows. Switch to the language you are speaking and say it again. - Q: Does Mac Dictation work offline in every language? A: No. For macOS 27, Apple lists 63 languages and regions for Dictation and 48 for on-device Dictation. Polish, Czech, Greek, Hungarian, Romanian, Ukrainian, Hindi and Portuguese for Portugal are among those not on the on-device list. Keyboard settings says, under Dictation, how yours is processed. - Q: Can I mix two languages in one sentence when dictating? A: Not reliably. Dictation listens for one language at a time, so a word from another language is usually written as something that sounds similar. Correct it by hand, or switch language for longer passages. - Q: Is there an app that switches dictation language automatically? A: Yes. Flit is a local dictation app for Mac: add the languages you speak, choose Auto, and it works out which one each dictation is in, or switch with one shortcut. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to start dictation with a mouse button on a Mac https://flit.fyi/mac-dictation-mouse-button (Mac dictation, published 2026-10-09) Short answer: macOS has no setting that starts Dictation from a mouse button: the Dictation shortcut is a keyboard shortcut. The route that works has two steps. Set a custom Dictation shortcut, such as Option-Z, in System Settings → Keyboard → Dictation → Shortcut, then use your mouse maker’s software to make a side button send that key combination. The button then starts Dictation, and a second press stops it. Flit, a local dictation app for Mac, lets you pick a side or middle button directly: hold it to talk and release it to insert the text. How to put Dictation on a mouse button: 1. Choose Apple menu → System Settings and click Keyboard in the sidebar. 2. Go to Dictation. If it is off, turn it on and click Enable. 3. Click the Shortcut menu, choose Customize, and press a key combination you do not use elsewhere, such as Option-Z. 4. Open the software that came with your mouse and select the button you want to use. 5. Choose the option that sends a keystroke, then press the same combination. 6. Click into a text field and press the mouse button. Press it again to stop. Ways to start dictation from a mouse button: - A setting in macOS | There is none: the Dictation shortcut takes keys only | Not available - Custom Dictation shortcut, sent by the mouse maker’s software | Press to start, press again to stop | A mouse with a spare button, and its maker’s software - The Microphone key or Fn twice, sent by the mouse | Often not possible: these are not ordinary key combinations | A custom shortcut, which the mouse can send - A dictation app with a mouse button setting | Depends on the app. In Flit: hold to talk, release to insert | The app, and Accessibility access for it Pick the button, then hold it to talk: Flit is a separate dictation app with a mouse button setting of its own, so there is no shortcut to invent and no mouse software to configure. - Open Flit’s Settings → Keyboard → Mouse button, click it, and press the side or middle button you want. - Hold the button to talk and release it to insert the text. There is no second press to remember. - For a long passage, tap the button once to keep recording hands-free, and press it again to finish. - The button works beside your keyboard key, so you can start from whichever hand is free. - Everything runs on the Mac, with no connection. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - While a button is set in Flit, that button does nothing else. A side button no longer goes back a page. - The left and right buttons cannot be chosen. - Flit has to receive the button press from macOS. If your mouse software has given the button another job, set it back to a plain button there first. - It needs Accessibility access, the same access Flit asks for to paste text. - Apple’s Dictation is free and already on your Mac. If it does what you need, you do not need Flit. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit’s spoken commands, such as "new paragraph", are in English today. Questions: - Q: Can I start dictation with a mouse button on a Mac? A: Not from a setting in macOS, where the Dictation shortcut is a keyboard shortcut. You can set a custom shortcut such as Option-Z, then use your mouse maker’s software to make a spare button send that combination. - Q: How do I assign Dictation to a mouse side button? A: In System Settings → Keyboard → Dictation, open the Shortcut menu, choose Customize and press a key combination. Then, in your mouse software, select the side button and assign that keystroke to it. - Q: Can I hold a mouse button to dictate, like push-to-talk? A: Not with Apple’s Dictation. Its shortcut starts and stops Dictation, so the button is a toggle: press to start and press again to stop. - Q: Can a mouse button press the Microphone key or Fn twice? A: Often not. Mouse software is built to send ordinary key combinations, and the Microphone key is a special key while Fn twice is a gesture on one key. Set a custom Dictation shortcut and send that. - Q: Does this work with a Magic Mouse? A: No. A Magic Mouse has no spare button to assign. Use the Dictation keyboard shortcut. - Q: Is there a dictation app with a mouse button for push-to-talk? A: Yes. Flit is a local dictation app for Mac: choose a side or middle button in its settings, hold it to talk, and release it to insert finished text in any app. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to transcribe an audio file on a Mac, offline and without uploading it https://flit.fyi/transcribe-audio-file-mac (Transcribe recordings, published 2026-10-12) Short answer: A Mac can turn a recording into text offline, without sending it anywhere. Voice Memos shows a transcript for its own recordings, where the feature is available. The open Whisper model runs on any audio file from Terminal, after a short install. Flit, a local dictation app for Mac, takes an audio or video file on its Transcribe tab and gives back a transcript with paragraphs, saved as a text file or exported as SRT and VTT, on the Mac itself. How to read the transcript of a recording in Voice Memos: 1. Open the Voice Memos app on your Mac. 2. Click the recording you want to read. 3. Click the Transcription button to view the transcript. 4. Select the text you want, Control-click it, and choose Copy. 5. Paste the text into Mail, a document, or any other app. 6. To find a recording by what was said, type the words into the search field. Voice Memos shows recordings with that text in the title or in the transcript. Three ways to transcribe a recording on a Mac: - Voice Memos | Nothing installed. A recording made in Voice Memos, in a region where transcription is available | A transcript to read, search and copy - Whisper in Terminal | Python, ffmpeg and a pip install. Comfort with a command line | A text file, plus SRT and VTT, for any audio file - Flit | The Flit app on macOS 14 or later, Apple silicon | A transcript with paragraphs from any audio or video file, saved as text, exported as Markdown, SRT or VTT Transcribe a file in Flit: Flit is a dictation app first: hold a key, speak, and finished text lands at your cursor. Since version 1.5 it also transcribes a recording you already have, with the same voice models, on the Mac. - Drop an audio or video file on Settings → Transcribe, on the Flit icon in the menu bar, or right-click it in Finder and choose Services → Transcribe with Flit. MP3, M4A, WAV, AIFF, FLAC, MP4 and MOV all work. - The transcript appears as it goes, in paragraphs that start where the speaker paused. A two-hour recording takes a few minutes at the Clean formatting level. - It is saved as a text file in a folder you choose and kept in the tab for later. Export it as Markdown, or as SRT and VTT subtitles with timestamps. - Choose a voice model and a formatting level for recordings, separate from your dictation settings. Polished makes the recording read as prose. - From a script: install the flit command from Settings → Transcribe, and flit transcribe meeting.m4a --srt does the same from Terminal. - Nothing is uploaded. The file is read on the Mac, in pieces, and Flit transcribes with Wi-Fi off. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does not label speakers, and it does not record a meeting as it happens. Both are on its roadmap. - Which languages a recording can be in depends on the voice model you choose: Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99 and Omnilingual more than 1,600, with accuracy that varies by language. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. Questions: - Q: How do I transcribe an audio file on a Mac for free? A: Two ways cost nothing. Voice Memos shows a transcript when you click the Transcription button on a recording, where the feature is available. The open Whisper model installs with pip and runs on any audio file from Terminal. - Q: Can I convert audio to text offline on a Mac? A: Yes. Whisper runs from Terminal once it is installed, and Flit transcribes any audio or video file with Wi-Fi off. Voice Memos shows a transcript of its own recordings where the feature is available, though Apple does not say whether that happens on the Mac. - Q: Can a Mac transcribe audio without uploading it? A: Yes. Voice Memos, Whisper in Terminal and Flit all run on the Mac itself. On Apple silicon the models are fast enough that a two-hour recording takes minutes, and the file never leaves the disk. - Q: Can I transcribe a video file on a Mac? A: Whisper reads video through ffmpeg, and Flit takes MP4 and MOV directly. Both use the sound and ignore the picture. Voice Memos only transcribes its own recordings. - Q: Which Mac app gives speaker labels? A: None of the three here. MacWhisper labels speakers in its Pro version, on the Mac, and meeting tools such as Otter do it in the cloud. - Q: Is there an app that transcribes audio files on a Mac? A: Yes. Flit is a local dictation app for Mac with a Transcribe tab: drop an audio or video file in and the transcript comes back on the Mac, saved as text and exported as Markdown, SRT or VTT. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Wispr Flow not working on Mac? The causes, and the fix for each https://flit.fyi/wispr-flow-not-working (Other dictation apps, published 2026-10-04) Short answer: Wispr Flow transcribes on its servers, so the first things to check are its status page and your own connection, including any VPN or firewall. If those are fine, the cause is on your Mac: the Accessibility or Microphone permission, the wrong input device, a shortcut blocked by Secure Keyboard Entry, or a sign-in that timed out. Flit, a local dictation app for Mac, transcribes on the Mac itself, so it has no server to be down and works with Wi-Fi off. How to fix Wispr Flow when it stops working: 1. Open Wispr Flow's status page. If it shows an incident, the fault is on their side and there is nothing to fix on your Mac. 2. Check your own connection. Wispr Flow says its transcription always happens in the cloud, so a dictation started offline fails. Reconnect, then retry it from History in the app. 3. Turn off your VPN or proxy and try again. If that fixes it, add Wispr Flow to the VPN or firewall allowlist. This is the cause behind "Connection lost" and "Couldn’t reach Wispr Flow". 4. Quit Wispr Flow from its menu bar icon and open it again. Closing the window does not quit it. Then install any update it offers. 5. If you spoke and nothing appeared, press ⌃⌘V to paste the last transcript. Then open System Settings → Privacy & Security → Accessibility and turn Wispr Flow on, or off and on again. 6. If it cannot hear you, open System Settings → Privacy & Security → Microphone and turn Wispr Flow on. In Wispr Flow, open Settings → General → Microphone and pick your microphone by name. 7. If Fn + Space or Escape stopped responding, a password manager or Terminal is holding Secure Keyboard Entry. Quit that app, or in Terminal untick Secure Keyboard Entry, then lock and unlock your Mac. 8. If you cannot sign in, start "Sign in via browser" again. The browser step times out after five minutes. What you see, and what causes it: - "Taking longer than usual", or a transcript that never arrives | A slow connection, or an incident on the service | Check the status page and your connection, then retry from History - "Connection lost" or "Couldn’t reach Wispr Flow" | A VPN, proxy, firewall or DNS filter is blocking it | Turn it off to test, then allowlist Wispr Flow - You spoke, and nothing was typed | The Accessibility permission was lost, often after an update | ⌃⌘V to paste the last transcript, then re-enable Accessibility - "No microphone detected" or "We couldn’t hear you" | The Microphone permission, or the wrong input device | Privacy & Security → Microphone, then pick the microphone in Wispr Flow - Fn + Space or Escape do nothing, but holding Fn works | Secure Keyboard Entry, held by a password manager or Terminal | Quit that app, then lock and unlock the Mac - The sign-in page spins or fails | The browser sign-in timed out after five minutes | Start "Sign in via browser" again - A long recording is cut off | Desktop recordings are capped at 20 minutes | Dictate in shorter stretches What Flit does differently: Flit turns speech into text on your Mac, so the two biggest causes on this page, the connection and the service, do not apply to it. - Transcription and formatting run on the Mac. There is no status page because there is no service. - It works with Wi-Fi off, on a plane, or behind any VPN or firewall. - No account, so there is no sign-in to time out. - Text lands in about 0.7 seconds, the same whether your connection is fast or gone. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit also needs the Microphone permission, and Accessibility to type for you. If macOS drops Accessibility, Flit can still copy the transcript to your clipboard. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: Is Wispr Flow down right now? A: Check its status page, which lists current and past incidents. If it shows none, test your own connection with any VPN turned off. - Q: Why is Wispr Flow not typing anything? A: Usually the Accessibility permission was lost, often after an update. Press ⌃⌘V to paste the last transcript, then turn Wispr Flow off and on in System Settings → Privacy & Security → Accessibility. - Q: Why does Wispr Flow say it could not hear me? A: Either the Microphone permission is off, or the app is listening to the wrong input. Turn it on in Privacy & Security → Microphone, then choose your microphone by name in Wispr Flow’s Settings → General → Microphone. - Q: Does Wispr Flow work without internet? A: No. Its privacy page says transcription always happens in the cloud. The offline guide covers what happens when you dictate with no connection. - Q: Is there a dictation app that does not depend on a server? A: Yes. Flit is a local dictation app for Mac: the voice model runs on the Mac, so it works offline and has no service that can be slow or down. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Does Wispr Flow work offline? No, and what to use when you have no connection https://flit.fyi/does-wispr-flow-work-offline (Other dictation apps, published 2026-10-04) Short answer: No. Wispr Flow has no offline mode. Its privacy page says "transcription always happens in the cloud", and its help page lists starting a dictation offline as a cause of failure. The audio is kept so you can retry once you are back online, for up to 14 days. To dictate with no connection on a Mac, use Apple’s built-in Dictation, which works without a connection when Keyboard settings says it is processed on your device, or Flit, a local dictation app for Mac that runs its voice models on the Mac and works with Wi-Fi off. What to do when you need to dictate offline: 1. If a Wispr Flow dictation failed while you were offline, reconnect, open History in the app and retry it. The audio is kept for 14 days. 2. For a quick note with nothing installed, use Apple’s Dictation: System Settings → Keyboard → Dictation. The text under the switch says whether it is processed on your device or needs a connection. 3. For dictation that types finished text into any app with no connection, install an app that runs its model on the Mac. Download the model while you are still online. 4. Test before you travel: turn Wi-Fi off and dictate one sentence. An app that works offline does not change when you do this. Dictating on a Mac with no connection: - Wispr Flow | No. Transcription is always in the cloud | A dictation started offline fails and can be retried from History for 14 days - Apple Dictation | For some languages. Keyboard settings says whether yours is processed on the device | Built in and free. Types what you say, with automatic commas, periods and question marks - Flit | Yes, always, once the voice model is downloaded | Finished text in any app; nothing you say leaves the Mac Flit works with Wi-Fi off: Flit runs the voice model and the formatting model on your Mac, so being offline changes nothing: hold a key, speak, and finished text lands at your cursor. - Every dictation is transcribed on the Mac, online or not. - Text lands in about 0.7 seconds, with no round trip to wait for. - Voice models you choose, covering more than 1,600 languages. - No account and no word limit. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit uses the network for three things only: downloading a model you choose, activating a license once, and checking for updates if you allow it. The privacy page lists them in full. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: Does Wispr Flow have an offline mode? A: No. Its privacy page says transcription always happens in the cloud, and there is no setting that changes that. - Q: What happens if I use Wispr Flow without internet? A: The dictation fails. The app keeps the audio, and you can retry it from History once you are back online, for up to 14 days. - Q: Does Apple Dictation work offline? A: It depends on the language. Open System Settings → Keyboard and read the text under Dictation: it says whether your dictation is processed on the device or sent to Apple, in which case it needs a connection. The Mac dictation guide has the details. - Q: Is there an app like Wispr Flow that works offline? A: Yes. Flit is a local dictation app for Mac that does the same job, finished text in any app while you hold a key, with the models running on the Mac. It works with Wi-Fi off. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. - Q: How can I check that a dictation app is really offline? A: Turn Wi-Fi off and dictate. If the text still appears, the speech model is on your Mac. For a stricter check, watch the app in a network monitor such as Little Snitch while you dictate. ### Is Wispr Flow safe? What it sends to the cloud, and how to limit it https://flit.fyi/is-wispr-flow-safe (Other dictation apps, published 2026-10-04) Short answer: Wispr Flow is a cloud service, so using it means sending your voice to its servers: its privacy page says transcription always happens in the cloud. With Context Awareness on, it also sends text from the app you are in. It says data is encrypted, that password fields are never read, and that it holds SOC 2 Type II, ISO 27001 and HIPAA attestations. Two settings limit what is shared: Context awareness, and model training. Whether that is safe enough depends on what you dictate. Flit, a local dictation app for Mac, transcribes on the Mac, so nothing you say is sent anywhere. How to limit what Wispr Flow shares: 1. Open Wispr Flow, then Settings → Data and Privacy. 2. Turn off Context awareness. This stops the app sending text from the app and text field you are dictating into. 3. Turn off the model-training option ("Improve the model for everyone"). Wispr Flow says this is a separate switch: turning it off does not turn off Context awareness. 4. For anything you would not put in an email to a stranger, such as client names, health details or credentials, do not dictate it into a cloud tool at all. What Wispr Flow says about your data: - Where is my voice transcribed? | "Transcription always happens in the cloud" | Privacy page - What else is sent? | With Context Awareness on: app information, the text before and after your cursor, text on screen, variable and file names in coding apps, conversation history | Help: Context Awareness - What is never read? | Password fields. Context reading is skipped in banking and financial apps | Help: Context Awareness - Is my data used for training? | Your choice, under Settings → Data and Privacy | Privacy page - Who else processes it? | Third-party AI providers, under zero-data-retention agreements, with exceptions for some features. They are not named | Data controls - How long is it kept? | Data shared with third-party models is "generally deleted within 30 days" | Privacy policy, 19 August 2026 - Which certificates? | SOC 2 Type II, ISO 27001, HIPAA | Privacy page Flit sends nothing you say: Flit records, transcribes and formats on your Mac. There is no server that receives your audio or your text, so there is no policy to read about what a server does with them. - Your voice and your text never leave the Mac. You can verify it with Wi-Fi off. - No account, no telemetry and no analytics. - Reading the screen for context is off until you turn it on, and what it reads stays in memory on the Mac for one dictation. - History is a local file you can cap, clear, or turn off. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does connect to the network for three things: a model download you start, license activation, and the update check if you allow it. None carries anything you dictate. The privacy page lists each one. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: Does Wispr Flow record everything I say? A: It records while you are dictating, and that audio is sent to its servers to be transcribed. Its privacy page says transcription always happens in the cloud. - Q: Does Wispr Flow read my screen? A: With Context Awareness on, it reads text from the app you are dictating into, including the text around your cursor and text on screen. It says password fields are never read. You can turn it off under Settings → Data and Privacy. - Q: Is Wispr Flow HIPAA compliant? A: Wispr Flow says it supports HIPAA, with HIPAA enforcement on its Growth plan. Whether that meets your obligations is a question for your compliance lead, and it needs a signed agreement, not a web page. - Q: Does turning off model training make Wispr Flow private? A: It stops your audio and edits being used to train models. It does not stop the audio being sent for transcription, and it does not turn off Context Awareness, which is a separate switch. - Q: Is there a dictation app that keeps everything on my Mac? A: Yes. Flit is a local dictation app for Mac: the voice and formatting models run on the Mac and nothing you say is sent anywhere. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to uninstall Wispr Flow on a Mac completely https://flit.fyi/uninstall-wispr-flow-mac (Other dictation apps, published 2026-10-04) Short answer: Dragging Wispr Flow to the Trash leaves its data, login item and permissions behind. To remove it fully: quit it from the menu bar, stop "Wispr Flow" and "Wispr Flow Helper" in Activity Monitor, move the app to the Trash, delete its folders in your Library, remove it from Login Items, and clear its entries under Privacy & Security. Uninstalling does not cancel a subscription, so cancel that first. If you are replacing it, Flit is a local dictation app for Mac that does the same job with no account to close. How to remove Wispr Flow from your Mac: 1. Cancel your subscription first if you have one. Deleting the app does not cancel it. 2. Click the Wispr Flow icon in the menu bar and quit it. 3. Open Activity Monitor, search for "Wispr", and quit Wispr Flow and Wispr Flow Helper if either is still listed. 4. In Finder, open Applications and move Wispr Flow to the Trash. 5. In Finder, choose Go → Go to Folder, type ~/Library and delete the Wispr Flow files listed below. 6. Open System Settings → General → Login Items & Extensions and remove Wispr Flow. 7. Open System Settings → Privacy & Security, then Accessibility and Microphone, and remove Wispr Flow from each. 8. Empty the Trash. The files Wispr Flow leaves in your Library: - Application Support | Wispr Flow and Flow | Settings, history and saved audio - Application Support/Caches | Wispr Flow | Cached files - Logs | Wispr Flow | Log files - Preferences | com.wispr.flowapp.plist, com.electron.wispr-flow.plist, com.electron.wispr-flow.accessibility-mac-app.plist | Preferences - Saved Application State | com.wispr.flowapp.savedState | Window state - HTTPStorages | com.wispr.flowapp | Network storage - Caches | com.wispr.flowapp and com.electron.wispr-flow | Cached files Replacing it? What Flit is: Flit does the job you installed Wispr Flow for, finished text in any app while you hold a key, and does all of it on your Mac. - Nothing you say is sent to a server, and it works with Wi-Fi off. - No account to create, and so none to delete later. - A native Mac app. Everything it stores is in one folder, ~/Library/Application Support/Flit. - No word limit and no subscription. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: Does deleting Wispr Flow cancel my subscription? A: No. Cancel it through whoever bills you: Wispr Flow’s billing page for a web or desktop purchase, or your Apple or Google subscriptions for a phone purchase. - Q: Where does Wispr Flow store its files on a Mac? A: In your user Library: mainly ~/Library/Application Support/Wispr Flow, plus preference files beginning com.wispr.flowapp and com.electron.wispr-flow, logs, caches and saved state. - Q: How do I stop Wispr Flow opening at login? A: System Settings → General → Login Items & Extensions, then remove Wispr Flow from "Open at Login". - Q: How do I delete my Wispr Flow account and data? A: Uninstalling removes what is on your Mac only. Deleting the account and the data held on Wispr Flow’s servers is a separate request to Wispr Flow, and neither cancels a subscription. - Q: Is there a dictation app with no account? A: Yes. Flit is a local dictation app for Mac with no account at all: you download it and dictate. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to cancel a Wispr Flow subscription, and what happens after https://flit.fyi/cancel-wispr-flow-subscription (Other dictation apps, published 2026-10-04) Short answer: You cancel Wispr Flow through whoever bills you. If you subscribed on the web or the desktop app, use the Cancel Subscription button on its billing page. If you subscribed on an iPhone, cancel in Settings → your name → Subscriptions; on Android, in Google Play. You keep access until the end of the period you paid for, then drop to the free plan of 2,000 words a week on desktop. Refunds for unused time are not automatic. Flit, a local dictation app for Mac, is paid for once, so there is nothing to cancel. How to cancel Wispr Flow: 1. Work out who bills you. Look at your receipt: it is from Wispr Flow (through Stripe), from Apple, or from Google. 2. Billed by Wispr Flow: open your account in the desktop app or on the web, find your plan, and click Cancel Subscription. 3. Billed by Apple: on your iPhone open Settings → your name → Subscriptions, choose Wispr Flow and cancel. 4. Billed by Google: open Google Play → your profile → Payments and subscriptions → Subscriptions, choose Wispr Flow and cancel. 5. Note the date your access ends. You keep the paid plan until then. What happens when you cancel: - When does access stop? | At the end of the period you already paid for | Monthly: within a month. Annual: at the end of the year - What do I keep? | The account drops to the free plan | 2,000 words a week on desktop, 1,000 on mobile - Is my data deleted? | No. Cancelling does not delete your data | Account deletion is a separate step - Do I get a refund? | Not automatically. You have to ask support | Apple handles refunds for App Store purchases - What does Pro cost? | $15 a user a month, or $12 a month billed annually | So a year is $144 to $180 Flit is paid for once: Flit is $19 once for up to two Macs, and it keeps working for as long as you keep it. There is no monthly charge to cancel. - A license that never expires, with a year of updates included. Lifetime updates are a separate, one-time option. - No word limit, during the trial or after. - No account. Buying gives you a key, and the key is all there is. - A 30-day refund on a first purchase. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: How do I cancel Wispr Flow? A: Through whoever bills you: the Cancel Subscription button in Wispr Flow’s billing settings for a web or desktop purchase, Apple Subscriptions for an iPhone purchase, or Google Play for Android. - Q: Does uninstalling Wispr Flow cancel the subscription? A: No. Neither does signing out or deleting the account, for App Store and Google Play purchases. Cancel the subscription itself. The uninstall guide covers removing the app. - Q: Will I get a refund if I cancel Wispr Flow? A: Not automatically. Its help page says to contact support for a refund of unused time, and Apple handles refunds for App Store purchases. - Q: Is Wispr Flow free after I cancel? A: Yes, with a limit: 2,000 words a week on desktop and 1,000 on mobile, according to its pricing page. - Q: Is there a dictation app without a subscription? A: Yes. Flit is a local dictation app for Mac that you pay for once. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Claude Code voice dictation: /voice, Mac Dictation, or a local app https://flit.fyi/dictate-to-claude-code (Dictate to AI tools, published 2026-10-08) Short answer: Claude Code has a voice mode built in, called voice dictation: type /voice, then hold Space while you speak, and the transcript is inserted at your cursor. It needs a Claude.ai login and a local microphone, and Anthropic’s documentation says the audio is streamed to its servers and is not processed locally. The Mac’s own Dictation also works in a terminal. Flit, a local dictation app for Mac, is the third way: hold a key and speak, and your exact words land in the prompt with your project’s file names spelled right, and no audio leaves the Mac. How to dictate to Claude Code with /voice: 1. Start Claude Code in your terminal, signed in with a Claude.ai account. 2. Type /voice and press Enter to turn dictation on. 3. Hold Space and speak. Recording runs while the key is held. 4. Release Space. The transcript is inserted at your cursor. 5. Read it, fix anything, and press Enter to send. 6. To tap instead of hold, type /voice tap: tap Space once on an empty prompt to start, and tap again to send. Three ways to dictate to Claude Code: - Claude Code’s /voice | Hold Space, or tap with /voice tap | On Anthropic’s servers - Mac Dictation | The Dictation shortcut in Keyboard settings | On the Mac or at Apple; Keyboard settings says which - Flit | Hold Fn or your own shortcut | On the Mac Dictating to Claude Code with Flit: Flit works at the level of the Mac, not of Claude Code, so it needs no account, no /voice and no connection. Hold a key, speak, let go, and the text is in the prompt. - Your exact words. In a terminal Flit does not reword what you said, so a command stays a command. - File names from your project. Flit reads the names in the terminal’s working directory, on the Mac, so "upload client dot swift" lands as UploadClient.swift. - Spoken symbols for flags and paths, such as "dash dash" and "tilde slash". - No audio leaves the Mac, and it works however you sign in to Claude Code, an API key included. - The same key in the Claude desktop app and in an editor’s chat panel, where Flit shapes what you say as a prompt. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - In a terminal you get exact words: Flit does not tidy a rambling brief into a clean paragraph there, as it does in a chat box. - Flit inserts by pasting, so Claude Code shows a long dictation as a pasted-text placeholder. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit types what you say. It does not talk back, read answers aloud or hold a spoken conversation with the tool. - Recognizing your project’s file names, and Flit’s spoken commands, work in English today. Questions: - Q: Does Claude Code have a voice mode? A: Yes. Type /voice to turn on voice dictation, then hold Space while you speak. It needs a Claude.ai account and a local microphone. - Q: Is Claude Code voice dictation processed locally? A: No. Anthropic’s documentation says voice dictation streams your recorded audio to Anthropic’s servers for transcription and that audio is not processed locally. - Q: Why does /voice not work in Claude Code? A: The documented causes are signing in with an API key or a cloud provider instead of a Claude.ai account, working over SSH or in a cloud session, having no microphone, or an organization policy that turns voice off. - Q: Does Claude Code voice dictation use up my usage? A: No. Anthropic says transcription does not consume Claude messages or tokens and does not count toward the limits shown in /usage. - Q: How do I add a line break in Claude Code without sending? A: Press Ctrl-J, or type a backslash and press Enter. Both work in every terminal. Shift-Enter works in some terminals without setup. - Q: Is there a dictation app for Claude Code that runs offline? A: Yes. Flit is a local dictation app for Mac: hold a key and speak, and your words land in the Claude Code prompt with your project’s file names spelled right. Nothing is sent anywhere. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Cursor voice input: how to dictate prompts, and what it needs https://flit.fyi/dictate-to-cursor (Dictate to AI tools, published 2026-10-08) Short answer: Cursor has voice input, also called Voice Mode, built in: click the microphone in the chat box, or press Cmd-Shift-Space, speak, and review the transcription before you send it. Cursor’s staff have said on its forum that transcription needs a paid plan, and its pages do not say where the audio is transcribed. The Mac’s own Dictation also types into the chat box. Flit, a local dictation app for Mac, works in all of Cursor: a prompt in the chat box, exact words in the editor and terminal, your file names spelled right, and no audio leaving the Mac. How to dictate a prompt in Cursor: 1. Open the chat in Cursor and click into the input. 2. Click the microphone icon in the chat input, or press Cmd-Shift-Space. 3. Speak. Name files and functions as you would to a colleague. 4. Stop, and read the transcription in the input. 5. Correct anything that was misheard, then send. Ways to dictate in Cursor: - Cursor’s voice input | The microphone in the chat box, or Cmd-Shift-Space | For the chat box. Staff say it needs a paid plan - Hold to talk in the Agents Window | Hold Ctrl-M | Added in Cursor 3.1. Records the whole clip, then transcribes - Mac Dictation | The Dictation shortcut in Keyboard settings | Types into the chat box; stops after 30 seconds of silence - Flit | Hold Fn or your own shortcut | Chat box, editor and terminal; runs on the Mac Dictating in Cursor with Flit: Flit reads what kind of field you are in and writes for it. That is decided by the field, on the Mac, each time you hold the key. - In the chat box, what you say is shaped as a prompt: punctuated, in paragraphs, with file names and identifiers kept as spoken. - In the editor pane and the integrated terminal, you get your exact words, and nothing is reworded. - File names from the open project. Flit reads them from the project on the Mac, so "user auth service dot swift" lands as UserAuthService.swift. - Spoken commands for casing and symbols, such as "camel case" and "dash dash". - No plan to check and no audio sent anywhere: everything runs on the Mac. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit is not a way to write code by voice. It types what you say; for syntax, the keyboard is still quicker. - Flit does not send the prompt for you. You press Return. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit types what you say. It does not talk back, read answers aloud or hold a spoken conversation with the tool. - Recognizing your project’s file names, and Flit’s spoken commands, work in English today. Questions: - Q: Does Cursor have voice input? A: Yes. Click the microphone icon in the chat input, or press Cmd-Shift-Space, which Cursor lists as Toggle Voice Mode. - Q: Is Cursor voice input free? A: Cursor’s documentation does not say. Its staff have said on the forum that voice transcription is available on the Pro plan and above. - Q: What is the keyboard shortcut for voice in Cursor? A: Cmd-Shift-Space toggles Voice Mode, and it can be remapped in Keyboard Shortcuts settings. In the Agents Window, holding Ctrl-M records while held. - Q: Where does Cursor send my voice? A: Cursor’s pages do not say where audio is transcribed. Its data-use page does not mention voice or audio. - Q: Can I use Mac Dictation in Cursor? A: Yes. Click into the chat input and press your Dictation shortcut. Apple’s Dictation types anywhere you can type. - Q: Is there a dictation app for Cursor that works offline? A: Yes. Flit is a local dictation app for Mac that writes a prompt in Cursor’s chat box and exact words in the editor and terminal, with your project’s file names spelled right. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Codex voice and dictation: the app, the command line, and the editor https://flit.fyi/dictate-to-codex (Dictate to AI tools, published 2026-10-08) Short answer: In the ChatGPT desktop app, which now contains Codex, select Dictate in the composer and start talking; your speech is transcribed into the composer for you to edit before sending. The Codex command line is different: its /voice command starts a spoken conversation, and there is no dictation into the prompt. For the terminal and the IDE extension, use the Mac’s Dictation or a dictation app. Flit, a local dictation app for Mac, types your exact words into the Codex prompt in any of the three, with your project’s file names spelled right and no audio leaving the Mac. How to dictate to Codex in the desktop app: 1. Open the ChatGPT desktop app and switch to Codex. 2. Click into the composer. 3. Select Dictate, then start talking. 4. Stop, and read the transcription in the composer. 5. Edit it if you need to, then send the prompt. Voice in each Codex: - ChatGPT desktop app, Codex mode | Dictate, in the composer | Dictation: speech becomes text you edit and send - ChatGPT desktop app, Codex mode | ChatGPT Voice (Start voice chat) | A live spoken conversation, on paid plans - Codex command line | /voice, toggled with F8 | A live spoken conversation, not dictation - Codex IDE extension | Nothing documented | Use Mac Dictation or a dictation app Dictating to Codex with Flit: Flit types into whichever Codex you use, because it works at the level of the Mac. Hold a key, speak, let go. - In a terminal, your exact words: nothing is reworded, so a command stays a command. - In the extension’s chat panel inside an editor, what you say is shaped as a prompt, with identifiers kept as spoken. - File names from your project, read on the Mac, so "orders controller dot rb" lands as orders_controller.rb when that file exists. - One key and one habit across the app, the command line and the editor. - No audio leaves the Mac. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - In a terminal you get exact words: Flit does not tidy a rambling brief into a clean paragraph there, as it does in a chat box. - Flit inserts by pasting, so a dictation arrives in the prompt as pasted text. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit types what you say. It does not talk back, read answers aloud or hold a spoken conversation with the tool. - Recognizing your project’s file names, and Flit’s spoken commands, work in English today. Questions: - Q: Does Codex have voice dictation? A: In the ChatGPT desktop app, yes: select Dictate in the composer. The Codex command line has /voice, which is a spoken conversation and not dictation. Nothing is documented for the IDE extension. - Q: What does /voice do in the Codex CLI? A: It starts or stops a live voice conversation with Codex, with a live transcript. F8 toggles it. It does not dictate text into the prompt. - Q: How do I dictate into the Codex CLI? A: Use a tool outside it. Click into the prompt and press the Mac’s Dictation shortcut, or use a dictation app that types at the cursor. - Q: Where does Codex send my dictation audio? A: OpenAI’s voice dictation page says recorded audio is sent to its models to be transcribed, and is kept for as long as the chat is in your history. - Q: What happened to the Codex app for Mac? A: In July 2026 it became the ChatGPT desktop app, which contains Chat, Work and Codex. The earlier ChatGPT app for Mac is now ChatGPT Classic. - Q: Is there a dictation app for Codex that runs on the Mac? A: Yes. Flit is a local dictation app for Mac that types your words into the Codex prompt in the app, the command line or the editor, with your project’s file names spelled right. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### ChatGPT dictation on a Mac: how to use it, and what happens to the audio https://flit.fyi/dictate-to-chatgpt-mac (Dictate to AI tools, published 2026-10-08) Short answer: In the ChatGPT desktop app for Mac, select Dictate in the composer, speak, and your words are transcribed into the composer for you to edit before sending. OpenAI says the recorded audio is sent to its models to be transcribed, and is kept for as long as the chat is in your history. Dictation is separate from ChatGPT Voice, which is a live spoken conversation. Flit, a local dictation app for Mac, types finished text into ChatGPT and every other app while you hold a key, and the audio never leaves the Mac. How to dictate to ChatGPT on a Mac: 1. Open the ChatGPT desktop app and start or open a chat. 2. Click into the composer. 3. Select Dictate and start talking. 4. Stop when you are done. The transcription appears in the composer. 5. Read it, edit it, and send. Dictation, Voice, and the other ways: - Dictate, in the composer | Turns speech into text you edit and send | To OpenAI’s models; kept with the chat - ChatGPT Voice | A live spoken conversation, on paid plans | To OpenAI - Mac Dictation | Types into the composer | On the Mac or at Apple; Keyboard settings says which - Flit | Types finished text into the composer | Stays on the Mac Dictating to ChatGPT with Flit: Flit turns your speech into text on the Mac and types it into the composer. ChatGPT receives the words you chose to send, and no recording. - Hold a key, speak, let go. The same gesture in ChatGPT, your browser, your editor and your mail. - Finished text: punctuation, paragraphs and lists, with hesitations and false starts taken out. - The audio is transcribed on the Mac and is not sent anywhere, so there is no recording to retain or delete. - A personal dictionary for your names and terms. - It works the same way at chatgpt.com in a browser. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - The text you send is still sent to OpenAI, as with anything you type. Flit keeps the audio on your Mac, not the prompt. - If your Flit shortcut is Option-Space, it collides with ChatGPT’s Chat Bar. Change one of them. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. - Flit types what you say. It does not talk back, read answers aloud or hold a spoken conversation with the tool. - Recognizing your project’s file names, and Flit’s spoken commands, work in English today. Questions: - Q: How do I use voice to text in ChatGPT on a Mac? A: In the ChatGPT desktop app, click into the composer, select Dictate, and start talking. The transcription appears in the composer for you to edit and send. - Q: Is ChatGPT dictation free? A: A June 2026 release note from OpenAI describes its dictation model rolling out across all plans. ChatGPT Voice, the spoken conversation, is listed for paid plans in the desktop app. - Q: Does ChatGPT save my dictation audio? A: Yes. OpenAI says audio from dictation is retained for as long as the chat is part of your chat history, and deleted within 30 days of your deleting the chat, with exceptions. - Q: What is the difference between ChatGPT Voice and dictation? A: Dictation turns your speech into text that you review, edit and send. Voice is a live spoken conversation in which ChatGPT answers aloud. - Q: What is the keyboard shortcut for dictation in ChatGPT on a Mac? A: OpenAI’s shortcut reference lists none for dictation. Voice chat is Control-Shift-V, and Option-Space opens the Chat Bar by default. - Q: Is there a way to dictate to ChatGPT without sending audio to OpenAI? A: Yes. Flit is a local dictation app for Mac that transcribes on the Mac and types the text into ChatGPT, so only the words you send leave your computer. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Superwhisper pricing: the free plan, Pro, and what the lifetime license costs https://flit.fyi/superwhisper-pricing (Other dictation apps, published 2026-10-12) Short answer: Superwhisper has a free plan and one paid plan, Pro, billed three ways: $8.49 a month, $84.99 a year, or $249.99 once for a lifetime license. All three give the same features. The free plan dictates without limit using local Whisper models; Pro adds cloud models, AI modes, file transcription and more. Students, educators and non-profits get 40% off, and there is a 30-day refund on every plan. Flit, a local dictation app for Mac, is priced differently: one payment, with no plan tiers, and everything runs on the Mac. How to pay the least for Superwhisper: 1. Start on the free plan. New users get 3,000 words with every Pro feature on, then the app falls back to Free and nothing is deleted. 2. Check whether Free is enough. It includes unlimited dictation with local Whisper models, custom vocabulary and text replacements. 3. If you qualify for the 40% discount, email Superwhisper from your academic or organization address before you buy. It is not applied afterwards. 4. Buy on superwhisper.com, not inside the iPhone or Android app. Superwhisper says the App Store and Google Play prices are higher because of the stores’ fees. 5. Choose how to be billed. Monthly costs more per year than Annual, and Lifetime is a single payment. 6. If it does not suit you, ask for a refund within 30 days of your first purchase. Superwhisper plans and prices: - Free | $0 | $0 - Pro, monthly | $8.49 a month | $305.64 - Pro, annual | $84.99 a year | $254.97 - Pro, lifetime | $249.99 once | $249.99 - Enterprise | Custom | Ask Superwhisper How Flit is priced, for comparison: Flit is a dictation app for the Mac only, sold as one payment. It is a narrower product than Superwhisper, and the price reflects that. - $19 once, for up to 2 of your own Macs. There is no monthly or yearly plan. - 12 months of updates are included. Another year is $19 and optional, and the app keeps working, with everything it has, if you never renew. - Over three years that is $19, or $19 plus two $19 renewals if you want every update. - Every feature is included, and everything runs on the Mac with no account. - Text appears as you speak, and is formatted on the Mac when you let go of the key. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Superwhisper’s free plan dictates with local Whisper models at no cost. If that is enough for you, you do not need to pay for anything, Flit included. - Superwhisper runs on Windows, iPhone and Android as well, on as many devices as you like. Flit covers 2 Macs and nothing else. - Superwhisper records meetings and separates speakers. Flit does not record meetings. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: How much does Superwhisper cost? A: Superwhisper is free with local Whisper models. Pro is $8.49 a month, $84.99 a year, or $249.99 once for a lifetime license, as of October 2026. - Q: Is Superwhisper free? A: Yes, there is a free plan with unlimited dictation using local Whisper models, up to two modes, custom vocabulary and text replacements. Cloud models, AI modes and file transcription need Pro. - Q: Does Superwhisper have a lifetime license? A: Yes. Pro Lifetime is $249.99 once. Superwhisper says lifetime keys never expire and a lost key can be recovered. - Q: Does Superwhisper have a student discount? A: Yes. Verified students, faculty and academic institutions get 40% off, and so do non-profits. Email Superwhisper from your academic address before you buy. - Q: Can I get a refund from Superwhisper? A: Yes, within 30 days of your first purchase on superwhisper.com, with no reason needed. Renewals are not separately refundable, and store purchases are refunded by Apple or Google. - Q: How many devices can I use Superwhisper Pro on? A: Superwhisper says one Pro license works on all your devices, with no device limit for Mac and Windows computers, plus the iPhone and Android apps. - Q: Is there an app like Superwhisper with no subscription? A: Superwhisper itself sells a lifetime license at $249.99. Flit is a local dictation app for Mac sold only as one payment. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Wispr Flow pricing: the free plan’s limit, Pro, Growth, and the student offer https://flit.fyi/wispr-flow-pricing (Other dictation apps, published 2026-10-12) Short answer: Wispr Flow has a free plan capped at 2,000 words a week on desktop and 1,000 on mobile. Pro removes the cap and costs $15 per user a month, or $12 a month billed annually. Growth, with single sign-on and admin controls, is $23 a month or $18 billed annually, and Enterprise is priced on request. Students get three months free and then pay $6 a month. Flit, a local dictation app for Mac, is priced differently: one payment, no word limit, and everything runs on the Mac. How to pay the least for Wispr Flow: 1. Start on the free plan. Wispr Flow says you never need a credit card to start. 2. Watch the weekly word count for a week or two. If you stay under 2,000 words on desktop, the free plan is enough. 3. If you are a student or educator, sign in with your education email. The student offer is three months free, then $6 a month. 4. If you do upgrade, choose annual billing: Pro is $12 a month billed annually, against $15 month to month. 5. Upgrade in the Flow app under Settings → Plan & Billing, or from the pricing page. 6. To stop, manage the subscription in the same place. The cancellation guide covers each way of being billed. Wispr Flow plans and prices: - Free | $0, capped at 2,000 words a week on desktop | $0 - Pro, monthly | $15 a month | $540 - Pro, annual | $12 a month, billed annually | $432 - Growth, monthly | $23 a month | $828 - Growth, annual | $18 a month, billed annually | $648 - Enterprise | Custom, available annually | Ask Wispr Flow How Flit is priced, for comparison: Flit is a dictation app for the Mac only, sold as one payment. Nothing is processed on a server, so there is no cost per sentence to pass on and no word limit. - $19 once, for up to 2 of your own Macs. There is no monthly or yearly plan. - 12 months of updates are included. Another year is $19 and optional, and the app keeps working, with everything it has, if you never renew. - Over three years that is $19, or $19 plus two $19 renewals if you want every update. - Every feature is included, and everything runs on the Mac with no account. - No weekly word limit, during the trial or after. - It works with no connection, since nothing is sent anywhere. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Wispr Flow’s free plan costs nothing. If you dictate less than 2,000 words a week and are comfortable with cloud transcription, it may be all you need. - Wispr Flow runs on Windows, iPhone and Android as well, and has a meeting Notetaker. Flit covers 2 Macs, and does not record meetings. - Flit has no shared team workspace, single sign-on or admin console. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: How much does Wispr Flow cost? A: Wispr Flow is free up to 2,000 words a week on desktop. Pro is $15 per user a month, or $12 a month billed annually. Growth is $23, or $18 billed annually. Enterprise is priced on request. These are the prices as of October 2026. - Q: Is Wispr Flow free? A: Yes, with a cap. The free plan allows 2,000 words a week on desktop and 1,000 a week on mobile, and no credit card is needed to start. - Q: What is the Wispr Flow free plan limit? A: 2,000 words a week on desktop and 1,000 words a week on mobile. Pro removes the limit. - Q: Does Wispr Flow have a student discount? A: Yes. Students and educators with an education email get three months of Flow Pro free, then pay $6 a month, which is 50% off. - Q: Does Wispr Flow have a lifetime license? A: No one-time price is listed on its pricing page. The paid plans are billed monthly or annually. - Q: Is Wispr Flow priced per user? A: Yes. Pro and Growth are priced per user per month, and a team can be created with no minimum number of seats. - Q: Is there an app like Wispr Flow with no subscription? A: Flit is a local dictation app for Mac sold as one payment, with no word limit and no account. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### Superwhisper not working? The fixes, by what you see https://flit.fyi/superwhisper-not-working (Other dictation apps, published 2026-10-13) Short answer: Most Superwhisper problems on a Mac come down to four things: it is listening to the wrong microphone, a macOS permission is missing (Microphone to hear you, Accessibility to paste), another app owns the shortcut, or a cloud model has no connection. Your words are rarely lost: Superwhisper saves each dictation in History, where it can be played back and reprocessed after the fix. Flit, a local dictation app for Mac, is a simpler setup: one voice model that you choose once, and formatting that runs on the Mac with no connection. How to fix Superwhisper on a Mac: 1. Open Superwhisper’s settings and look at the top-right corner of any pane. That is the active microphone. Switch it if it is the wrong one. 2. Open System Settings → Privacy & Security → Microphone and make sure Superwhisper is switched on. 3. Open System Settings → Privacy & Security → Accessibility and switch Superwhisper on there too. Without it, nothing can be pasted into other apps. 4. Start a recording and speak. If the waveform does not move, audio is not reaching the app: try the Mac’s built-in microphone. 5. If nothing happens when you press the shortcut, try an uncommon key combination. If that works, another app owns the first one. 6. If a dictation hangs on processing, check your connection, or switch that mode to a local model. 7. Quit and reopen Superwhisper, then retry the dictation from History. The audio is already saved. 8. Choose Check for Updates in Settings → Configuration. What you see, and what causes it: - "No voice found" | The recording had no detectable speech: the wrong or a muted microphone | Check the active microphone and the Microphone permission - The text is in History but was not pasted | Accessibility permission is off, or the cursor moved | Enable Superwhisper under Accessibility; keep focus in the field - Old clipboard text is pasted | The clipboard was restored before the app read it | Set Clipboard content to Replace with dictation - The text is pasted twice | Simulate keypresses is on and the app also accepted the paste | Turn Simulate keypresses off for that app - Nothing happens on the shortcut | Another app or the system owns the key | Pick a key no other app uses - Raw, unformatted text | The AI step failed, usually with no connection | Reconnect, or reprocess from History - Stuck on processing | No connection for a cloud model, or a local model still downloading | Check Settings → Models library - Slow, or heavy on memory | A local model too large for the Mac | Choose a smaller model, or keep it loaded for longer - The app will not open | macOS older than 14 | Update macOS, or install an older release If you want fewer things to configure: Flit is built for one thing, dictation into any app while you hold a key. It has no modes to set up and no cloud step, which removes several of the causes above. - One voice model, chosen once. There is no separate language model to pick, and no API key. - Formatting runs on the Mac, so being offline cannot leave you with raw text. - One gesture: hold a key to talk, release to finish. - Its own microphone setting, in Settings → General. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit needs the same two macOS permissions, Microphone and Accessibility, and it pastes at the cursor as Superwhisper does. Those causes do not go away. - Flit has no custom modes or prompts of your own, no cloud models, and it does not record meetings. If you use those in Superwhisper, Flit does not replace them. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: Why is Superwhisper not working on my Mac? A: The usual causes are the wrong microphone, a missing Microphone or Accessibility permission, a shortcut that another app owns, or a cloud model with no connection. Your dictation is normally still in History. - Q: What does "No voice found" mean in Superwhisper? A: That the recording had no detectable speech. Superwhisper says the cause is almost always the audio input: check the active microphone, the Microphone permission, and whether the waveform moves when you speak. - Q: Why does Superwhisper not paste my text? A: Most often because Accessibility permission is off. Enable Superwhisper under System Settings → Privacy & Security → Accessibility. The text is still in History and on your clipboard. - Q: Why did Superwhisper paste unformatted text? A: The AI formatting step failed, so it pasted the raw transcription. Most language models run in the cloud, so check your connection, then reprocess the dictation from History. - Q: Does Superwhisper work offline? A: With a local voice model, transcription does. Its documentation says most language models run in the cloud, so AI formatting needs a connection unless you choose a local language model. - Q: Which macOS does Superwhisper need? A: Current releases need macOS 14 or later. Superwhisper lists older releases for earlier versions of macOS. - Q: Is there a simpler dictation app than Superwhisper for Mac? A: Flit is a local dictation app for Mac with one voice model and no modes to configure: hold a key, speak, and finished text lands in any app. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to cancel a Superwhisper subscription, and what you keep https://flit.fyi/cancel-superwhisper-subscription (Other dictation apps, published 2026-10-13) Short answer: To cancel Superwhisper Pro, open superwhisper.com/billing, sign in with the email on your account, and choose Cancel subscription. If you subscribed in the iPhone or Android app, cancel through the App Store or Google Play instead, because Superwhisper cannot see those subscriptions. Pro stays active until the end of the period you have paid for, then the account moves to the free plan, and your modes, vocabulary and history stay on your device. Flit, a local dictation app for Mac, has no subscription to cancel: it is one payment. How to cancel Superwhisper: 1. Work out who bills you. If you bought on superwhisper.com, it is Superwhisper. If you bought inside the iPhone or Android app, it is Apple or Google. 2. If you bought on the website: open superwhisper.com/billing, sign in with the email on your account, and choose Cancel subscription. 3. If you bought on an iPhone: open the App Store, tap your profile picture, tap Subscriptions, select Superwhisper, and cancel there. 4. If you bought on Android: open the Play Store, tap your profile icon, then Payments & subscriptions → Subscriptions, select Superwhisper, and cancel there. 5. Check the date your paid period ends. Pro keeps working until then. 6. If you are inside the first 30 days and want your money back, request a refund as well. Cancelling alone does not refund anything. Where to cancel, by where you bought: - superwhisper.com | superwhisper.com/billing → Cancel subscription | Superwhisper, within 30 days of first purchase - The iPhone app | App Store → your profile → Subscriptions | Apple - The Android app | Play Store → Payments & subscriptions → Subscriptions | Google - An Enterprise plan | The organization’s admin, in the Enterprise dashboard | Your account manager If you are cancelling because of the subscription: Flit is sold as one payment, so there is nothing to cancel and no renewal date to watch. Superwhisper’s own Lifetime plan is the other way to stop paying monthly. - $19 once, for up to 2 of your own Macs. - 12 months of updates are included. Another year is $19 and optional, and nothing renews by itself. - No account. The license is a key, and everything runs on the Mac. - A 30-day refund on a first purchase. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Superwhisper’s free plan still dictates after you cancel. If local Whisper models are enough, you do not need to buy anything. - Flit has no custom modes, no cloud models and no meeting recording, and it runs on the Mac only. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: How do I cancel my Superwhisper subscription? A: Open superwhisper.com/billing, sign in with the email on your account, and choose Cancel subscription. If you bought in the iPhone or Android app, cancel in the App Store or Google Play. - Q: What happens when I cancel Superwhisper Pro? A: Pro features stay active until the end of the billing period you paid for, then the account moves to the free plan. Your modes, vocabulary and history stay on your device. - Q: Can I get a refund from Superwhisper? A: Yes, within 30 days of your first purchase on superwhisper.com, using its refund request form. Renewal charges are not separately refundable. - Q: Does deleting the Superwhisper app cancel my subscription? A: No. Billing is handled in the billing portal, or by Apple or Google. Cancel there first. - Q: How do I delete my Superwhisper account? A: Email support@superwhisper.com from the address on your account with the subject "Delete my account". Superwhisper says it completes the deletion within 30 days. - Q: Can I switch from a Superwhisper subscription to Lifetime? A: Yes. Buy Lifetime, then email Superwhisper’s support from the same address, and it refunds the unused part of your subscription. - Q: Is there a dictation app for Mac with no subscription? A: Superwhisper itself sells a Lifetime plan. Flit is a local dictation app for Mac sold only as one payment, with no account. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to uninstall Superwhisper on a Mac completely https://flit.fyi/uninstall-superwhisper-mac (Other dictation apps, published 2026-10-13) Short answer: To uninstall Superwhisper on a Mac, quit it from its menu bar icon and delete Superwhisper from Applications. That removes the app and leaves your data. To remove the data as well, delete the ~/superwhisper folder in your home directory, which holds your recordings, history, modes and vocabulary. Deleting the app does not cancel a subscription; that is done in the billing portal. Flit, a local dictation app for Mac, keeps everything it stores in one folder too, and has no account or subscription to close. How to uninstall Superwhisper on a Mac: 1. If you pay for Pro, cancel it first at superwhisper.com/billing, or in the App Store if you bought it there. 2. If you might come back, copy the ~/superwhisper folder somewhere safe. It holds everything you would want again. 3. Click the Superwhisper icon in the menu bar and choose Quit. 4. Open the Applications folder and move Superwhisper to the Trash. 5. To remove your data too, choose Go → Home in Finder and delete the superwhisper folder. 6. Empty the Trash. What each step removes: - The app, from Applications | Superwhisper itself | Recordings, history, modes, vocabulary - The ~/superwhisper folder | Recordings, history, modes, vocabulary, replacements | Your account and any subscription - The subscription, in the billing portal | Future charges | The account, on the free plan - The account, by email to support | Your email on Superwhisper’s records | Anonymized billing records Replacing it? What Flit is: Flit does the core job you installed Superwhisper for, finished text in any app while you hold a key, and does all of it on your Mac. - Nothing you say is sent to a server, and it works with Wi-Fi off. - No account to create, and so none to delete later. - A native Mac app. Everything it stores is in one folder, ~/Library/Application Support/Flit. - One payment, with no subscription to cancel before you uninstall. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit has no custom modes, no cloud models and no meeting recording. If you used those in Superwhisper, Flit does not replace them. - Flit runs on a Mac with Apple silicon and macOS 14 or later. There is no Windows, iPhone or Android app today; Windows is in the works. - The voice model is a download of about 0.5 to 2.8 GB that you choose once. After that Flit needs no connection. - Flit holds no SOC 2, ISO 27001 or HIPAA certificate. Nothing you say reaches a server, so there is no server to certify. If your organization requires a certificate, Flit does not have one. Questions: - Q: How do I completely uninstall Superwhisper on a Mac? A: Quit it from the menu bar icon, delete Superwhisper from Applications, then delete the ~/superwhisper folder in your home directory to remove your recordings, history and modes. - Q: Where does Superwhisper store its files on a Mac? A: In ~/superwhisper in your home directory. Older installs use ~/Documents/superwhisper, and a custom location is shown in Settings → Advanced settings under App folder location. - Q: Does uninstalling Superwhisper delete my recordings? A: No. Deleting the app leaves the data folder in place. Delete the ~/superwhisper folder to remove recordings and history. - Q: Does uninstalling Superwhisper cancel my subscription? A: No. Cancel at superwhisper.com/billing, or in the App Store or Google Play if you subscribed there. - Q: Can I reinstall Superwhisper and keep my settings? A: Yes, as long as you did not delete the data folder. Superwhisper says reinstalling the app never touches it. - Q: Is there a dictation app for Mac with no account? A: Yes. Flit is a local dictation app for Mac with no account and no subscription: hold a key, speak, and finished text lands in any app. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to transcribe a voice memo on a Mac https://flit.fyi/transcribe-voice-memo-mac (Transcribe recordings, published 2026-10-16) Short answer: Voice Memos on a Mac transcribes its own recordings. Click a recording, click the transcription button, and the text appears; select it, Control-click and choose Copy to paste it anywhere. It needs macOS 15 or later on a Mac with Apple silicon, and it is not available in every country or region. You can also watch the transcript appear while you record, and search all your recordings by what was said. Flit, a local dictation app for Mac, takes a voice memo dragged out to Finder and gives back a transcript in paragraphs, saved as a file and exported as Markdown or subtitles. How to transcribe a voice memo on a Mac: 1. Open Voice Memos on your Mac. 2. Click the recording you want. 3. Click the transcription button. The transcript replaces the waveform. 4. Select the text you want, Control-click it, and choose Copy. 5. Paste it into a document, a note or an email. 6. To find a recording by what was said, type the words into the search field. The list shows recordings with that text in the title or in the transcript. What Voice Memos does with a transcript: - Read it | Click the recording, then the transcription button | The current word is highlighted as it plays - Watch it as you record | Click the transcription button while recording | The waveform steps aside for the text - Copy it | Select, Control-click, Copy | Copying is the only way out that Apple describes - Search every recording | Type in the search field | Matches the title or the transcript - Jump to a moment | Select a word in the transcript | The playhead moves to that point - Summarize it | Writing Tools, with Apple Intelligence | Not in every language or region When you need more than a copy: Flit is a dictation app with a Transcribe tab. Drag a voice memo to Finder, drop it on Flit, and the transcript comes back on the Mac, with the things Voice Memos leaves out. - A transcript in paragraphs, split where the speaker paused, saved as a text file in a folder you choose. - Export as Markdown, or as SRT and VTT subtitles with timestamps. - A switch between the formatted text and exactly what was heard. - It takes recordings from anywhere: MP3, M4A, WAV, MP4, MOV and the rest of what macOS plays. - Nothing is uploaded, and it works with Wi-Fi off. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - For a short memo you only want to read, Voice Memos is already there and costs nothing. You do not need Flit for that. - Flit does not say who said what. A transcript is one flow of text in paragraphs. Speaker labels are planned. - Which languages a recording can be in depends on the voice model you choose: Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99 and Omnilingual more than 1,600, with accuracy that varies by language. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. Questions: - Q: How do I transcribe a voice memo on a Mac? A: Open Voice Memos, click the recording, then click the transcription button. Select the text, Control-click it and choose Copy to use it elsewhere. - Q: Why is there no transcript in Voice Memos on my Mac? A: Apple’s requirement is macOS 15 or later on a Mac with Apple silicon, and the feature is not available in every country or region. On an Intel Mac or an older macOS, there is no transcription button. - Q: Can I export a Voice Memos transcript? A: Apple describes copying it: select the text, Control-click and choose Copy. Its guide has no option to export the transcript as a file. - Q: Can Voice Memos transcribe an audio file I already have? A: Apple’s guide describes transcripts of recordings made in Voice Memos and has no way to import a file. Recordings made in macOS 14 or earlier are transcribed when you open them. - Q: Can I search voice memos by what was said? A: Yes. Type in the search field and the list shows recordings with that text in the title or the transcript. Selecting a word moves the playhead to it. - Q: Is there an app that turns a voice memo into subtitles? A: Flit is a local dictation app for Mac with a Transcribe tab: drop the memo on it and export SRT or VTT subtitles with timestamps, on the Mac. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to make subtitles (SRT) from a video on a Mac https://flit.fyi/make-subtitles-video-mac (Transcribe recordings, published 2026-10-16) Short answer: A Mac can make subtitles from the speech in a video three ways. Final Cut Pro’s Generate Captions creates captions in the timeline on a Mac with Apple silicon, in U.S. English only, which you convert to SRT and export with File → Export Closed Captions. The open Whisper model writes SRT and VTT files from Terminal. iMovie and QuickTime Player do not make subtitles. Flit, a local dictation app for Mac, takes the video on its Transcribe tab and exports SRT or VTT with timestamps, in any language its voice model covers, on the Mac. How to make subtitles in Final Cut Pro: 1. Open your project in Final Cut Pro. 2. Choose Edit → Generate Captions → Closed Captions, or press Shift-Command-C. 3. The first time, Final Cut Pro downloads its language model, which needs an internet connection once. 4. Check the captions in the timeline and correct any words. 5. Convert the captions to SRT, since generated captions are made in the iTT format. 6. Choose File → Export Closed Captions, pick SRT, and save the file. Three ways to make subtitles on a Mac: - Final Cut Pro | $299.99 once, Apple silicon, speech in U.S. English | Captions in the timeline, exported as SRT or iTT, or burned in as titles - Whisper in Terminal | Python, ffmpeg and a pip install | SRT, VTT and text files, in the languages Whisper knows - Flit | macOS 14 or later, Apple silicon | SRT and VTT with one line per sentence, plus text and Markdown Subtitles from the Transcribe tab: Flit is a dictation app with a Transcribe tab. Drop the video on it and export subtitles, with no editor to learn and nothing uploaded. - MP4 and MOV go straight in. The sound is read and the picture is left alone. - Export SRT or VTT subtitles with timestamps, one line per sentence, or the text as Markdown. - Speech in any language your voice model covers, not only U.S. English. - A two-hour recording took about three and a half minutes at the Clean formatting level in our tests. - From a script: flit transcribe talk.mp4 --srt --out ~/Subtitles. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit makes the subtitle file. It does not edit video or burn the subtitles into the picture, as Final Cut Pro can. - Flit does not say who said what. A transcript is one flow of text in paragraphs. Speaker labels are planned. - Which languages a recording can be in depends on the voice model you choose: Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99 and Omnilingual more than 1,600, with accuracy that varies by language. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. Questions: - Q: How do I make SRT subtitles from a video on a Mac? A: In Final Cut Pro, choose Edit → Generate Captions → Closed Captions, convert the captions to SRT, then File → Export Closed Captions. Without Final Cut Pro, Whisper in Terminal or a transcription app can write the SRT file. - Q: Can iMovie make subtitles automatically? A: No. Apple’s iMovie guide has no captioning or subtitle feature. - Q: Does Final Cut Pro generate captions in other languages? A: Apple says Generate Captions is available in U.S. English only, on a Mac with Apple silicon. - Q: What is the difference between SRT and VTT? A: Both are text files of lines with start and end times. SRT and VTT are the two common formats, and VTT is the one web video players read. Final Cut Pro’s generated captions start as a third format, iTT. - Q: Can I make subtitles on a Mac for free? A: Yes, with the open Whisper model in Terminal, which writes SRT and VTT files after a short install. - Q: Is there an app that makes subtitles from a video on a Mac? A: Flit is a local dictation app for Mac with a Transcribe tab: drop a video on it and export SRT or VTT subtitles, on the Mac. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to transcribe a Zoom or Teams recording on a Mac https://flit.fyi/transcribe-zoom-teams-recording-mac (Transcribe recordings, published 2026-10-16) Short answer: Zoom makes a transcript only of cloud recordings, on a paid plan, when the host has turned on Create audio transcript; it arrives as a VTT file beside the recording. A Zoom recording saved to your Mac gets no transcript from Zoom. In Microsoft Teams, recording a meeting starts transcription, and the organizer can download the transcript as a .docx or .vtt file after the meeting. For a recording without a transcript, Flit, a local dictation app for Mac, transcribes the saved MP4 or M4A file on the Mac itself. How to get a transcript of a Zoom cloud recording: 1. Before the meeting, sign in to the Zoom web portal and open Settings → Recording. 2. Under Advanced cloud recording settings, select Create audio transcript. It cannot be changed during a meeting. 3. In the meeting, choose Record to the Cloud. Cloud recording needs a Pro, Business or Enterprise account, and you have to be the host. 4. After the meeting, wait for Zoom’s email saying the transcript is ready. It can take longer than the recording. 5. Open Recordings & Transcripts → Cloud recordings in the web portal and click the meeting. 6. The transcript is a separate VTT file in the list. Download it, or open it in any text editor. Do you get a transcript?: - Zoom, saved to the cloud | Yes, as a VTT file, if Create audio transcript was on and the plan is paid | Transcribe the downloaded file - Zoom, saved to your Mac | No. Zoom says computer recordings do not support audio transcription | Transcribe the MP4 or M4A in Documents/Zoom - Zoom, no recording at all | A meeting transcript saved as a .txt file, if the host turned it on | Nothing to transcribe afterwards - Teams, recorded | Yes. Recording starts transcription, and the organizer downloads .docx or .vtt | Generate one from the recording in OneDrive or SharePoint - Teams, transcribed but not recorded | Yes, from the meeting’s Recap | Download it from Recap Transcribe the recording on your Mac: Flit is a dictation app with a Transcribe tab. Hand it the MP4 or M4A Zoom saved, or the MP4 the organizer shared, and the transcript comes back on the Mac, with nothing uploaded. - Drop the file from Documents/Zoom on the Transcribe tab, or on the Flit icon in the menu bar. - A two-hour recording took about three and a half minutes at the Clean formatting level in our tests, and you can keep working while it runs. - The transcript is split into paragraphs where the speaker paused, and saved as a text file in a folder you choose. - Export it as Markdown, or as SRT and VTT with timestamps. - It works with any plan, any administrator settings, and no connection. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does not say who said what. Zoom and Teams transcripts can name speakers; Flit’s is one flow of text. - Flit does not join or record the meeting. You need the recording file first. - Recording other people may need their consent where you are. That is for you to check. - Which languages a recording can be in depends on the voice model you choose: Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99 and Omnilingual more than 1,600, with accuracy that varies by language. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. Questions: - Q: Does Zoom transcribe local recordings? A: No. Zoom says computer recordings do not support audio transcription, and points to cloud recording if you need it. - Q: How do I get a transcript of a Zoom recording? A: Turn on Create audio transcript in the web portal under Settings → Recording before the meeting, record to the cloud, then download the VTT file from Recordings & Transcripts. It needs a paid plan. - Q: Where are Zoom recordings saved on a Mac? A: By default in /Users/[your name]/Documents/Zoom, as an MP4 video and an M4A audio file. You can change the folder in Zoom’s Settings → Recording. - Q: How do I download a Teams meeting transcript? A: Open the meeting chat, then Recap → Transcript → Download, and choose .docx or .vtt. By default only the organizer and co-organizers can download it. - Q: Can I transcribe a Teams recording after the meeting? A: The video’s owner can generate a transcript from the recording in OneDrive or SharePoint, unless the organization has turned that off. - Q: Is there a way to transcribe a meeting recording without uploading it? A: Yes. Flit is a local dictation app for Mac with a Transcribe tab: drop the MP4 or M4A on it, and the transcript comes back on the Mac. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ### How to record and transcribe a phone call on a Mac https://flit.fyi/transcribe-phone-call-mac (Transcribe recordings, published 2026-10-16) Short answer: On a Mac with macOS 26 or later, you can record a call from the Phone app or a FaceTime audio call: open the call’s options and click Call Recording. Both people hear a notice, and the recording and its transcript are saved to a Call Recordings folder in Notes, with each speaker identified. Transcripts are only made in some languages and regions, and call recording is not available in the European Union and several other places. Flit, a local dictation app for Mac, transcribes a saved call recording on the Mac in languages Apple’s transcripts do not cover, such as German and French. How to record and transcribe a call on a Mac: 1. Make the call from the Phone app on your Mac, or start a FaceTime audio call with one other person. 2. During the call, open the call’s options and click Call Recording. 3. Both of you hear an audio notice that the call is being recorded. It plays again from time to time during the call. 4. End the call. The recording is saved to the Call Recordings folder in Notes. 5. Open the note and double-click the recording to see the transcript. 6. To use the text elsewhere, open the transcript’s options and choose Copy Transcript or Add Transcript to Note. Call recording on Apple devices: - iPhone, iOS 18 or later | During the call, open the options and tap Call Recording | Notes, Call Recordings folder - Phone app on a Mac, macOS 26 or later | During the call, open the options and click Call Recording | Notes, Call Recordings folder - FaceTime audio on a Mac, macOS 26 or later | The same, with one other person on the call | Notes, with the transcript - The European Union and several other places | Not available | Nothing is recorded When the call is in another language: Apple transcribes calls in seven languages. Flit transcribes a saved call recording on the Mac in any language its voice model covers, German, French and Italian among them. - Save the recording from Notes with Save Attachment, then drop the file on Flit’s Transcribe tab. - A transcript in paragraphs, saved as a text file, and exported as Markdown, SRT or VTT. - Nothing is uploaded, and it works with Wi-Fi off. - Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. Good to know: - Flit does not record calls. Apple’s call recording, where it is available, makes the file. - Apple’s transcript labels each speaker. Flit’s does not. - Which languages a recording can be in depends on the voice model you choose: Parakeet Ultra covers 25 European languages, Cohere Transcribe 14, Qwen3-ASR 30, Whisper 99 and Omnilingual more than 1,600, with accuracy that varies by language. - Flit runs on a Mac with Apple silicon and macOS 14 or later, and its voice model is a download of about 0.5 to 2.8 GB that you choose once. Questions: - Q: Can I record a phone call on a Mac? A: Yes, on macOS 26 or later, in the Phone app or on a FaceTime audio call with one other person: open the call’s options and click Call Recording. It is not available in the European Union and some other regions. - Q: Where do call recordings go on a Mac? A: To a Call Recordings folder in the Notes app, with the transcript, in languages Apple transcribes. - Q: Does the other person know the call is being recorded? A: Yes. Apple says both participants hear an audio notice that the call is being recorded, and that it replays during the call. - Q: Which languages get a call transcript? A: Apple lists Cantonese, English, Japanese, Korean, Mandarin Chinese, Brazilian Portuguese and Spanish (Mexico and the United States). - Q: Why is call recording missing on my iPhone or Mac? A: Call recording is not available in the European Union and several other countries and regions that Apple lists. Where it is available, it is on by default. - Q: How do I transcribe a call recording in German or French? A: Save the recording from Notes with Save Attachment and transcribe the file with a tool that covers the language. Flit is a local dictation app for Mac that transcribes recordings on the Mac in German, French and many more. Flit is free for 7 days with every feature on, then $19 once, with no account and no subscription. ## Glossary ### Globe key (Fn key, Function key) URL: https://flit.fyi/glossary/globe-key The Globe key is the key marked with a globe symbol, and usually also with fn, on Mac keyboards. Apple calls it the Fn (Function) or Globe key. It works as a modifier for shortcuts and function keys, and pressing it on its own can change the input source, open the emoji picker, or start Dictation. Most people meet the Globe key by accident. They brush it while typing and an emoji picker appears, or the keyboard language changes, or a microphone shows up next to the cursor. None of that is a fault. It is one setting in System Settings, and you can change it in a few seconds. The key does two jobs. Held down with another key, it is a modifier, the same way Command and Option are. Pressed alone, it does whatever the "Press 🌐 key to" setting says. On some Mac models that setting is labeled "Press fn key to" instead, and it works the same way. #### What does the Globe key do on a Mac? Pressed on its own, the Globe key does one of four things. Apple lists them in Keyboard settings, and you choose one. The label on the setting depends on your Mac model: it reads either "Press 🌐 key to" or "Press fn key to". Held down with another key, it acts as a modifier. Apple’s list of Mac keyboard shortcuts names it "Fn (Function) or Globe" and puts it first among the modifier keys. - Change Input Source: switches between keyboard layouts for writing in other languages. - Show Emoji & Symbols: opens the Character Viewer for entering emoji, symbols, and more. - Start Dictation: starts dictation when you press the key twice. You may be asked to enable Dictation first. - Do Nothing: nothing happens when you press the key alone. #### How to change what the Globe key does The setting is at the top level of Keyboard settings, not inside the Dictation section. 1. Choose Apple menu → System Settings. 2. Click Keyboard in the sidebar. You may need to scroll down. 3. Find the pop-up menu labeled "Press 🌐 key to" or "Press fn key to". 4. Choose Change Input Source, Show Emoji & Symbols, Start Dictation, or Do Nothing. 5. Press the key once, or twice if you chose Start Dictation, to confirm it does what you expect. #### How to start Dictation with the Globe key With the Start Dictation option, you press the Globe key twice to begin dictating. Apple notes that the option and the Dictation shortcut are linked. If you choose Press Fn (Function) Key Twice as the Dictation shortcut, the Keyboard setting changes to Start Dictation (Press Fn Twice) automatically. Apple also lists Fn-D as a shortcut to start or stop dictation, and says you may be able to set a different one in Keyboard settings. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and turn it on. Click Enable if you are asked. 3. Set "Press 🌐 key to" or "Press fn key to" to Start Dictation. 4. Click in a text field and press the Globe key twice. 5. Speak when the cursor is highlighted and pulsing, or when you hear the tone. 6. Press Escape or the Dictation shortcut again when you are done. #### Why does the Globe key open the emoji picker? If the Character Viewer opens every time you touch the key, the setting is on Show Emoji & Symbols. Apple describes this as an option so that you only need to press the Fn key or the Globe key to get emoji suggestions or open the Character Viewer. To stop it, change the setting to Do Nothing or to one of the other options. You do not lose the emoji picker. Fn-E and Control-Command-Space still open it, and so does Edit → Emoji & Symbols in most apps. #### Globe key shortcuts on a Mac Apple’s keyboard shortcut list includes a set of shortcuts that use the Globe key as a modifier. Hold the key, press the letter, then release both. On a keyboard with no globe symbol, the Fn key does the same job. - Fn-E: show the Character Viewer, where you choose emoji and other symbols. - Fn-D: start or stop dictation. - Fn-C: show or hide Control Center. Fn-N: show or hide Notification Center. - Fn-A: show or hide the Dock. Fn-H: show the desktop. - Fn-Q: create a Quick Note. - Fn-Delete: forward delete. Fn with the arrow keys: Page Up, Page Down, Home, and End. #### Globe key not working When the key seems dead, the cause is often a setting and not the hardware. Work through these checks in order. - Open Keyboard settings and check that the "Press 🌐 key to" or "Press fn key to" option is not set to Do Nothing. - If it is set to Start Dictation, remember that the key must be pressed twice, and that Dictation must be turned on. - If Voice Control is on, standard Dictation is not available, so the key will not start it. - If the top row does not respond as F1, F2 and so on, hold the Globe key while you press the function key. - On a keyboard not made by Apple there may be no Fn key. Apple suggests trying the Control key with a function key, and checking with the keyboard manufacturer. Questions: - Q: Is the Globe key the same as the Fn key? A: Yes. Apple refers to it as the Fn (Function) or Globe key, and its shortcut lists write Globe key shortcuts as Fn shortcuts, such as Fn-E. The setting in Keyboard settings is labeled "Press fn key to" or "Press 🌐 key to" depending on your Mac model. - Q: How do I stop the Globe key from opening emoji? A: Choose Apple menu → System Settings, click Keyboard, then set "Press 🌐 key to" (or "Press fn key to") to Do Nothing or another option. You can still open the emoji picker with Fn-E or Control-Command-Space. - Q: How do I turn off the Globe key on a Mac? A: Set "Press 🌐 key to" or "Press fn key to" to Do Nothing in System Settings → Keyboard. The key then does nothing when pressed alone, but it still works as a modifier for function keys and Fn shortcuts. - Q: Why does my keyboard language change when I press the Globe key? A: The key is set to Change Input Source, which switches between the keyboard layouts you have added. Change the option in System Settings → Keyboard if you do not want that. - Q: How do I use F1 to F12 with the Globe key? A: Press and hold the Fn or Globe key while you press a key in the top row. To make the top row act as standard function keys all the time, go to System Settings → Keyboard → Keyboard Shortcuts → Function Keys and select "Use F1, F2, etc. keys as standard function keys". - Q: What does pressing the Globe key twice do? A: When the key is set to Start Dictation, pressing it twice starts Dictation in the current text field. Dictation must be turned on in System Settings → Keyboard → Dictation. In Flit: Holding the Globe key is how you dictate in Flit by default, which is why Flit asks you to set the key to Do Nothing. Fn shortcuts such as Fn-E still work while Flit is running. (https://flit.fyi/globe-key-opens-emoji-picker-mac) Related terms: dictation, push-to-talk, voice-typing ### Voice Control (Mac voice commands) URL: https://flit.fyi/glossary/voice-control Voice Control is an accessibility feature built into macOS that lets you operate a Mac entirely by voice. You speak commands to navigate the desktop and apps, interact with what is on the screen, and dictate and edit text, instead of using a keyboard, mouse, or trackpad. Voice Control is often confused with two other Apple features. Dictation turns speech into text and does nothing else. VoiceOver is the screen reader that speaks what is on the screen. Voice Control is the one that listens all the time and treats what you say as instructions: "Open Mail", "Scroll down", "Click Done". It lives in Accessibility settings, and Apple groups it with its features for mobility. It also takes over text entry while it is on. In Apple’s words, when Voice Control is on, you use Voice Control to dictate text, and standard macOS Dictation is not available. #### How to turn on Voice Control on a Mac The first time you turn on Voice Control, your Mac must be connected to the internet so it can complete a one-time file download from Apple. After the download is complete, you do not need an internet connection to use it. You can also ask Siri to "Turn on Voice Control". 1. Choose Apple menu → System Settings. 2. Click Accessibility in the sidebar, then click Voice Control. You may need to scroll down. 3. Turn Voice Control on. 4. Wait for the Voice Control icon in the menu bar to change, which shows that the download is complete and Voice Control is listening. 5. Say a command, for example "Open Mail" or "Scroll down". 6. Click Open Tutorial in Voice Control settings if you want to practice the essential commands. #### How to turn off Voice Control There are two levels of off. Pausing keeps the feature on but stops it from acting on what you say. Turning it off in settings ends it and makes standard Dictation available again. Pause when you take a call or talk to someone in the room. Turn it off if you only wanted to try it, or if you want the regular Dictation shortcut back. 1. To pause, say "Stop listening". Say "Start listening" to resume. 2. To turn it off, choose Apple menu → System Settings. 3. Click Accessibility in the sidebar, then click Voice Control. 4. Turn Voice Control off. #### Voice Control vs Dictation Dictation is for entering text. You start it with a key or shortcut, speak, and it stops when you end it or when no speech is detected for 30 seconds. It is set up under System Settings → Keyboard → Dictation. Voice Control is for running the whole Mac by voice, and it keeps listening until you pause it or turn it off. The two do not run side by side. While Voice Control is on, it handles dictation itself, in one of several modes that you switch between by saying the mode name. - Dictation mode: any words that are not Voice Control commands are entered as text. This mode is on by default. - Spelling mode: you dictate character by character, which helps with passwords and web addresses. It is not available in all Voice Control languages. - Command mode: Voice Control responds only to commands, and other words are ignored. - Swift mode: Voice Control writes Swift code and handles the formatting. #### Voice Control vs VoiceOver The names are close and the jobs are opposite. Voice Control takes your voice as input so that you can control the Mac without your hands. VoiceOver is the screen reader built into macOS, which speaks what is on the screen so that you can use the Mac without seeing it. VoiceOver is turned on and off with Command-F5, or under System Settings → Accessibility → VoiceOver. The two features work together: with both on, you can say commands such as "VoiceOver read all". Apple suggests headphones in that case, so that Voice Control hears only you. #### Is Voice Control private? Apple’s instructions say that after the one-time download, you do not need an internet connection to use Voice Control. That is the clearest statement Apple makes about where it runs. A separate choice covers whether Apple may keep samples of your audio. Under System Settings → Privacy & Security → Analytics & Improvements, the Improve Siri & Dictation switch controls whether Apple can store and review a sample of your Siri and Dictation audio. Apple says that data helps improve Siri, Dictation, and other language features such as Voice Control. Leave it off if you do not want to share recordings. #### How to add custom words to Voice Control If Voice Control keeps getting a name or a technical term wrong, add it to the custom vocabulary. Apple says you can add up to 1000 terms, in languages that support a custom vocabulary. If the Vocabulary option is dimmed, the selected language does not support it. 1. Choose Apple menu → System Settings, click Accessibility, then click Voice Control. 2. Click Vocabulary. 3. Click the add button, type the word or phrase as you want it entered, then click Next. 4. Record yourself saying the phrase and click Save, or click Save Without Recording. Recording requires Dictation to be turned on. 5. Repeat for other terms, then click Done. 6. While you work, you can also select a word and say "Add to vocabulary". Questions: - Q: Why did Dictation stop working after I turned on Voice Control? A: Apple states that when Voice Control is on, you use Voice Control to dictate text and standard macOS Dictation is not available. Turn Voice Control off under System Settings → Accessibility → Voice Control to get standard Dictation back. - Q: Does Voice Control on Mac need the internet? A: Only the first time. Your Mac must be online to complete a one-time file download from Apple, and after that Voice Control does not need an internet connection. - Q: How do I see all Voice Control commands? A: Say "Show commands" to see the commands available for what you are doing. For the full list with examples, go to System Settings → Accessibility → Voice Control and click Commands. - Q: How do I click something with Voice Control when I do not know its name? A: Say "Show names" or "Show numbers" to label the items on the screen, then say the name or number. Say "Show grid" to place a numbered grid over the screen for precise locations. - Q: Can I create my own Voice Control commands? A: Yes. In System Settings → Accessibility → Voice Control, click Commands, then add a command, give it a phrase of two or more words, and choose an action such as Paste Text, Press Keyboard Shortcut, or Run Shortcut. - Q: Is Voice Control the same as Siri? A: No. Voice Control is an accessibility feature for operating the Mac and entering text by spoken command. Siri is a separate assistant, although you can ask Siri to turn Voice Control on. In Flit: Flit types what you say into any app. It does not control your Mac by voice, so it stands in for Dictation, not for Voice Control. (https://flit.fyi/features) Related terms: dictation, spoken-punctuation, personal-dictionary, on-device-processing ### Dictation commands (Mac dictation commands, voice punctuation commands) URL: https://flit.fyi/glossary/dictation-commands Dictation commands are the words and phrases you say while using Dictation on a Mac to enter punctuation, symbols, capital letters, and formatting instead of ordinary text. Saying "comma" types a comma, and saying "new paragraph" starts a new paragraph. Apple publishes the full list in the Mac User Guide under "Commands for dictating text on Mac". It says the commands are always available while you are dictating. There is nothing to switch on beyond Dictation itself, and no special mode to enter. The list is longer than most people expect. It covers punctuation, typography symbols, formatting, capitalization, math and currency signs, a few emoticons, and the copyright, registered, and trademark signs. In some languages you can also insert emoji by name. #### How to use dictation commands on a Mac You say a command in the flow of your sentence, with no pause or trigger word. Dictation replaces the command with the mark or the formatting. On a Mac with Apple silicon, you can also keep typing while you dictate, which is a quick way to fix a mark by hand. 1. Choose Apple menu → System Settings, click Keyboard in the sidebar, go to Dictation, and turn it on. 2. Click where you want the text to go. 3. Press the Microphone key if your keyboard has one, use the Dictation keyboard shortcut, or choose Edit → Start Dictation. 4. Speak, and name the marks as you go, for example "dear Sam comma new paragraph thanks for the notes period". 5. Press Escape, the Microphone key, or the Dictation shortcut to stop. Dictation also stops when no speech is detected for 30 seconds. #### Punctuation and symbol commands These are the commands most people need. Each one is the plain name of the mark. Apple lists four ways to say the sentence-ending dot: period, point, dot, or full stop. - Sentence marks: comma, period, question mark, exclamation mark, colon, semicolon, ellipsis. - Quotes and joins: quote and end quote, begin single quote and end single quote, apostrophe, hyphen, dash. - Pairs: open and close parenthesis, open and close square bracket, open and close brace, open and close angle bracket. - Typography: ampersand, asterisk, at sign, backslash, forward slash, caret, underscore, vertical bar, percent sign, degree sign, and hashtag or pound sign. - Math and currency: equal sign, plus sign, minus sign, multiplication sign, greater than sign, less than sign, dollar sign, cent sign, euro sign, yen sign, pound sterling sign. - Other: copyright sign, registered sign, trademark sign, and emoticons such as smiley face, frowny face, and winky face. #### How to dictate a new line, numbers, and capital letters Formatting commands change how the next word or phrase is written. Apple notes that "new line" is the same as pressing Return once and "new paragraph" is the same as pressing it twice, and that the break appears when you are done dictating. The capitalization commands come in two kinds. Some apply to the next word only. The ones ending in "on" stay active until you say the matching "off" command. - "New line" starts a new line. "New paragraph" starts a new paragraph. - "Tab key" moves the cursor to the next tab stop. - "Numeral" formats the next phrase as a number. "Roman numeral" formats it as a Roman numeral. - "No space on" formats the next phrase without spaces. "No space off" resumes default spacing. - "Caps on" formats the next phrase in title case. "Caps off" resumes the default case. - "All caps" formats the next word in capitals. "All caps on" and "all caps off" do the same for a phrase. #### How to dictate emoji on a Mac Apple says that in some languages you can insert emoji using dictation. You say the name of the emoji followed by the word "emoji". Its macOS Feature Availability page lists the supported languages under "Dictation: Dictate Emoji". 1. Start Dictation in any text field. 2. Say the name of the emoji and then the word "emoji", for example "heart emoji" or "car emoji". 3. Carry on dictating the rest of your sentence. 4. If nothing appears, check that your dictation language supports emoji, or press Fn-E to open the Character Viewer and pick one by hand. #### Dictation commands vs Voice Control commands Dictation commands only insert characters and formatting. They cannot select text, delete a word, or click a button. For that, Apple points to Voice Control, the accessibility feature for controlling a Mac by voice, which has its own text editing commands such as "Replace cat with dog". You cannot mix the two. When Voice Control is on, you use Voice Control to dictate text, and standard macOS Dictation is not available. The list of what you can say in Voice Control is under System Settings → Accessibility → Voice Control → Commands. #### Dictation commands not working When Dictation types the word "comma" instead of the mark, or ignores a command, check the following. - Use Apple’s exact wording. The list says "open parenthesis", "end quote", and "all caps", and a different phrasing may be typed as plain words. - Check the dictation language. Choose the correct language and region under System Settings → Keyboard → Dictation → Languages. - Emoji by name work only in some languages, and automatic punctuation only in supported languages. - Wait until you stop dictating before you judge a line break. A new line or new paragraph appears when you are done. - Click a word underlined in blue to choose an alternative. Dictation marks text it is unsure about that way. - Reduce background noise and speak at a normal volume, or use a headset microphone. Questions: - Q: How do I add punctuation when dictating on a Mac? A: Say the name of the mark, such as "comma", "period", or "question mark". In supported languages, Dictation also inserts commas, periods, and question marks automatically. - Q: How do I start a new paragraph with Mac Dictation? A: Say "new paragraph". Say "new line" for a single line break. The break appears when you are done dictating. - Q: How do I dictate in all caps on a Mac? A: Say "all caps" before a single word. For a longer phrase, say "all caps on", speak the phrase, then say "all caps off". - Q: How do I dictate a number as digits on a Mac? A: Say "numeral" before the number. Apple’s command list describes it as formatting the next phrase as a number. - Q: Where is the list of Mac dictation commands? A: Apple publishes it in the Mac User Guide on the page "Commands for dictating text on Mac". It is grouped into punctuation, typography, formatting, capitalization, mathematical, currency, emoticons, and intellectual property. - Q: Can I delete or select text with Mac Dictation commands? A: Not with standard Dictation, whose published commands insert characters and formatting. Voice Control, under System Settings → Accessibility, has commands for editing text and controlling the Mac. In Flit: Flit has more than 60 spoken commands of its own for punctuation, symbols, casing and structure. They are in English today. (https://flit.fyi/commands) Related terms: spoken-punctuation, dictation, formatting-pass, verbatim-mode ### Auto-punctuation (automatic punctuation) URL: https://flit.fyi/glossary/auto-punctuation Auto-punctuation is a setting of Dictation on Mac that automatically inserts commas, periods, and question marks as you dictate. It works in supported languages and can be turned on or off in System Settings under Keyboard, in the Dictation section. Speech has no punctuation, so dictated text needs it from somewhere. Before this setting existed in your language, you had to say every mark out loud. With Auto-punctuation on, Dictation works out where a sentence ends or pauses and adds the mark for you. It covers three marks and no more. Anything else, such as a colon, a quotation mark, or a new paragraph, still has to be spoken as a command. And when the automatic marks land in the wrong place for the way you talk, the fix is one switch. #### How to turn off Auto-punctuation on a Mac Turn it off if Dictation keeps ending sentences where you only paused to think, or if you prefer to say every mark yourself. Turning it off does not remove punctuation from Dictation. It only stops the automatic marks. Every spoken command, from "comma" to "question mark", keeps working, so you decide where each one goes. 1. Choose Apple menu → System Settings. 2. Click Keyboard in the sidebar. You may need to scroll down. 3. Go to the Dictation section. 4. Turn off Auto-punctuation. 5. Dictate a test sentence and say the marks by name, for example "see you at noon period". #### How to turn on Auto-punctuation Auto-punctuation is part of Dictation, so Dictation has to be on first. The switch sits with the other Dictation options: Languages, Microphone source, and Shortcut. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and turn it on. Click Enable if you are asked. 3. Check that the language you speak is selected next to Languages. 4. Turn on Auto-punctuation. 5. Click in a text field, start Dictation, and speak a full sentence without naming any marks. #### Why does Dictation add punctuation I did not say? Because Auto-punctuation is on. Apple’s Dictation guide says that in supported languages, Dictation automatically inserts commas, periods, and question marks for you as you dictate. That is the full list of marks Apple names. If you also say a mark out loud while the setting is on, check the result. Many people who learned to say "period" at the end of every sentence can stop doing so. If you do not want the automatic marks, turn the setting off and keep speaking them. Auto-punctuation is not a separate service. It is part of Dictation, so the same privacy note applies: the text under the Dictation switch in Keyboard settings tells you whether your voice inputs and transcripts for general text dictation are processed on your device and not sent to Apple. #### Auto-punctuation vs spoken punctuation These are two ways to get the same characters. Auto-punctuation infers a mark from the shape of the sentence. Spoken punctuation types exactly the mark you name. Apple says its dictation commands are always available while you are dictating, so you can use both together. A practical rule: leave Auto-punctuation on for messages, notes, and email, where a missed comma costs nothing. Name the mark whenever the exact character matters. - Added automatically: commas, periods, and question marks. - Spoken by name: exclamation mark, colon, semicolon, quote and end quote, parentheses, brackets, dash, and hyphen. - Spoken by name: new line and new paragraph. - Spoken by name: symbols such as at sign, ampersand, percent sign, and dollar sign. #### Which languages support Auto-punctuation? Not all of them. Apple notes in Keyboard settings help that Auto-punctuation is not available in all languages, and it publishes the list on its macOS Feature Availability page under "Dictation: Auto-Punctuation". For macOS 27 that list covers regional variants of the languages below. The list changes between macOS versions, so check the Feature Availability page for the exact language and region you dictate in. - English, including the United States, United Kingdom, Canada, Australia, and India. - French, German, Italian, Spanish, and Dutch. - Chinese (Simplified and Traditional), Mandarin Chinese (Taiwan), and Cantonese. - Japanese, Korean, and Thai. #### Auto-punctuation not working If your dictated text comes out as one long unpunctuated line, check these things. Start with the setting itself, since it can be switched off without you noticing, then move on to the language. - Open System Settings → Keyboard → Dictation and confirm that Auto-punctuation is turned on. - Confirm that the language you are dictating in is on Apple’s Auto-punctuation list. If you dictate in more than one language, check which one is selected next to the cursor. - Check whether Voice Control is on. When it is, standard Dictation is not available and you are dictating through Voice Control instead. - Speak in full sentences at a normal pace, and reduce background noise. Apple suggests a headset microphone in a noisy or echoing room. - As a fallback, say the marks by name. Spoken commands work whether or not Auto-punctuation is on. Questions: - Q: How do I turn off auto-punctuation on Mac? A: Choose Apple menu → System Settings, click Keyboard, go to Dictation, then turn off Auto-punctuation. After that, Dictation types only the marks you say by name. - Q: Where is the Auto-punctuation setting on a Mac? A: It is in System Settings → Keyboard, in the Dictation section, below the Languages, Microphone source, and Shortcut options. - Q: What punctuation does Mac Dictation add automatically? A: Commas, periods, and question marks, in supported languages. Every other mark, such as an exclamation mark or a colon, has to be spoken by name. - Q: Does Mac Dictation add exclamation marks automatically? A: Apple lists only commas, periods, and question marks for Auto-punctuation. To get an exclamation mark, say "exclamation mark". - Q: Can I still say punctuation when Auto-punctuation is on? A: Yes. Apple says the dictation commands are always available while you are dictating, so you can say "comma", "colon", or "new paragraph" at any time. - Q: Why is there no punctuation when I dictate on my Mac? A: Either Auto-punctuation is turned off in System Settings → Keyboard → Dictation, or the language you are dictating in does not support it. Apple’s macOS Feature Availability page lists the supported languages. In Flit: Flit punctuates on your Mac as part of formatting, and you can still say a mark aloud when you want it exactly there. (https://flit.fyi/commands) Related terms: spoken-punctuation, formatting-pass, dictation, verbatim-mode ### Input volume (microphone volume, input level, mic gain) URL: https://flit.fyi/glossary/microphone-input-volume Input volume is the setting on a Mac that controls how strongly sound from the selected microphone or other input device is captured. You set it with a slider in System Settings → Sound, under Input, where you also choose which input device the Mac uses. Input volume is the first thing to check when people say you sound quiet on a call, when a recording is faint, or when dictation keeps missing words. It is separate from the output volume keys on your keyboard, which only change what you hear. People also call this the input level. Strictly, the volume is the setting you choose and the level is how strong the signal turns out to be once your voice, your distance from the microphone and that setting are combined. #### How to change the microphone input volume on a Mac Apple puts the control in Sound settings. Apple’s guidance is simple: if the source is loud, lower the input volume so the sound is not too loud or distorted, and if someone is speaking softly, increase it so your Mac can better capture the voice. 1. Choose Apple menu → System Settings, then click Sound in the sidebar. You may need to scroll down. 2. Click Input. 3. Select the device you want to use in the list of sound input devices. 4. Drag the input volume slider to the right to raise the volume, or to the left to lower it. 5. Speak at your normal distance and volume in the app you use, and adjust again if you are still too quiet or too loud. #### How to choose which microphone your Mac uses A Mac can listen through more than one device. Apple lists the computer’s internal microphone, your display’s microphone if it has one, your iPhone microphone, or an external microphone plugged into the sound port. All available input devices appear in one list, and a display with an internal microphone is listed as the Mac model followed by Microphone. The wrong device is a more common problem than the wrong volume. If a headset or a display is selected and you are talking to the laptop, you will sound distant however high the slider is. 1. Choose Apple menu → System Settings, then click Sound. 2. Click Input. 3. Read the list of sound input devices and click the one you are actually speaking into. 4. Set the input volume for that device. 5. If you use Apple’s Dictation, also open System Settings → Keyboard, go to Dictation, and check the Microphone source pop-up menu. Automatic lets your Mac listen to the device you are most likely to use for Dictation. #### Where is the input level on a Mac? The control Apple documents is the input volume slider in System Settings → Sound → Input. Apple’s FaceTime guide gives the same path with one more label on current macOS: go to Output & Input, click Input, then drag the input volume slider. To judge the result, use the app you are recording or speaking in. Apple’s troubleshooting article for recording describes audio meters that move when sound is reaching an app, and says that if you do not see meters move or waveforms appear, the app may not have permission to use the input. #### Why is my Mac microphone so quiet? A low level usually has an ordinary cause. Check these in order, because the first three account for most cases. - The wrong input device is selected. Check the list under System Settings → Sound → Input. - The input volume slider is set low for that device. Drag it up and test again. - You are too far from the microphone. Move closer, or use a headset so your voice is clearly louder than the room. - The app is not allowed to use the microphone. Open System Settings → Privacy & Security → Microphone and turn access on for that app. - The device controls its own level. Apple’s help for macOS Sequoia 15 notes that you cannot adjust the input volume for a digital input device with your computer’s controls and must adjust it at the source. - An external microphone or audio interface needs attention. Apple suggests checking with the manufacturer for software updates and compatibility information. #### Input volume vs Mic Mode These two settings solve different problems. Input volume changes how loud everything is, your voice and the room together. Raising it makes a quiet voice louder, and it makes the fan behind you louder too. Mic Mode changes what is kept. Voice Isolation filters out background sounds and Wide Spectrum includes them, in apps that support Mic Modes. If people can hear you but also hear your surroundings, change the mic mode. If they can barely hear you at all, change the input volume or the device. #### How to tell when the microphone is in use macOS shows you when an app is listening. Apple says the Recording Indicator in Control Center shows when your computer’s microphone is in use or when it was used recently. It also turns on when you open the Input options in Sound settings, so seeing it there is expected. You decide which apps can use the microphone at all. Open System Settings → Privacy & Security, click Microphone, and turn access on or off for each app in the list. An app with access turned off receives no sound, which can look exactly like a microphone with the volume at zero. Questions: - Q: How do I turn up the microphone volume on a Mac? A: Choose Apple menu → System Settings, click Sound, click Input, select your microphone, and drag the input volume slider up. - Q: Where is the input volume setting on a Mac? A: It is in System Settings → Sound, under Input. Select an input device in the list and the input volume slider applies to that device. - Q: How do I change which microphone my Mac uses? A: Open System Settings → Sound, click Input, and select the device you want in the list of sound input devices. All sound input devices available to your Mac are listed there. - Q: Why is my microphone so quiet on my Mac? A: The most common causes are the wrong input device being selected, a low input volume slider, and sitting too far from the microphone. Check all three in System Settings → Sound → Input. - Q: Why can I not change the input volume on my Mac? A: Some devices set their own level. Apple’s help for macOS Sequoia 15 says you cannot adjust the input volume for a digital input device with your computer’s controls and must adjust it from its source. - Q: Why does the microphone indicator appear when I open Sound settings? A: Apple says the Recording Indicator turns on when you access the Input options in Sound settings. It shows that the microphone is in use or was used recently. In Flit: Flit has its own microphone setting, in Settings → General, so you can choose the microphone it listens to by name. (https://flit.fyi/setup#permissions) Related terms: voice-activity-detection, dictation, word-error-rate, speech-to-text ### Mic Mode (Mic Modes, Voice Isolation, Wide Spectrum) URL: https://flit.fyi/glossary/mic-mode Mic Mode is a macOS control that changes how your Mac’s microphone treats the sound around you in apps that support it. Voice Isolation filters out background sounds, Wide Spectrum includes the sounds around you, and Standard is the standard mic mode with neither applied. Most people find Mic Mode during a call, when someone says they can hear a dog, a fan or traffic. It is a quick fix because it lives in the menu bar and changes what the other side hears without any new hardware. It is also easy to miss. Apple’s instructions point to the menu bar, not to a pane in System Settings, and they start with going to an app that supports Mic Modes. #### Where is Mic Mode on a Mac? The location depends on your version of macOS. Apple says that in macOS Sonoma 14 or later you click Mic Mode from the Video menu or Audio menu that appears in the menu bar when a video call is in progress. In macOS Ventura 13 or macOS Monterey 12, you click Control Center in the menu bar, then click Mic Mode. In both cases there is a condition. You have to be in an app that supports Mic Modes, such as FaceTime, and Apple notes that the options vary by app. #### How to turn on Voice Isolation on a Mac Voice Isolation is the mode most people want. Apple says it gives priority to your voice and minimizes the sounds of traffic, dogs barking and other background noise around you. These steps are for macOS Sonoma 14 or later. 1. Open an app that supports Mic Modes, such as FaceTime, and start a call. 2. Look in the menu bar for the Video menu, or the Audio menu if it is an audio call. 3. Click it, then click Mic Mode. During a FaceTime video call, click the arrow next to Mic Mode. 4. Select Voice Isolation. 5. Ask the other person whether the background noise has gone, and switch back if your voice sounds worse. #### How to switch back to Standard or use Wide Spectrum Wide Spectrum does the opposite of Voice Isolation. Apple says it captures your voice as well as the sound around you, such as the sound of music or nature. Use it when the room is the point: a music lesson, a group around one laptop, or a sound you want the other side to hear. 1. While the call is in progress, click the Video menu or Audio menu in the menu bar. 2. Click Mic Mode. 3. To include the sounds around you, select Wide Spectrum. 4. To revert to the standard mic mode, select Standard. #### Standard vs Voice Isolation vs Wide Spectrum The three modes are three answers to one question: how much of the room should come through with your voice. None of them is a volume control. They change what the microphone keeps and what it removes. - Standard: the standard mic mode. Choose it to go back to normal after trying the others. - Voice Isolation: filters out background sounds and gives priority to your voice. Choose it for calls from a noisy room. - Wide Spectrum: includes the sounds around you. Choose it when the other side should hear the room, music or several people. #### Which Macs and apps support Mic Mode? Apple’s FaceTime guide says Voice Isolation and Wide Spectrum are available on Mac models 2018 and later. Apple’s article on video effects and mic modes describes the mic modes for macOS Monterey 12, macOS Ventura 13, and macOS Sonoma 14 or later. On the app side, Apple describes these as features you can use in FaceTime and other videoconferencing apps, and it tells you to go to an app that supports Mic Modes. Apple does not publish a list of those apps, and it says the options vary by app, so one app may show fewer modes than another. #### Mic Mode not showing or not working If you cannot find Mic Mode, the usual reason is that no app that supports it is in a call at that moment. Work through the basics before you assume the feature is missing. - Start a call first. Apple says the Video menu or Audio menu appears in the menu bar when a video call is in progress. - Check your macOS version. On macOS Ventura 13 or Monterey 12, Mic Mode is under Control Center, not a Video menu. - Check your Mac. Apple lists Voice Isolation and Wide Spectrum for Mac models 2018 and later. - Try FaceTime. If Mic Mode appears there but not in another app, that app may not support Mic Modes, or may offer different options. - Check the microphone itself. Open System Settings → Sound, click Input, and make sure the right device is selected and the input volume is not too low. - Check permission. Open System Settings → Privacy & Security → Microphone and make sure the app is allowed to use the microphone. Questions: - Q: Where is Mic Mode on my Mac? A: In macOS Sonoma 14 or later, click Mic Mode from the Video menu or Audio menu that appears in the menu bar when a video call is in progress. In macOS Ventura 13 or Monterey 12, click Control Center in the menu bar, then click Mic Mode. - Q: What does Voice Isolation do on a Mac? A: Voice Isolation gives priority to your voice and minimizes background noise such as traffic or a barking dog. It is one of the Mic Modes, alongside Wide Spectrum and Standard. - Q: What is Wide Spectrum on a Mac? A: Wide Spectrum is a Mic Mode that captures your voice as well as the sound around you, such as music or nature. It is the opposite of Voice Isolation. - Q: Why can I not see Mic Mode in the menu bar? A: Mic Mode appears when an app that supports Mic Modes is in a call. Start a call in an app such as FaceTime, then look for the Video menu or Audio menu in the menu bar, or open Control Center on macOS Ventura 13 or Monterey 12. - Q: Which Macs support Voice Isolation? A: Apple’s FaceTime guide says Voice Isolation and Wide Spectrum are available on Mac models 2018 and later. - Q: How do I turn off Voice Isolation on a Mac? A: During a call, click the Video menu or Audio menu in the menu bar, click Mic Mode, and select Standard. That reverts to the standard mic mode. Related terms: voice-activity-detection, dictation, word-error-rate ### Secure Keyboard Entry (Secure Event Input, secure input) URL: https://flit.fyi/glossary/secure-keyboard-entry Secure Keyboard Entry is a Terminal setting on a Mac that prevents other apps on your computer or the network from detecting and recording what you type in Terminal. It is built on a macOS mechanism called Secure Event Input, which stops keystrokes from being passed to any process that intercepts the keyboard. You meet this setting in two ways. Either you turn it on yourself because you type passwords into a terminal, or you find out about it because a keyboard shortcut, a text expander or a push-to-talk key has stopped responding and you are trying to work out why. The protection and the side effect are the same thing. macOS cannot tell a helpful tool that watches the keyboard from a harmful one, so while secure input is on, none of them receive your keys. #### How to turn on Secure Keyboard Entry in Terminal The setting is a single menu item in Apple’s Terminal app. Apple gives one warning before you use it: make sure other apps do not require keystrokes from Terminal. 1. Open the Terminal app on your Mac. 2. Click Terminal in the menu bar. 3. Choose Secure Keyboard Entry. 4. Open the Terminal menu again and confirm that a checkmark appears next to Secure Keyboard Entry. The checkmark means it is on. #### How to turn off Secure Keyboard Entry If your shortcuts or text snippets stopped working after you opened a terminal, check this menu item first. The same item switches the setting on and off. 1. Bring Terminal to the front. 2. Click Terminal in the menu bar. 3. If Secure Keyboard Entry has a checkmark, choose it once to remove the checkmark. 4. If you use iTerm2, click iTerm2 in the menu bar and choose Secure Keyboard Entry there instead. That menu item is iTerm2's own. 5. Switch to another app and try the shortcut or snippet that had stopped working. #### What is Secure Event Input? Secure Event Input is the system mechanism underneath the menu item. Apple’s developer note on it, Technical Note TN2150, says the function was added in Mac OS X 10.3 to protect keyboard entry so that keyboard events cannot be intercepted by a keyboard intercept process. That note is from 2007 and sits in Apple’s documentation archive, but it is still Apple’s own explanation of the behavior. Apps do not need a menu item to use it. Apple’s note says standard password fields secure keyboard events only when necessary, which means while you are typing in the field. Secure Keyboard Entry in a terminal is different because you choose to leave the protection on for ordinary typing. #### Why do keyboard shortcuts and text expanders stop working? Tools that react to keys while another app is in front have to watch the keyboard. Apple’s note lists the ways a process can do that, including an event tap, and names macros, accessibility tools and keyboard customizers as useful examples. A text expander, a global hotkey and a hold-to-talk key all fall in that group. The note then states the rule: the system stops passing keyboard events to any intercept process whenever any process has enabled secure event input, whether that process is in the foreground or the background. So an app that leaves secure input on can silence keyboard tools across the whole Mac, not just inside its own window. Apple tells developers to enable it only when needed and to disable it when focus moves away. - Text expansion: the tool never sees the abbreviation you typed, so nothing expands. - Global shortcuts: a shortcut that another app listens for does nothing. - Push-to-talk keys: a tool waiting for a key press or release does not see it. - Key remappers and macro tools: they do not receive the keys they are meant to change or replay. #### Secure Keyboard Entry in iTerm2 iTerm2 is a separate terminal app, not an Apple product, and it has its own switch under iTerm2 → Secure Keyboard Entry. iTerm2's documentation says that when it is enabled the operating system prevents other programs on your computer from seeing what you type, and that it may disable global hotkeys in other programs. iTerm2's FAQ lists two more effects of its setting. When you use the open command, the app’s window may not come to the foreground, which the FAQ describes as a macOS security feature that prevents an app from stealing keyboard focus while you enter sensitive text. It also names scrolling problems with a Logitech mouse. In both cases the FAQ says to turn off iTerm2 → Secure Keyboard Entry. #### Is Secure Keyboard Entry worth leaving on? It depends on what you type in a terminal and which tools you rely on. If you often type passwords or secrets at a prompt and do not use keyboard tools, leaving it on costs you nothing. If you depend on a text expander, a launcher shortcut or a dictation key, you will lose them whenever it is on. It is also not the only protection you have. macOS asks for your permission before an app can monitor the keyboard, and you can review that list under System Settings → Privacy & Security → Input Monitoring. Removing apps you do not trust from that list protects every app, not just the terminal. Questions: - Q: What does Secure Keyboard Entry do in Terminal? A: Apple says it can prevent other apps on your computer or the network from detecting and recording what you type in Terminal. You turn it on by choosing Terminal → Secure Keyboard Entry, and a checkmark shows that it is on. - Q: How do I turn off Secure Keyboard Entry on a Mac? A: Open Terminal, click Terminal in the menu bar, and choose Secure Keyboard Entry so the checkmark disappears. In iTerm2 the item is under the iTerm2 menu. - Q: Why did my text expander stop working on my Mac? A: One common cause is secure input. While any app has Secure Event Input enabled, macOS stops passing keystrokes to tools that watch the keyboard, so the expander never sees what you typed. Check Secure Keyboard Entry in your terminal app and click out of any password field. - Q: Why do my global hotkeys not work when Terminal or iTerm2 is open? A: Secure Keyboard Entry is probably turned on. iTerm2's documentation says its setting may disable global hotkeys in other programs, and Apple warns you to make sure other apps do not require keystrokes from Terminal before you turn it on. - Q: Is Secure Event Input the same as Secure Keyboard Entry? A: Secure Event Input is the macOS mechanism that hides keystrokes from keyboard intercept processes. Secure Keyboard Entry is the menu item in Terminal, and separately in iTerm2, that keeps that protection on while you type in the terminal. - Q: Do password fields turn on secure input? A: Yes. Apple’s developer note says the standard password fields secure keyboard events only when necessary, so the protection applies while you are typing in the field. If a keyboard tool stops responding, click out of the password field and try again. Related terms: text-expansion, push-to-talk, accessibility-permission ### Live Captions (real-time captions, live transcription) URL: https://flit.fyi/glossary/live-captions Live Captions is a macOS accessibility feature that shows a real-time transcription of spoken audio in a window on your screen, for audio in any app and for live conversations around you. Audio is processed on the device, and the feature is available only on Mac computers with Apple silicon. Live Captions is one of the accessibility features for hearing on a Mac, and it is useful to anyone who wants to follow a call, a video or a podcast with the sound low. It can caption what your Mac is playing or what its microphone hears, and you switch between the two with one click. It is a reading tool, not a writing tool. Apple describes a window of captions that follows the audio. If you want your own speech typed into a document, that is the job of Dictation, which is a different feature. #### How to turn on Live Captions on a Mac Live Captions is in Accessibility settings. The first time you turn it on, language data is downloaded to your Mac, so you must be connected to the internet for that step. 1. Choose Apple menu → System Settings, then click Accessibility in the sidebar. 2. Click Live Captions. You may need to scroll down. 3. Turn Live Captions on. 4. Wait for the language data to download if this is the first time. 5. Play audio in any app, or start a call. The Live Captions window appears and transcribes the audio in real time. 6. To keep the window visible when nothing is playing, click the Live Captions icon in the menu bar and choose Keep Onscreen. #### How to turn off or pause Live Captions You can pause captioning at any time without turning the feature off. The window is also hidden automatically if no audio is detected, so a missing window does not always mean Live Captions is off. 1. To pause, click the pause button in the Live Captions window, or click the Live Captions icon in the menu bar and choose Stop Live Captions. 2. To resume, click the button in the window again, or choose Start Live Captions from the menu bar icon. 3. To turn the feature off completely, choose Apple menu → System Settings and click Accessibility. 4. Click Live Captions. 5. Turn Live Captions off. #### How to caption the microphone or the computer audio Live Captions has two sources. Computer audio is the sound coming out of your Mac, such as a video or the other side of a call. Microphone is the sound picked up by the computer’s microphone, such as an in-person conversation. To switch, click the button in the top-right corner of the Live Captions window, or click the Live Captions icon in the menu bar and choose Computer Audio or Microphone. By default Live Captions transcribes audio from all apps. A setting called Live Captions in FaceTime limits it to FaceTime only. #### Is Live Captions private? Apple states that for Live Captions, audio is processed on device. For FaceTime it says all audio is processed on your Mac. The only internet requirement Apple lists is the download of language data the first time you turn the feature on. Apple also sets a limit on how far to trust it. The accuracy of Live Captions may vary, and Apple says it should not be relied upon in high-risk or emergency situations. #### Live Captions vs Dictation Both features turn speech into text, and they are easy to confuse. The difference is where the text goes and whose speech it is. - Purpose: Live Captions lets you read audio you are listening to. Dictation lets you enter text by speaking. - Output: Live Captions shows text in its own window. Dictation puts text at the insertion point in the app you are using. - Source: Live Captions can follow computer audio or the microphone. Dictation listens to the microphone source chosen in Keyboard settings. - Location: Live Captions is under System Settings → Accessibility. Dictation is under System Settings → Keyboard. - Hardware: Live Captions is available only on Mac computers with Apple silicon. #### Live Captions not working or not showing Most problems come down to hardware, language, or the wrong audio source. Check each of these. - Your Mac does not have Apple silicon. Apple says Live Captions is available only on Mac computers with Apple silicon. - Your language or region is not supported. Apple’s macOS Feature Availability page lists Live Captions for varieties of English, French, German, Spanish, Cantonese and Mandarin Chinese, plus Japanese and Korean. - The language data has not downloaded. Connect to the internet and turn Live Captions on again. - The window is hidden because no audio is detected. Choose Keep Onscreen from the Live Captions icon in the menu bar. - The wrong source is selected. Switch between Computer Audio and Microphone. - Live Captions in FaceTime is turned on in Live Captions settings, which limits captions to FaceTime. Questions: - Q: How do I turn on Live Captions on my Mac? A: Choose Apple menu → System Settings, click Accessibility, click Live Captions, and turn it on. The first time, language data is downloaded, so you must be connected to the internet. - Q: Does Live Captions work on Intel Macs? A: No. Apple says Live Captions is available only on Mac computers with Apple silicon, and it is not available in all languages, countries, or regions. - Q: Does Live Captions send my audio to Apple? A: Apple states that Live Captions audio is processed on device. An internet connection is needed the first time you turn it on, to download language data. - Q: Can Live Captions caption a FaceTime call? A: Yes. During a FaceTime video call you can turn on Live Captions from the call controls, and Apple says all audio is processed on your Mac. Apple also notes that Live Captions uses additional power. - Q: Can I change the size or color of Live Captions? A: Yes. In System Settings → Accessibility → Live Captions you can change the font family, font size, font color and background color. You can also resize the window by dragging any edge. - Q: What is Type to Speak in Live Captions? A: It is a field in the Live Captions window where you type what you want to say and press Return, and your Mac speaks it out loud. If you do not see the field, click the Live Captions icon in the menu bar and choose Type to Speak. In Flit: Flit does not caption sound coming from your Mac. It types what you say into the app you are in. (https://flit.fyi/features) Related terms: transcription, streaming-transcription, on-device-processing, dictation ### Enhanced Dictation (Use Enhanced Dictation, offline dictation) URL: https://flit.fyi/glossary/enhanced-dictation Enhanced Dictation was a checkbox in the Dictation pane of Keyboard preferences on older versions of macOS that downloaded speech recognition software to the Mac so you could dictate offline. Apple’s Mac User Guide describes it for macOS High Sierra 10.13 and macOS Mojave 10.14, and it is absent from the guide for macOS Catalina 10.15 and every version since. If you remember ticking a box called Use Enhanced Dictation and cannot find it now, you are not missing a hidden setting. Current macOS has no option by that name. The job it did, turning speech into text without sending it to a server, is now part of ordinary Dictation on a Mac with Apple silicon, and part of Voice Control on any Mac that supports it. Apple’s current documentation does not announce a removal or give a date for one. What can be checked is which editions of the Mac User Guide mention the feature and which do not. #### What did Enhanced Dictation do? In the macOS Mojave guide, Apple says that turning on Enhanced Dictation makes it possible to dictate text even when you are offline, and to use a dictation keyword phrase to start dictating. Selecting the checkbox started a download of the software it needed. If you added more languages while it was on, those languages were downloaded too. With the checkbox off, the Mac used what Apple called server-based dictation. Your words were sent to Apple servers, and the final converted text appeared only after you ended dictation. - Dictation worked without an internet connection. - You could edit with the keyboard, mouse, trackpad or spoken commands while dictating. Without it, starting to edit stopped the dictation session. - You could start dictating by saying a keyword phrase. - You could speak commands to control the Mac, such as selecting text, scrolling a document or pressing Return, and you could turn on a set of advanced commands or create your own. #### Where is Enhanced Dictation in current macOS? It is not there. In macOS High Sierra and macOS Mojave the checkbox sat in System Preferences, under Keyboard, in the Dictation pane. The macOS Catalina 10.15 edition of the same help page no longer mentions Enhanced Dictation at all. It describes keyboard dictation as a feature where your words are sent to Apple servers and converted into text in real time, and it points people who want to dictate and control the Mac by voice to Voice Control. Current macOS uses System Settings instead of System Preferences, and Dictation is at System Settings, Keyboard, Dictation. You will find the Dictation switch, the languages, the shortcut and an Auto-punctuation switch there. There is no Enhanced Dictation option, and Apple’s current help page for Dictation does not use the phrase. #### What replaced Enhanced Dictation? Two things replaced it. The first is Dictation itself on a Mac with Apple silicon. The macOS Monterey 12 guide says that on these Macs general text dictation requests are processed on the device in many languages and no internet connection is required. It also says you can dictate text of any length without a timeout, and that dictation stops automatically when no speech is detected for 30 seconds. The second is Voice Control, which took over the spoken commands. Apple’s guide says Voice Control needs a one-time download the first time you turn it on, and after that it does not need an internet connection. When Voice Control is on, you use it to dictate text, and standard macOS Dictation is not available. If you want offline speech recognition outside of Apple’s features, open models such as Whisper and Parakeet also run entirely on the Mac. #### How to check whether Dictation works offline on your Mac Apple does not publish one list that says which Mac and language pairs are processed on the device. Instead, the Mac tells you in Keyboard settings. Apple’s current help page says to look at the text below the Dictation switch, and that Dictation must be turned on for the text to appear. 1. Choose Apple menu, then System Settings. 2. Click Keyboard in the sidebar. You may need to scroll down. 3. Go to Dictation and turn it on if it is off, then click Enable. 4. Read the text below Dictation. It says whether your voice inputs and transcripts for general text dictation are processed on your device and not sent to Siri servers. 5. Click About Ask Siri, Dictation & Privacy at the bottom of Keyboard settings if you want Apple’s full explanation. #### How to dictate without internet using Voice Control Voice Control is the closest thing in current macOS to the old Enhanced Dictation with dictation commands. It is an accessibility feature for entering text and controlling the Mac by speaking commands, instead of using a keyboard, mouse or trackpad. 1. Choose Apple menu, then System Settings. 2. Click Accessibility in the sidebar, then click Voice Control. 3. Turn on Voice Control. The first time, your Mac completes a one-time download. 4. When the download finishes, click in a text field and speak to dictate. No internet connection is needed from this point. 5. Turn Voice Control off again when you want to go back to standard Dictation, because the two cannot be used at the same time. #### Enhanced Dictation vs Dictation today The name is gone, but most of what it offered is now the default on a Mac with Apple silicon. The main differences are where the setting lives and which Macs get on-device processing. - Where it is: Enhanced Dictation was in System Preferences, Keyboard, Dictation. Dictation is now in System Settings, Keyboard, Dictation. - What you switch on: Enhanced Dictation was a separate checkbox with its own download. Today there is only the Dictation switch, and the text under it tells you where processing happens. - Typing while you talk: on a Mac with Apple silicon you can keep using the keyboard while you speak without stopping dictation. - Spoken commands: selecting text, scrolling and clicking by voice now belong to Voice Control, not Dictation. - Punctuation: in supported languages Dictation now inserts commas, periods and question marks for you, with an Auto-punctuation switch under Keyboard, Dictation. Questions: - Q: Where is Enhanced Dictation on my Mac? A: Current macOS has no setting called Enhanced Dictation. Apple’s Mac User Guide documents it for macOS High Sierra 10.13 and macOS Mojave 10.14, where it was a checkbox in System Preferences, Keyboard, Dictation. - Q: When was Enhanced Dictation removed from macOS? A: Apple’s help pages do not give a removal date. The Mac User Guide for macOS Mojave 10.14 describes Enhanced Dictation, and the guide for macOS Catalina 10.15 does not mention it. - Q: Can I still dictate offline on a Mac? A: Yes, on many Macs. Apple says that on a Mac with Apple silicon, dictation requests are processed on the device in many languages with no internet connection required. Voice Control also works offline after its one-time download. - Q: How do I know if Mac Dictation is sending my voice to Apple? A: Open System Settings, click Keyboard, and read the text under the Dictation switch. It states whether your voice inputs and transcripts are processed on your device and not sent to Siri servers. - Q: Did Enhanced Dictation send audio to Apple? A: Apple described Enhanced Dictation as working offline, using software downloaded to the Mac. With the checkbox off, the Mac used server-based dictation and your words were sent to Apple servers. - Q: What happened to dictation commands on Mac? A: The spoken commands that came with Enhanced Dictation are now part of Voice Control, at System Settings, Accessibility, Voice Control. When Voice Control is on, standard Dictation is not available. In Flit: Flit runs its voice models on the Mac and works with no connection, which is the job Enhanced Dictation was for. (https://flit.fyi/privacy) Related terms: dictation, on-device-processing, spoken-punctuation, whisper ### Improve Siri & Dictation (Share Audio Recordings, Siri & Dictation History) URL: https://flit.fyi/glossary/improve-siri-and-dictation Improve Siri & Dictation is an opt-in privacy setting that lets Apple store and review a sample of the audio from your Siri and Dictation interactions to develop and improve those features. On a Mac it is under System Settings, Privacy & Security, Analytics & Improvements. You usually meet this setting as a question. When you turn on Dictation or Siri, the Mac may ask whether you want to share audio recordings. Clicking Share Audio Recordings turns the setting on, and clicking Not Now leaves it off. You can change your mind at any time. The setting controls whether Apple may keep audio. It does not turn Dictation on or off, and it does not decide whether your dictation is processed on the Mac or on a server. Those are separate things with separate controls. #### What is shared with Apple when Improve Siri & Dictation is on? Apple’s privacy page says that if you opt in, Apple may store the audio of your interactions with Siri and Dictation. That audio, along with transcripts and related request data, may be used to develop and improve Siri, Dictation, Search and limited other language processing features such as Translate and Voice Control. The request history is associated with a random, device-generated identifier that rotates multiple times per hour and is not tied to your Apple Account or email address. Apple says it may retain and use the data for up to two years, and that the small subset of interactions that has been reviewed may be kept beyond two years. - Audio of your Siri and Dictation interactions. - Transcripts, including your request and the response Siri gave. - Related request data, such as the category of request, device specifications and performance statistics. - The approximate location of your device at the time of the request. - Audio, text input and transcripts of Translate requests, which the same setting covers. #### How to turn off Improve Siri & Dictation on a Mac The switch sits with the other analytics choices, not in Keyboard or Siri settings. Turning it off stops Apple from storing new audio from this Mac. Apple notes that the setting is per device, so check your other devices as well. 1. Choose Apple menu, then System Settings. 2. Click Privacy & Security in the sidebar. You may need to scroll down. 3. Go to Analytics & Improvements. 4. Turn off Improve Siri & Dictation. 5. If you also want to remove what was already shared, delete your Siri and Dictation history as described in the next section. #### How to delete Siri & Dictation history on a Mac Apple’s Mac User Guide for macOS Tahoe 26 says you can delete request history, both audio recordings and transcripts, that is associated with a random identifier and less than six months old. You must be connected to the internet to do it. Deleting history does not change your choice about sharing audio recordings, so turn the setting off first if that is what you want. The steps below are from Apple’s guide for macOS 26. On macOS 27 the sidebar item is named Siri, so look for the button in Siri settings. 1. Connect your Mac to the internet. 2. Open System Settings. 3. Click Apple Intelligence & Siri, or Siri, in the sidebar. You may need to scroll down. 4. Click Delete Siri & Dictation History. 5. Click Delete to confirm. #### What happens when Improve Siri & Dictation is off? With the setting off, Apple says your audio data is not stored by Apple. That is not the same as nothing being sent. If your dictation is not processed on the device, the things you dictate are still sent to and processed on Apple’s servers, but they are not stored unless you opt in. Apple’s Siri, Dictation & Privacy page also says Apple stores related request data by default, such as the category of request, device specifications and performance statistics. Request history, which includes transcripts and related request data, may be retained for up to two years under a random identifier, and Apple may review a subset of Siri transcripts. A history of your Siri and Dictation interactions is also kept on your device. Apple says that if you turn off both Siri and Dictation, the transcripts stored on your device and used to personalize your experience are deleted. #### Is Improve Siri & Dictation safe to leave on? Apple’s terms are specific about who can hear recordings. A subset of the data may be reviewed by graders, and audio review is conducted only by Apple employees, who are subject to strict confidentiality obligations. Depending on the request, a reviewer may see contextual data directly relevant to it, such as names of your contacts or apps installed on your devices. Apple says request history is not used to build a marketing profile and is never sold to anyone. Even so, a sample of what you say may be listened to by a person. If you dictate things you would not want another person to hear, such as client details or medical notes, leave the setting off. Dictation works the same either way. #### Improve Siri & Dictation vs on-device Dictation These are two different questions. Improve Siri & Dictation is about whether Apple may keep audio. On-device processing is about whether audio leaves your Mac to be transcribed in the first place. Apple says your device indicates in Keyboard settings whether your audio and transcripts are processed on your device and not sent to Apple servers. To check, open System Settings, click Keyboard, and read the text below the Dictation switch. Dictation must be turned on for the text to appear. A related switch in the same Analytics & Improvements list, Improve Assistive Voice Features, covers audio recordings of Voice Control and Adaptive Voice Shortcuts interactions. It is separate from Improve Siri & Dictation. - Improve Siri & Dictation: Privacy & Security, Analytics & Improvements. Controls whether Apple stores and reviews audio. - Dictation on or off: Keyboard, Dictation. Controls whether you can dictate at all. - Processed on device or on a server: shown as text below the Dictation switch in Keyboard settings. - Siri and Dictation history: deleted from Siri settings. Questions: - Q: Should I turn off Improve Siri & Dictation? A: It is optional, and Siri and Dictation work without it. When it is on, Apple may store audio of your interactions and Apple employees may review a sample, so turn it off if you dictate anything sensitive. - Q: Where is Improve Siri & Dictation on a Mac? A: Open System Settings, click Privacy & Security, go to Analytics & Improvements, and turn Improve Siri & Dictation on or off. - Q: Does Apple listen to my dictation? A: Only if you opt in to Improve Siri & Dictation. In that case Apple may store audio, and a subset may be reviewed, with audio review conducted only by Apple employees. - Q: How do I delete my Siri and Dictation history on a Mac? A: Apple’s macOS 26 guide says to open System Settings, click Apple Intelligence & Siri or Siri, click Delete Siri & Dictation History, then click Delete. You need an internet connection. - Q: How long does Apple keep Siri and Dictation recordings? A: Apple says it may retain and use request history for up to two years, and the small subset of interactions that has been reviewed may be kept beyond two years. - Q: Does turning off Improve Siri & Dictation stop audio being sent to Apple? A: No. It stops Apple from storing your audio. Whether audio is sent for processing depends on whether Dictation is processed on your device, which Keyboard settings tells you below the Dictation switch. In Flit: Flit sends nothing you say to anyone, so it has no sharing setting to turn off. (https://flit.fyi/privacy) Related terms: dictation, on-device-processing, automatic-speech-recognition ### Apple Intelligence URL: https://flit.fyi/glossary/apple-intelligence Apple Intelligence is Apple’s set of generative AI features built into macOS, including Writing Tools for rewriting, proofreading and summarizing text. It runs on Mac models with M1 or later and MacBook Neo, uses on-device models in many cases, and sends more complex requests to Apple’s Private Cloud Compute servers. For someone who writes or dictates on a Mac, Apple Intelligence matters after the words are on the page. It can rewrite a paragraph in a different tone, proofread it, or boil it down to key points. It works on text you select, in most places you can type. It is a separate thing from Dictation. Dictation turns speech into text and has its own settings and its own privacy page. None of the Apple Intelligence pages Apple publishes for the Mac list Dictation as an Apple Intelligence feature. #### Which Macs can run Apple Intelligence? Apple’s requirements page for the current release says Apple Intelligence in macOS 27 is available on Mac with M1 or later and on MacBook Neo (A18 Pro). Intel Macs are not on the list. Apple says the features are automatically available on eligible devices once you update to the 27.0 release, in supported languages and regions. Apple Intelligence was already in earlier releases. Apple’s Mac User Guide for macOS Sequoia 15 describes it as available in beta on Mac models with M1 or later, and the guide for macOS Tahoe 26 has a Turn on Apple Intelligence button under Apple Intelligence & Siri in System Settings. - Mac: M1 or later, or MacBook Neo (A18 Pro). - System: macOS 27 Golden Gate for the current feature set. Writing Tools and many other features are listed for macOS Tahoe 26 and later. - Storage: up to 14 GB on Mac models with M3 or later and at least 12GB of unified memory, and up to 8 GB on other eligible Macs. - Language: the device language and the Siri language must be set to the same supported language. - Region: Apple says the features do not currently work in China mainland, and some features vary by region and language. #### How to use Writing Tools on a Mac Writing Tools is the part of Apple Intelligence that edits text. Apple says it is available in most places you write, including third-party apps and websites. It works the same way on text you typed and text you dictated, because it only sees the selected text. 1. Select the text you want to refine or summarize. 2. Control-click the selection, choose Writing Tools, then choose Show Writing Tools. 3. Click Proofread to check the text, or click Rewrite to get a new version. 4. To set a tone for the rewrite, click Friendly, Professional or Concise. To ask for something specific, click Describe a change and type what you want. 5. Click Summarize or Key Points if you want a shorter version instead. 6. Click Done when you are finished. #### Does Apple Intelligence change Dictation? Apple does not say that it does. The requirements page lists every feature Apple Intelligence powers in macOS 27 and macOS 26, and Dictation is not among them. The Mac User Guide pages for Apple Intelligence, Writing Tools and Apple Intelligence privacy do not mention Dictation either. Dictation stays where it was, at System Settings, Keyboard, Dictation, and it works on Macs that cannot run Apple Intelligence. Its privacy terms are set out on Apple’s separate Siri, Dictation & Privacy page. In practice the two fit together in sequence: you dictate a rough draft, then select it and use Writing Tools to proofread or rewrite it. On macOS 27, Apple also offers Siri AI in beta, which can compose, rewrite, proofread and give feedback on writing. Apple says Writing Tools is available for users without Siri AI writing support. #### Is Apple Intelligence private? On device vs Private Cloud Compute Apple says that in many cases Apple Intelligence models run entirely on device, so a task can be completed without data leaving your Mac. For requests that need more computing power, the Mac can send the request to Private Cloud Compute, which runs larger models on Apple silicon servers. Apple states that data processed in Private Cloud Compute is not stored or made accessible to Apple. It is used only to fulfill your request, the results are returned to your device, and nothing is retained. Apple also says independent researchers can inspect the software that runs on those servers. Apple’s pages do not say, feature by feature, whether a given Writing Tools request stays on the Mac or goes to Private Cloud Compute. The Mac decides for each request. Apple notes that it may collect limited information about a server request, such as its approximate size and how long it took, and that this does not include the content. #### How to see what your Mac sent to Private Cloud Compute macOS can keep a log of requests that left your Mac for Private Cloud Compute, and you can export it. Apple’s Mac User Guide calls the setting Apple Intelligence Report, and Apple’s privacy page calls it Apple Intelligence & PCC Report. 1. Open System Settings and click Privacy & Security in the sidebar. 2. Click the Apple Intelligence report setting. 3. Choose a report duration: the last 15 minutes or the last 7 days. 4. Use Apple Intelligence as you normally would, so that there are requests to record. 5. Come back and click Export Activity, choose where to save the file, then click Export. 6. Open the saved file, named Apple_Intelligence_Report.json, in a text reader. #### How to turn off Apple Intelligence features How you turn it off depends on the version of macOS. In macOS Tahoe 26, Apple’s guide describes a single control under Apple Intelligence & Siri in System Settings. In macOS 27, Apple’s guide lists the features one by one, each with its own switch. - Siri AI: open System Settings, click Siri in the sidebar, click Turn Off Siri, then confirm. - Message summaries: in Messages, choose Messages, Settings, then turn off Summarize Messages. - Email summaries: in Mail, choose Mail, Settings, click Viewing, then turn off Summarize Message Previews. - Notification summaries: open System Settings, click Notifications, click Summarize Notifications, then turn it off. - Limits for yourself or a child: use the Siri restrictions in Screen Time settings. Questions: - Q: Does my Mac support Apple Intelligence? A: Apple says Apple Intelligence in macOS 27 is available on Mac with M1 or later and on MacBook Neo (A18 Pro). Intel Macs are not listed. - Q: Does Apple Intelligence make Mac Dictation better? A: Apple does not say so. Dictation does not appear in Apple’s list of Apple Intelligence features, and Dictation has its own settings and privacy page. - Q: How do I open Writing Tools on a Mac? A: Select some text, Control-click it, choose Writing Tools, then choose Show Writing Tools. From there you can proofread, rewrite, summarize or describe a change. - Q: Does Apple Intelligence send my writing to Apple? A: Apple says many requests are handled entirely on device, and more complex ones go to Private Cloud Compute. Apple states that data sent there is not stored or made accessible to Apple and is used only to fulfill the request. - Q: How much storage does Apple Intelligence need on a Mac? A: Apple lists up to 14 GB on Mac models with M3 or later and at least 12GB of unified memory, and up to 8 GB on other eligible Macs. - Q: Can I use Writing Tools on text I dictated? A: Yes. Writing Tools works on any text you select in most places you write, so it does not matter whether the text was typed or dictated. In Flit: On macOS 26 or later, Flit can use Apple’s on-device model to format what you say. The other choice is a formatting model you download. (https://flit.fyi/setup#formatting) Related terms: dictation, on-device-processing, formatting-pass, tone-matching ### SpeechAnalyzer (SpeechTranscriber, Speech framework) URL: https://flit.fyi/glossary/speechanalyzer SpeechAnalyzer is a class in Apple’s Speech framework, introduced with macOS 26 and iOS 26, that apps use to turn spoken audio into text. It runs an analysis session with modules such as SpeechTranscriber, using speech models that are downloaded from Apple and managed by the system. You will not find SpeechAnalyzer in System Settings. It is a developer API, the plumbing an app calls when it wants speech-to-text without shipping its own model. Apple says the same technology already powers features in Notes, Voice Memos and Journal. It matters to Mac users for two reasons. Apps built on it transcribe on the device, and they share one set of system-managed models instead of each app adding its own model to your disk. #### What does SpeechAnalyzer do? Apple’s documentation describes SpeechAnalyzer as the object that analyzes spoken audio and manages the analysis session. The analyzer itself does not transcribe. It holds modules, and each module does one kind of analysis. Apple says many use cases only need the SpeechTranscriber module, which performs speech-to-text. The API is written for Swift concurrency. An app feeds audio in as an asynchronous sequence and reads results from another asynchronous sequence, so input, output and session control are decoupled. An analyzer can only analyze one input sequence at a time. - SpeechAnalyzer: holds the modules, accepts audio input and controls the overall analysis. - SpeechTranscriber: a transcription module that Apple describes as appropriate for normal conversation and general purposes. - DictationTranscriber: a transcription module similar to system dictation features and compatible with older devices. - SpeechDetector: a voice activity detection module that can gate transcription on the presence of speech to save power. - AssetInventory: installs and tracks the model assets the modules need. #### SpeechAnalyzer vs SFSpeechRecognizer SFSpeechRecognizer is the older speech API, available since iOS 10 and macOS 10.15. In its WWDC25 session Apple said that class gave access to the speech-to-text model powering Siri, worked well for short-form dictation and could use Apple servers on resource-constrained devices. Apple’s documentation for it still describes speech recognition as a network-based service with usage limits. Apple says the new model behind SpeechTranscriber is both faster and more flexible than the one available through SFSpeechRecognizer, and good for long-form and distant audio such as lectures, meetings and conversations. Apple’s reference page does not mark SFSpeechRecognizer as deprecated, so existing apps keep working. - Where it runs: Apple says transcription with the new API is entirely on device. SFSpeechRecognizer may require an internet connection for some languages. - Audio length: Apple tells developers to plan for a one-minute limit on audio duration with SFSpeechRecognizer. The new model is built for long-form audio. - Keeping audio local: with SFSpeechRecognizer a developer must set requiresOnDeviceRecognition, and Apple notes on-device requests will not be as accurate. - Languages: Apple says the old class relied on the user to add languages. The new API lets the app request the model for a language itself. - Results: the new API can deliver quick volatile results that are later replaced by finalized ones. #### SpeechTranscriber vs DictationTranscriber SpeechTranscriber uses Apple’s new speech-to-text model. Apple says it is available for all platforms but watchOS, with certain hardware requirements, and supports a set of languages with more to come. An app is expected to check whether the current device supports it before using it. DictationTranscriber is the fallback for an unsupported language or device. Apple’s documentation says it uses the same speech-to-text models as system dictation features, or as SFSpeechRecognizer when configured for on-device operation. It does not support languages that SFSpeechRecognizer only supports through network access. Apple adds that, unlike the old class, the user does not have to go into Settings and turn on Siri or keyboard dictation for a language. #### How to transcribe audio with SpeechAnalyzer Apple’s reference page gives a general sequence for analyzing audio files and streams. The steps below follow that sequence for a developer building a Mac app in Swift on macOS 26 or later. 1. Create and configure the modules you need, for example a SpeechTranscriber for a locale. 2. Make sure the relevant assets are installed or already present, using AssetInventory. 3. Create an input sequence that will carry the spoken audio, from a file or from a capture device such as a microphone. 4. Create the analyzer with the modules and the input sequence. 5. Supply audio and start the analysis. 6. Read results from the module as they arrive and act on them. 7. Finish the analysis when you are done, so the result streams end. #### How to install the speech models with AssetInventory The models are not bundled with the app. Apple describes the assets as machine-learning models downloaded from Apple’s servers and managed by the system. In the WWDC25 session Apple says the model is retained in system storage, does not increase the download or storage size of the app, and runs outside the app’s memory space. Once an asset is downloaded, the system retains and updates it automatically and shares it with other apps. Each app gets a limited number of locale reservations. 1. Create the analyzer modules in the configurations you plan to use. 2. Assign your app’s asset reservations to those locales. AssetInventory does this automatically if needed, or you can reserve a locale manually. 3. Request an asset installation for those modules and call its download and install method. 4. Wait for the download to finish. It may finish immediately if the assets were preinstalled or another app already downloaded them. 5. When your app no longer needs a locale, release the reservation so the system can remove the assets later. #### Is SpeechAnalyzer private? Which platforms support it? Apple’s WWDC25 session states that transcription is entirely on device but the models need to be fetched. The download comes from Apple’s servers, and after that the audio an app transcribes with SpeechTranscriber is processed on the device. What an app then does with the resulting text is up to that app, so its own privacy policy still applies. Apple’s documentation lists SpeechAnalyzer and SpeechTranscriber for iOS 26, iPadOS 26, Mac Catalyst 26, macOS 26, tvOS 26 and visionOS 26. DictationTranscriber is listed for the same platforms except tvOS. On a Mac, apps that have asked for speech recognition access appear at System Settings, Privacy & Security, Speech Recognition. SpeechAnalyzer is one way for a Mac app to transcribe locally. Open models such as Whisper and Parakeet are another, and they are bundled or downloaded by the app instead of being managed by the system. Questions: - Q: What is SpeechAnalyzer? A: SpeechAnalyzer is a class in Apple’s Speech framework that analyzes spoken audio and manages the analysis session. With the SpeechTranscriber module it performs speech-to-text in apps on macOS 26, iOS 26 and later. - Q: Does SpeechAnalyzer work offline? A: Apple says transcription is entirely on device, but the models need to be fetched first. Once the assets are installed, the system retains them and shares them with other apps. - Q: Which macOS version does SpeechAnalyzer need? A: Apple’s documentation lists SpeechAnalyzer as available from macOS 26.0. It is also available on iOS 26, iPadOS 26, Mac Catalyst 26, tvOS 26 and visionOS 26. - Q: Is SFSpeechRecognizer deprecated? A: Apple’s reference page for SFSpeechRecognizer does not mark it as deprecated. Apple introduced SpeechAnalyzer as a new API that supports more use cases, and says its model is faster and more flexible. - Q: What is the difference between SpeechTranscriber and DictationTranscriber? A: SpeechTranscriber uses Apple’s new model for general and long-form transcription on supported devices. DictationTranscriber uses the same models as system dictation features and is the fallback for older devices and other languages. - Q: Do SpeechAnalyzer models take up space in my app? A: No. Apple says the model is retained in system storage and does not increase the download or storage size of the app, and it runs outside the app’s memory space. In Flit: On macOS 26 or later, Apple Speech is one of the voice models you can choose in Flit, with nothing to download. (https://flit.fyi/setup#engine) Related terms: speech-to-text, automatic-speech-recognition, on-device-processing, voice-activity-detection ### Dictation URL: https://flit.fyi/glossary/dictation Dictation is producing written text by speaking, with the words transcribed in or near real time at the place you are writing. It differs from transcription, which converts an existing recording into text after the fact. Modern dictation software listens while you speak and inserts text at the cursor of whatever app you are using, so the output lands where an email, document, or message is being written. Quality differs mainly in three places: how accurately speech is recognized, how finished the resulting text is (punctuation, capitalization, numbers), and how reliably it works across applications. People adopt dictation for speed (speech runs well over 100 words a minute against roughly 31 to 40 for composing at a keyboard), for health reasons such as RSI, or because thinking out loud produces better first drafts. #### What a dictation tool actually does Four things happen between the key going down and text appearing. The microphone is opened and audio is captured. A speech recognition model turns that audio into words, either as it arrives or once you stop. Some kind of cleanup turns the words into writing: punctuation, capitals, numbers, fillers removed. Then the text is inserted into the focused app, usually through the operating system’s accessibility layer, so it lands at the cursor as if typed. Products differ in each step. Cloud tools do the second and third steps on a server. Raw transcribers skip the third. Some tools can only insert into their own window and make you paste. When you compare dictation apps, you are really comparing where each of the four steps happens and how well. #### Dictation versus transcription The two words get used interchangeably and should not be. Dictation is live: you are composing, the tool is keeping up, and the result lands where you are working. Transcription is after the fact: a recording exists, and a tool turns it into a document. Meeting notes, interview write-ups, and podcast subtitles are transcription jobs, and they need features dictation does not, like speaker labels and timestamps. Most tools are built for one job and do the other badly or not at all. A meeting transcriber is a poor way to write an email, and a dictation app is a poor way to process an hour of recorded audio. Pick by the job. #### What to look for Ask where the audio is processed, because that decides privacy and offline behavior in one go. Ask how finished the text is when it lands, because a raw transcript is faster to speak than to fix. Ask whether it types into the apps you actually use, including browser-based ones, because a tool that only works in a notes window is a tool you will stop opening. And try it in a terminal or a code editor, where any helpful rewrite becomes a bug. #### How to turn on Dictation on a Mac Every Mac has dictation built in, and it is off until you switch it on. The setup takes about a minute and happens in one place, Keyboard settings. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and turn it on. 3. Click the Edit button next to Languages and select the language and region you speak. 4. Open the pop-up menu next to Microphone source and choose a microphone, or leave it on Automatic. 5. Open the Shortcut pop-up menu and choose a shortcut, or choose Customize and press the keys you want. 6. Click into any text field and press the shortcut, press the Microphone key if your keyboard has one, or choose Edit → Start Dictation. 7. Press the shortcut again or press Escape to stop. #### Why does dictation stop on its own? Apple’s Dictation has no limit on length. Apple says you can dictate text of any length without a timeout. What it does have is a silence rule: it stops automatically when no speech is detected for 30 seconds. If dictation will not start at all, check whether Voice Control is on. When Voice Control is on, standard Dictation is not available, and you dictate through Voice Control instead. If it starts but hears nothing, look at Microphone source in Keyboard settings and confirm that the microphone you are speaking into is the one selected. #### How to dictate punctuation and formatting by voice In supported languages, Apple’s Dictation inserts commas, periods, and question marks for you. Everything else you can say out loud. These are spoken commands from Apple’s own list, and many other dictation tools accept similar ones. - Say "new line" or "new paragraph" to break the text. - Say "caps on" and "caps off" to capitalize the first letter of each word in between. - Say "all caps" to put the next word in capital letters. - Say "numeral" before a number to get digits instead of a spelled-out word. - Say "no space on" and "no space off" to join words together. - Say the name of a mark, such as "semicolon", "open parenthesis", or "question mark", to insert it. Questions: - Q: Is dictation faster than typing? A: For producing text, yes, by roughly three to four times. Conversational speech runs 120 to 160 words a minute; composing original text at a keyboard runs around 31 to 40. The gap narrows when you count the time spent fixing a raw transcript, which is why the formatting quality of a dictation tool matters as much as its recognition accuracy. - Q: Does dictation work in every app? A: It depends on how the tool inserts text. Tools that type through the operating system’s accessibility layer work anywhere a cursor can go, including browser apps and Electron apps. Tools that only work inside their own window make you copy and paste, which is where most people give up. - Q: Do I need to speak differently to dictate? A: Less than you would think, with a tool that removes fillers and false starts. You can talk the way you talk to a colleague. The one habit worth building is finishing a sentence before you release the key, so the tool has a whole thought to punctuate. - Q: How do I turn on dictation on a Mac? A: Open System Settings, click Keyboard, go to Dictation, and turn it on. Then choose a shortcut in the same section and press it in any text field to start speaking. - Q: Can I type while dictating on a Mac? A: On a Mac with Apple silicon, yes. Apple says you can keep using the keyboard while you speak, with no need to stop dictation first. - Q: Is there a time limit on Mac dictation? A: No. Apple says you can dictate text of any length without a timeout. Dictation only stops by itself when it detects no speech for 30 seconds. In Flit: Flit is dictation in this exact sense: hold a key, speak, and finished text lands at your cursor, entirely on your Mac. (https://flit.fyi/#demo) See it on flit.fyi: - Watch a dictation land in Messages, Slack, Mail, and an editor: https://flit.fyi/#video - Everything Flit does: https://flit.fyi/features - Dictation vs typing: the numbers: https://flit.fyi/blog/dictation-vs-typing Related terms: speech-to-text, voice-typing, transcription, streaming-transcription, words-per-minute ### Voice typing URL: https://flit.fyi/glossary/voice-typing Voice typing is another name for dictation: entering text into an application by speaking instead of using the keyboard. The term was popularized by the Voice Typing feature in Google Docs. In practice, "voice typing" and "dictation" describe the same activity. Search usage differs slightly: voice typing tends to be used for typing replacement in everyday apps, while dictation is the older word inherited from the era of dictating to a person or a recorder. The practical questions are identical either way: where does the audio go, how finished is the text, and how many of your applications does it work in. #### Where the phrase comes from Google Docs shipped a menu item called Voice Typing in 2015, and Android’s keyboard uses the same words for its microphone key. Both send audio to Google’s servers, transcribe it there, and return text into a Google product. That is where most people first met the phrase, which is why "voice typing" tends to carry an unspoken assumption: it works in one app, and it works in the cloud. Neither assumption holds for the category as a whole. A voice typing tool on a Mac can type into every app on the machine, and it can run entirely on the machine. #### Built-in voice typing versus a dedicated tool Every modern operating system has voice typing built in: Apple Dictation on the Mac, Voice Typing on Windows, Gboard on Android. They are free, already installed, and fine for a short message. What they tend to lack is the finishing: fillers stay in, numbers and times come out as words, the tone is the same in every app, and there is no vocabulary that learns your names. A dedicated tool earns its price in the cleanup and in the apps that built-in dictation struggles with, like terminals, code editors, and browser-based apps. If you write a few sentences a day, the built-in one is enough. If you write for a living, the difference is measured in hours a week. #### How to start voice typing on a Mac from the keyboard Voice typing on a Mac is the Dictation feature. Once it is on, you can start it three ways, and you can decide which key does it. 1. Choose Apple menu → System Settings, click Keyboard, go to Dictation, and turn it on. 2. Open the Shortcut pop-up menu in the Dictation section and choose a shortcut, or choose Customize and press your own keys. 3. If you want the 🌐 key to do the job, find "Press 🌐 key to" in Keyboard settings and choose Start Dictation. 4. Click where you want the text to go, in any app. 5. Press your shortcut, press the Microphone key in the row of function keys if your keyboard has one, or choose Edit → Start Dictation. 6. Speak, then press the shortcut again or press Escape when you are done. #### How to voice type in another language Dictation listens for one language at a time, so the language has to be added before you can speak it. You can keep several and switch between them without opening settings again. 1. Choose Apple menu → System Settings, then click Keyboard. 2. Go to Dictation and click the Edit button next to Languages. 3. Select each language and region you want, then confirm. 4. Start dictation in a text field. 5. Click the language shown next to the cursor, or press the 🌐 key, to switch to another language you added. #### Voice typing vs Voice Control on a Mac macOS has two features that turn speech into text, and they are built for different jobs. Dictation is voice typing: you start it, speak, and stop it. Voice Control, at System Settings → Accessibility → Voice Control, is for running the whole Mac by voice, including clicking, scrolling, and editing. Voice Control has three modes. In dictation mode, which is the default, any words that are not commands are entered as text. In command mode, it responds only to commands and ignores other words. In spelling mode, you dictate character by character, which helps with web addresses and similar strings. You cannot run both at once. When Voice Control is on, standard Dictation is not available. The first time you turn Voice Control on, the Mac must be online to complete a one-time file download from Apple. Questions: - Q: Is voice typing the same as dictation? A: Yes. They describe the same activity: speaking, and having the words appear as text where you are writing. "Voice typing" is the newer, more casual phrase; "dictation" is the older word and the one most macOS tools use. - Q: Does the Mac have voice typing built in? A: Yes, as Apple Dictation, available from a keyboard shortcut in most apps. It is free and processes supported languages on the Mac. It produces a fairly raw transcript with limited formatting, and it does not behave well in terminals or code editors. - Q: How do I voice type on a MacBook? A: Turn on Dictation in System Settings → Keyboard, click into a text field, and press the Dictation shortcut or the Microphone key in the function row. Speak, then press Escape or the shortcut again to stop. - Q: Why is voice typing not working on my Mac? A: Check three things. Dictation must be turned on in System Settings → Keyboard, Voice Control must be off because standard Dictation is not available while it is on, and Microphone source in the Dictation section must point at the microphone you are using. - Q: Can I add emoji with voice typing on a Mac? A: Yes. With Apple’s Dictation you say the name of the emoji followed by the word emoji, such as "heart emoji" or "car emoji". In Flit: Flit is a voice typing tool for the whole Mac: every app with a cursor, not one product’s document. (https://flit.fyi/alternatives/apple-dictation) See it on flit.fyi: - How Flit compares with the built-in dictation: https://flit.fyi/alternatives/apple-dictation - The best dictation apps for Mac, ranked: https://flit.fyi/best-dictation-app-for-mac Related terms: dictation, speech-to-text, accessibility-permission ### Transcription (file transcription) URL: https://flit.fyi/glossary/transcription Transcription is converting an existing audio or video recording into text after the fact. It is the after-the-fact counterpart to dictation, which produces text live as you speak. Transcription starts from a file: a recorded meeting, an interview, a lecture, a voice memo. The tool processes the whole recording, often faster than real time, and returns a document. Because the audio already exists, transcription tools can afford slower and more accurate models, run several passes, and add structure that live dictation never needs: timestamps, speaker labels, paragraphing by topic, summaries. The confusion with dictation is understandable, because the same speech recognition models sit inside both. What differs is everything around the model, and it is rare for one product to be excellent at both jobs. #### What transcription needs that dictation does not Speaker diarization, so the transcript says who said what. Timestamps, so a quote can be found in the recording. Long-form handling, because an hour of audio is a different problem from a sentence. Summaries and action items, because nobody reads a full meeting transcript. Import from files and links, including video. None of this helps you write an email faster. A dictation tool spends its effort elsewhere: latency, formatting at the cursor, working inside every app, and behaving in a terminal. #### Choosing the right kind of tool If your recording already exists, you want a transcriber. On the Mac, MacWhisper and Superwhisper both transcribe files locally; Otter.ai transcribes meetings in the cloud with speaker labels and summaries. If you are composing, you want dictation. Some people need both, and two focused tools tend to beat one that stretches. #### How to transcribe a recording on a Mac with Voice Memos The Voice Memos app that comes with macOS can show a transcript of a recording. Apple notes that audio transcription is not available in all countries or regions, so the button may be missing on your Mac. 1. Open the Voice Memos app on your Mac. 2. Click the recording you want to read. 3. Click the Transcription button to view the transcript. 4. Select the text you want, Control-click it, and choose Copy. 5. Paste the text into Mail, a document, or any other app. 6. To find a recording by what was said, type the words into the search field. Voice Memos shows recordings with that text in the title or in the transcript. #### How to transcribe an audio file in Terminal with Whisper Whisper is an open speech recognition model, and its code and model weights are released under the MIT License. It ships with a command-line tool that takes audio files and returns text. The project lists Python 3.8 to 3.11 as compatible, and it needs the ffmpeg tool to read audio. 1. Open Terminal. 2. Install ffmpeg. With Homebrew the command is: brew install ffmpeg 3. Install Whisper with: pip install -U openai-whisper 4. Run it on a file with: whisper audio.mp3 --model turbo 5. If you know the language of the recording, name it, for example: whisper japanese.wav --language Japanese 6. To get English text from speech in another language, add --task translate and choose a model other than turbo, for example --model medium. #### Is automatic transcription accurate? It depends on the recording and on the model. With Whisper you choose the trade-off yourself. The project publishes six sizes, from tiny at 39 million parameters to large at 1,550 million, and rates tiny at roughly ten times the speed of large. The turbo model has 809 million parameters and is rated at roughly eight times the speed of large. Long files are not handled in one pass. Whisper reads the whole file and processes it with a sliding 30-second window, so an error in one stretch does not have to spoil the rest. Whatever tool you use, the honest test is the same: pick a few minutes that matter, play them back, and read along with the transcript before you trust the full document. Questions: - Q: Is transcription the same as dictation? A: No. Transcription turns an existing recording into text; dictation turns your live speech into text where you are writing. The same recognition models can power both, but the products are built differently and are rarely good at both jobs. - Q: Can a dictation app transcribe a recording? A: Some can, as a secondary feature. Flit does, since 1.5, on the Mac, with timestamps in its SRT and VTT exports but no speaker labels. If speaker labels are the job, choose a transcriber built for them. - Q: Which local Mac tools transcribe files? A: MacWhisper is the strongest, with a free tier and a paid Pro version with speaker labels. Superwhisper and Flit both transcribe recorded files as well as dictating. All three run local models on Apple silicon. - Q: How do I transcribe audio to text on a Mac for free? A: Two options cost nothing. The Voice Memos app shows a transcript when you click the Transcription button on a recording, where the feature is available. The open Whisper model can be installed with pip and run on an audio file from Terminal. - Q: Can Whisper translate audio into English? A: Yes. Adding --task translate to the whisper command produces English text from speech in another language. The project says the turbo model is not meant for translation, so choose another multilingual model such as medium or large. - Q: Can I copy the text from a Voice Memos transcript? A: Yes. Open the transcript on your Mac, select the text, Control-click it, and choose Copy. You can then paste it into any other app. In Flit: Flit transcribes files since 1.5: drop an audio or video recording on the Transcribe tab and the transcript comes back on your Mac, exported as text, Markdown, SRT or VTT. No speaker labels yet. (https://flit.fyi/features#transcribe) See it on flit.fyi: - Flit vs MacWhisper, the file transcriber: https://flit.fyi/alternatives/macwhisper - Flit vs Superwhisper, which does both: https://flit.fyi/alternatives/superwhisper - Flit vs Otter.ai, the meeting transcriber: https://flit.fyi/alternatives/otter-ai Related terms: dictation, speaker-diarization, speech-to-text, whisper ### Push-to-talk (hold-to-talk) URL: https://flit.fyi/glossary/push-to-talk Push-to-talk is an activation style where recording runs only while a key is held down: press, speak, release, done. It borrows its name from two-way radios. For dictation, holding a key is the lowest-friction contract available: there is no window to open, nothing to click, and the release of the key is an unambiguous signal that you are finished, which lets the software commit the text immediately. Common alternatives are toggle activation (tap to start, tap to stop) for longer passages, and voice activation, where the tool arms itself and starts when it hears speech. #### Why the key-up matters The moment you release the key tells the software two things at once: the utterance is complete, and you want the text now. A tool that streams recognition while the key is held has almost nothing left to do at key-up, which is where sub-second results come from. A tool that waits for key-up to start recognizing has to process the whole recording before a word appears, and the pause grows with the length of what you said. Push-to-talk also solves the hardest problem in voice input, which is knowing when you have stopped. Voice activation has to guess from a pause, and every guess is either too eager or too slow. A key is never wrong. #### Which key The Fn key is the common default on a Mac because it sits alone in the bottom corner, is rarely used for anything else, and can be held with one finger while the other hand keeps working. Any modifier or key combination works; the practical constraints are that it should be reachable without looking and should not collide with a shortcut in the app you are dictating into. #### Hold, toggle, or hands-free Hold for a sentence or two, which is most dictation. Toggle, where a tap locks recording on until the next tap, for a paragraph or a stream of thought when holding a key gets tiring. Hands-free, where the tool is armed once and starts and stops on its own, for people who cannot use the keyboard at all or want to dictate while doing something else. A good tool offers all three and lets you cancel any of them with Escape. #### How to choose the key that starts dictation on a Mac You can try a single dictation key today with the dictation that is built into macOS. The same steps tell you which keys are already taken before you assign one in any other tool. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and turn it on. 3. Open the Shortcut pop-up menu and choose one of the listed shortcuts. 4. To use your own keys, choose Customize and press the keys you want. 5. Click into a text field and press the shortcut to start. 6. Press the shortcut again or press Escape to stop. #### Is Apple Dictation push-to-talk? No. Apple’s Dictation is a toggle with a silence timeout. You press the shortcut to start, and it keeps listening until you press the shortcut again, press Escape, or say nothing for 30 seconds. That difference changes how it feels. With a toggle, the microphone stays open after your last word until you remember to close it. With a held key, the microphone closes the instant your finger lifts, so there is never a question about whether it is still listening. #### Why does the Fn key open the emoji picker or switch languages? On current Mac keyboards the Fn key is also the 🌐 key, and macOS gives it a job of its own. If a tap of your push-to-talk key keeps opening the emoji picker or changing the input source, that setting is the cause. You can change it in Keyboard settings. 1. Choose Apple menu → System Settings, then click Keyboard. 2. Find the pop-up menu labeled "Press 🌐 key to". 3. Choose Do Nothing if another tool should own the key. 4. Choose Start Dictation instead if you want the key to start Apple’s Dictation. 5. Go back to a text field and press the key to confirm that the old behavior is gone. Questions: - Q: What is push-to-talk in a dictation app? A: Recording runs only while you hold a key. Release the key and the text is committed. It is the fastest way to dictate a sentence because the release is an exact signal that you are done. - Q: Which key should I use for push-to-talk on a Mac? A: Fn is the usual default: it sits alone in the bottom corner and rarely does anything else. Any key or combination that you can hold without looking and that no app you use has claimed will do. - Q: What if I want to dictate for a long time? A: Use a toggle mode, where one tap locks recording on and another tap ends it, or a hands-free mode that starts when you speak and stops when you pause. Most tools that offer push-to-talk offer at least one of these too. - Q: How do I change the dictation shortcut on a Mac? A: Open System Settings, click Keyboard, and go to Dictation. Open the Shortcut pop-up menu and choose a shortcut, or choose Customize and press the keys you want to use. - Q: Is Apple Dictation hold-to-talk or a toggle? A: It is a toggle. You press the shortcut once to start and press it again or press Escape to stop, and it also stops by itself when no speech is detected for 30 seconds. - Q: How do I stop the Fn key from opening the emoji picker on a Mac? A: Open System Settings, click Keyboard, and change "Press 🌐 key to" from Show Emoji & Symbols to Do Nothing. The other choices are Change Input Source and Start Dictation. In Flit: Flit defaults to holding Fn, with tap-to-lock and a hands-free mode as alternatives, and Esc cancels any of them. (https://flit.fyi/#start) See it on flit.fyi: - Hold, tap, and hands-free, side by side: https://flit.fyi/#start - Flit for hands-free and RSI: https://flit.fyi/use-cases/accessibility - Picking your key on first run: https://flit.fyi/download Related terms: voice-activity-detection, wake-word, latency, dictation ### Voice activity detection (VAD) URL: https://flit.fyi/glossary/voice-activity-detection Voice activity detection (VAD) is the technique of detecting when someone is speaking versus when there is silence or background noise. It lets software start and stop recording automatically. A VAD model is small and fast, and runs continuously on the audio stream without transcribing anything: its only job is a yes/no answer to "is this speech?". Dictation tools use it for hands-free modes (start when speech begins, stop at a pause) and to trim silence before the recognizer runs. A well-behaved implementation only arms the microphone when you ask it to, and discards audio that never became a dictation. #### How it works The classic approach is energy-based: if the microphone signal is louder than a threshold for long enough, it is speech. That breaks in any room with a fan, a keyboard, or a second person. Modern VAD is a tiny neural network trained to tell human speech from everything else, and it runs in a few milliseconds per chunk of audio, cheap enough to run all the time on a laptop without anyone noticing. The output is a stream of speech and non-speech segments. A dictation tool uses the first transition to start the recognizer and a long enough non-speech stretch to decide that you have finished. #### Why it matters for privacy A hands-free mode means a tool is listening before you have said anything. What separates a trustworthy implementation from a creepy one is what happens to audio that turns out not to be speech, or speech that never becomes a dictation. The right answer is that it is never stored, never sent anywhere, and evaporates as soon as the detector says no. A tool that runs VAD on-device can make that promise; a tool that streams to a server to decide whether you are speaking cannot. #### Where VAD falls short Detecting a pause is not the same as detecting the end of a thought. People pause mid-sentence to think, and a VAD-driven stop will cut them off; set the pause longer and the tool feels sluggish. That is why hands-free modes suit steady, fluent dictation and push-to-talk suits everything else, and why a good tool offers both. #### How to see silence detection at work on your Mac You do not need special software to watch a speech detector make a decision. Apple’s Dictation stops automatically when no speech is detected for 30 seconds, and you can trigger that rule yourself. 1. Choose Apple menu → System Settings, click Keyboard, go to Dictation, and turn it on. 2. Click into an empty note or document. 3. Press the Dictation shortcut and say one sentence. 4. Stop talking and leave the Mac alone. 5. Watch dictation end without a key press once 30 seconds pass with no speech. 6. Repeat the test with music or a fan running nearby to see whether background sound changes the result on your setup. #### Why does hands-free dictation cut me off or miss my first word? A speech detector can only judge the signal it is given. If the wrong microphone is selected, or the input is very quiet, your voice and the room start to look alike, and the start and stop decisions get worse. Before you blame the tool, check what the Mac is actually listening to. - In System Settings → Sound, click Input and select the microphone you are really speaking into. - On the same screen, drag the input volume slider up if your voice is too quiet, or down if it is too loud. - For Apple’s Dictation, check Microphone source under System Settings → Keyboard → Dictation. Automatic lets the Mac pick the device you are most likely to use. - Move closer to the microphone or use a headset, so your voice is clearly louder than the room. - If you pause to think in the middle of sentences, use a held key or a toggle for those passages instead of a hands-free mode. #### How to check whether your microphone is open Any tool that waits for speech has to keep the microphone open while it waits. macOS shows you when that is happening and lets you decide which apps may do it at all. 1. Open Control Center and look for the Recording Indicator. It shows when the microphone is in use or was used recently. 2. Choose Apple menu → System Settings, then click Privacy & Security in the sidebar. 3. Click Microphone. 4. Read the list of apps that have asked for the microphone. 5. Turn access off for any app that should not be listening. Questions: - Q: What does VAD do in a dictation app? A: It decides whether the microphone is picking up speech. In a hands-free mode it starts recognition when you begin talking and stops when you pause, so you never touch a key. - Q: Is a hands-free mode always listening? A: Only once you arm it, in a well-designed tool. Even then the detector is deciding "speech or not" rather than transcribing, and audio that never becomes a dictation should be discarded immediately. Ask where that decision runs; on the device is the answer you want. - Q: What does VAD stand for? A: VAD stands for voice activity detection. It is the part of a voice tool that decides whether the incoming sound is speech or not, without working out what the words are. - Q: How long does Mac dictation wait before it stops listening? A: Apple’s Dictation stops automatically when no speech is detected for 30 seconds. You can end it sooner by pressing the Dictation shortcut again or pressing Escape. - Q: How can I tell if an app is using my microphone on a Mac? A: Open Control Center and look for the Recording Indicator, which shows when the microphone is in use or was used recently. To see which apps are allowed to use it, go to System Settings → Privacy & Security → Microphone. In Flit: Flit’s hands-free mode is armed by you, starts the moment it hears speech, stops when you pause, and records nothing until it hears you. (https://flit.fyi/#start) See it on flit.fyi: - The three activation modes: https://flit.fyi/#start - Dictating without touching the keyboard: https://flit.fyi/use-cases/accessibility - What Flit records and when: https://flit.fyi/privacy Related terms: push-to-talk, wake-word, streaming-transcription, on-device-processing ### Wake word (hotword) URL: https://flit.fyi/glossary/wake-word A wake word is a spoken phrase that turns a listening device on, like "Hey Siri" or "Alexa". A small always-on model listens for that one phrase and ignores everything else until it hears it. Wake words solve the problem of starting a voice interaction when there is no key to press: a speaker on a shelf, a phone in a pocket. The detector is deliberately tiny and deliberately narrow, tuned to one phrase, so it can run constantly at very low power without transcribing the room. For desktop dictation, a wake word is usually the wrong tool. You are sitting at a keyboard, so a key is faster, more precise, and never triggers by accident. Hands-free dictation on a computer is better served by voice activity detection: arm the tool once, and it starts when you speak. #### Wake word versus voice activity detection Both are always-on listeners, and both run small models, but they answer different questions. A wake-word detector asks "was that my phrase?" and treats everything else as silence. A voice activity detector asks "is that speech at all?" and treats every utterance as a potential dictation. The first is for waking a device that is otherwise asleep; the second is for a tool you have already told to listen. #### The privacy question A wake-word detector has to hear everything to catch the phrase, which is why the question with any always-listening product is what happens to the audio it decides was not the phrase. On a well-built device it is discarded on the device within moments. The recurring news stories about smart speakers come from the moment after the wake word, when audio is sent to a server, and from false triggers that send audio nobody meant to send. #### How to set or change the Siri wake word on a Mac The Mac has one wake word feature built in, and it belongs to Siri. Apple notes that "Hey Siri" or "Siri" may not be available for every device, language, or region, so you may not see every option below. 1. Open the System Settings app on your Mac. 2. Click the Siri settings in the sidebar. Depending on your version of macOS the item is named Siri, or Apple Intelligence & Siri. 3. Find the setting for the phrase Siri listens for. 4. Choose the phrase you want to speak to start Siri: "Hey Siri" or "Siri". 5. Decide whether to turn on "Allow Siri when locked". With it on, the phrase also works when the Mac is locked or in sleep. 6. Say the phrase from where you normally sit to confirm that it responds. #### How does a wake word detector work? Apple’s machine learning team has published a description of the "Hey Siri" detector on iPhone, and it is a good picture of the general design. A very small speech recognizer runs all the time and listens for just those two words. It turns the incoming sound into a confidence score that the phrase was spoken. The check happens in two stages. A low-power, always-on processor runs a small model first. When the score passes a threshold, it wakes the main processor, which checks the same sound again with a larger model. The setup also has you say the phrase several times, so the detector can compare later attempts with your voice. #### Why does a wake word trigger by accident, or not at all? Every wake word detector comes down to a threshold on a confidence score. Set it low and the device wakes for phrases that only sound similar. Set it high and it ignores you. Engineers measure the two errors separately: the false-accept rate is the number of false activations per hour, and the false-reject rate is the share of real attempts that fail. Apple describes one way to soften the trade-off. When a score lands just under the threshold, the system becomes more sensitive for a few seconds, so repeating the phrase triggers it without extra effort. That is why saying a wake word a second time so often works. Questions: - Q: Does dictation software need a wake word? A: On a computer, no. A key you hold or tap is faster and never misfires. Wake words exist for devices without a keyboard. Hands-free dictation at a desk is better done with voice activity detection, which starts on any speech once you have armed it. - Q: What is the difference between a wake word and a hotkey? A: A hotkey is a physical key you press to start listening. A wake word is a phrase the device is always listening for. The hotkey is exact and private; the wake word is convenient when your hands are not on a keyboard. - Q: Can I say just "Siri" instead of "Hey Siri" on a Mac? A: On supported Macs, languages, and regions, yes. Open System Settings, open the Siri settings, and choose either "Hey Siri" or "Siri" as the phrase it listens for. - Q: Does "Hey Siri" work when my Mac is locked? A: Only if you allow it. Apple says that when "Siri" or "Hey Siri" is on and you turn on "Allow Siri when locked", you can use Siri even if the Mac is locked or in sleep. - Q: What is a false accept in wake word detection? A: A false accept is the device waking up when nobody said the phrase. It is counted as false activations per hour, and it is balanced against the false-reject rate, which is the share of real attempts that fail. In Flit: Flit has an optional wake word. Turn it on, and with voice activation armed a dictation starts only when you begin with "Hey Flit", or a phrase of your own. Apple's on-device recognizer checks for it, so nothing leaves your Mac. Off by default, and it needs macOS 26. (https://flit.fyi/#start) See it on flit.fyi: - Hold, tap, and hands-free: https://flit.fyi/#start - When the microphone is open: https://flit.fyi/privacy Related terms: voice-activity-detection, push-to-talk ### Accessibility permission (Accessibility access) URL: https://flit.fyi/glossary/accessibility-permission Accessibility permission is the macOS setting that lets one app read and control the interface of other apps. Dictation tools ask for it because it is how they type text at your cursor in whatever app you are using. macOS exposes every window, text field, and button through an accessibility layer that was built for screen readers and assistive devices. The same layer is the only sanctioned way for a third-party app to insert text into another app, which is why every dictation tool that "works everywhere" asks for it. Without it, a tool can only put text on the clipboard and leave the pasting to you. The permission is broad by design: an app that has it can, in principle, read what is on your screen. That is why the prompt is worth taking seriously and why the answer to "what does this tool actually read?" belongs in its privacy statement. #### What a dictation tool does with it Three things, in most tools. It inserts the finished text at the cursor, which is the whole point. It may read the focused text field, so replies can match the thread you are in and pick up the names already on screen. And it identifies the frontmost app, so a per-app style rule can apply. The second of those is the one to ask about: reading screen text is useful for tone, and it is also the feature where cloud tools have been caught sending screenshots to their servers. #### How macOS asks The first time a tool tries to type into another app, macOS shows a prompt and sends you to System Settings, Privacy and Security, Accessibility, where you flip a switch for that app. The permission survives updates but is revoked if the app’s code signature changes, which is why an unsigned or self-built tool keeps asking. You can withdraw it any time from the same switch. #### Declining it A tool built with care keeps working without it, in a reduced form: the transcript goes to the clipboard and you paste it. That is slower, but it is a legitimate choice for a locked-down machine or a first trial, and it should be a mode rather than an error. #### How to turn Accessibility access on or off for an app When an app first tries to control your Mac through accessibility features, macOS shows an alert with an Open System Settings button. You can also get to the same list yourself at any time, to grant access or to take it back. 1. Choose Apple menu → System Settings. 2. Click Privacy & Security in the sidebar. 3. Click Accessibility. You may need to scroll down to find it. 4. Find the app in the list. 5. Turn its switch on to grant access, or off to remove it. 6. Go back to the app and try the action again. #### Why is the app missing from the Accessibility list? An app normally appears in the list after it has asked for access once. If you dismissed the alert, or the app never showed one, the list will not include it. You can add the app by hand. 1. Open System Settings → Privacy & Security → Accessibility. 2. Click the add button at the bottom of the list. 3. Search for the app. 4. Select the app and click Open. 5. Confirm that the app now appears in the list with its switch turned on. #### Accessibility vs Microphone permission A dictation tool usually asks for two permissions, and they cover different things. Microphone lets the app hear you. Accessibility lets the app act on other apps, which is how the text gets typed at your cursor. Granting one does not grant the other. They also live in different lists. Microphone is at System Settings → Privacy & Security → Microphone, with one switch per app. Accessibility is at System Settings → Privacy & Security → Accessibility. If a tool hears you but nothing appears at the cursor, check the Accessibility switch. If it types nothing because it heard nothing, check the Microphone switch. Apple’s advice for the Accessibility list is direct: be cautious and grant access only to apps that you know and trust. Apple also points out that what a third-party app does with your information is governed by that app’s terms and privacy policy, not by Apple’s. Questions: - Q: Why does a dictation app need Accessibility access? A: Because Accessibility is the only sanctioned way on macOS for one app to type into another. Without it, a dictation tool can only copy the transcript to the clipboard for you to paste. - Q: Is granting Accessibility permission safe? A: It is broad: an app with it can read the interface of other apps. Grant it to tools you trust, and read the tool’s privacy statement to learn what it reads and whether any of it leaves the machine. A local-only tool with a published, verifiable statement is the safe case. - Q: Can I use dictation without it? A: With some tools, in a copy-only mode: the text goes to your clipboard and you paste it yourself. It is slower, but it works, and it is a fair way to trial a tool before trusting it. - Q: Where is Accessibility permission in System Settings on a Mac? A: Choose Apple menu → System Settings, click Privacy & Security in the sidebar, then click Accessibility. You may need to scroll down to see it. - Q: How do I remove Accessibility access from an app? A: Open System Settings → Privacy & Security → Accessibility and turn off the switch next to the app. You can turn it back on from the same place. - Q: Does a dictation app need Microphone permission as well as Accessibility? A: Yes, they are separate. Microphone access, at System Settings → Privacy & Security → Microphone, lets the app hear you, and Accessibility access lets it type the result into other apps. In Flit: Flit asks for Accessibility to type at your cursor and to read the focused field for tone, locally and in memory only. Decline it and Flit runs in copy-only mode. (https://flit.fyi/download#requirements) See it on flit.fyi: - The two permissions, explained before macOS asks: https://flit.fyi/download#requirements - What Flit reads and what it never stores: https://flit.fyi/privacy - Types into every app on your Mac: https://flit.fyi/#apps Related terms: dictation, on-device-processing, tone-matching ### Speech-to-text (STT) URL: https://flit.fyi/glossary/speech-to-text Speech-to-text (STT) is the technology that converts spoken audio into written words. It is the umbrella term covering live dictation, file transcription, captioning, and voice commands. The conversion is performed by a speech recognition model, which may run in the cloud (audio is uploaded to a server) or on-device (audio never leaves the machine). The choice determines both privacy properties and offline behavior. Raw STT output is a transcript: lowercase-ish, lightly punctuated, with fillers and false starts intact. Products built on STT differ mostly in what they do after recognition, from nothing at all to a full formatting pass. #### The pipeline, in four stages Audio capture turns sound into a stream of samples. Feature extraction converts those samples into a representation the model can read, usually a spectrogram. The recognition model maps that representation to text, sometimes through an intermediate step of phonemes or sub-word tokens. Post-processing turns the model’s output into something a person would write: punctuation, capitalization, numbers as digits, and in the better tools, filler removal and tone. When people say a tool "uses AI", they almost always mean the third stage. The fourth is where most of the daily-use quality lives, and it is the stage the marketing tends to skip. #### Cloud and on-device Cloud STT sends audio to a server, runs a large model there, and returns text. It can use models too big for a laptop and it works on any device with a network. On-device STT runs the model on your own hardware, so audio never leaves the machine, it works offline, and there is no round trip. On Apple silicon the gap in accuracy has narrowed to the point where the choice is about privacy and cost far more than quality. #### What the words mean on a product page "Speech-to-text" on a spec sheet usually means the raw model. "Dictation" means the whole experience, including where the text lands. "Transcription" means files. "Streaming" means words appear as you speak rather than after. "Local" or "on-device" means the audio stays home. Knowing which stage each word describes is most of what you need to read a comparison honestly. #### How to turn on speech-to-text on a Mac Every Mac has speech-to-text built in. Apple calls it Dictation, and it types into any text field where the cursor is. You can set it up in about a minute. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and turn it on. 3. Click the pop-up menu next to Microphone source and choose the microphone you want. Automatic lets the Mac listen to the device you are most likely to use. 4. Click the Shortcut pop-up menu and choose a shortcut to start Dictation. Choose Customize to make one that is not in the list. 5. Click in any text field, press the shortcut, and speak. 6. Stop when you are done. Dictation also stops automatically when no speech is detected for 30 seconds. #### Is speech-to-text accurate? What the published numbers show It depends on the audio more than most people expect. NVIDIA publishes results for its open Parakeet TDT 0.6B v3 model on eight English test sets, and the same model scores a word error rate of 1.93% on clean read speech (LibriSpeech test-clean) and 11.31% on meeting recordings (AMI). Noise moves the number too. On the same model card, the average word error rate rises from 6.34% on clean audio to 11.66% when noise is as loud as the speech (a signal-to-noise ratio of 0). A quiet room and a close microphone are the cheapest accuracy upgrade you can make. - Kind of speech: read sentences are easier than a meeting with people talking over each other. - Background noise and music: error rates climb as the noise gets closer to the level of your voice. - Accent and dialect: the Whisper model card reports different performance across accents and dialects. - Language: the Whisper model card reports lower accuracy on languages with less training data. - Vocabulary: the Parakeet model card warns that a word outside the model’s vocabulary is not likely to be recognized. #### How to check whether your speech stays on your Mac Speech-to-text is either processed on your device or sent to a server, and you can usually find out which. For Apple’s Dictation the answer is written in Settings. For any other tool, the permission list tells you which apps can hear the microphone at all. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and read the text under the Dictation switch. It says whether your voice input and transcripts are processed on your device. 3. Go to System Settings → Privacy & Security → Microphone to see every app that has asked to use the microphone. 4. Turn off any app in that list that you do not want listening. 5. For a third-party tool, read its privacy policy for where audio is processed. The Microphone list shows who can listen, not where the audio goes. Questions: - Q: What is the difference between speech-to-text and dictation? A: Speech-to-text is the underlying technology: audio in, words out. Dictation is a product built on it that inserts those words where you are writing, live, with formatting. Every dictation app uses speech-to-text; not every speech-to-text product is dictation. - Q: Which speech-to-text is most accurate? A: For English on a good microphone, the leading cloud engines and the best open models are close enough that your accent, room, and vocabulary matter more than the model. Accuracy is measured as word error rate on benchmark audio, and a per-user dictionary closes more of the real-world gap than switching models does. - Q: How do I turn on speech-to-text on a Mac? A: Choose Apple menu → System Settings, click Keyboard, go to Dictation, and turn it on. Then press the shortcut shown there in any text field and speak. - Q: Does Mac speech-to-text stop after a time limit? A: Apple says you can dictate text of any length without a timeout. Dictation does stop automatically when no speech is detected for 30 seconds. - Q: Why is speech-to-text less accurate in a noisy room? A: Noise hides the parts of the sound the model relies on. On NVIDIA’s published results for Parakeet TDT 0.6B v3, the average word error rate goes from 6.34% on clean audio to 11.66% when the noise is as loud as the speech. In Flit: Flit runs the whole pipeline on your Mac: capture, recognition, formatting, and insertion, with nothing sent anywhere. (https://flit.fyi/features#privacy) See it on flit.fyi: - Every stage of the pipeline, as Flit does it: https://flit.fyi/features - Speech-to-text for developers, compared: https://flit.fyi/blog/speech-to-text-for-developers - Twenty tools compared, cell by cell: https://flit.fyi/alternatives Related terms: automatic-speech-recognition, dictation, on-device-processing, formatting-pass ### Automatic speech recognition (ASR) URL: https://flit.fyi/glossary/automatic-speech-recognition Automatic speech recognition (ASR) is the machine-learning field, and the class of models, that turns audio of human speech into text. ASR models are the engine inside every dictation and transcription product. Modern ASR models are neural networks trained on large amounts of recorded speech. Well-known families include Whisper (OpenAI, open source), Parakeet (NVIDIA, open source), and the proprietary engines inside cloud services from Google, Apple, and others. ASR quality is usually reported as word error rate (WER) on benchmark audio. Real-world quality also depends on your microphone, room, accent, and vocabulary, which is why per-user dictionaries and domain vocabularies matter as much as the headline model. #### How the models are built An ASR model is trained on pairs of audio and the correct transcript, hundreds of thousands of hours of them, until it can predict the text for audio it has never heard. Older systems separated the acoustic model (sounds to phonemes) from the language model (phonemes to likely words). Current systems are end-to-end: one network maps audio directly to text tokens, which is simpler, more accurate, and easier to run on consumer hardware. Two architectures dominate today. Encoder-decoder models, like Whisper, read a whole chunk of audio and then write the transcript, which makes them accurate and batch-oriented. Transducer models, like Parakeet, emit tokens as audio arrives, which makes them streamable. #### What decides accuracy in practice Model size and training data set the ceiling. Everything below it is yours: microphone quality, distance from it, room echo, background noise, accent, and vocabulary. Names, jargon, and product terms are where every model struggles, because they are rare in training data, which is why a personal dictionary and a context-aware correction step tend to improve day-to-day accuracy more than switching to a bigger model. #### Open models changed the category Until 2022, competitive ASR lived behind cloud APIs. Whisper’s open release, followed by Parakeet and others, put state-of-the-art recognition in anyone’s hands, and Apple silicon gave laptops the compute to run it. That is the reason private, local dictation went from a compromise to a default in a few years. #### How to compare two ASR models on your own voice Benchmark scores describe benchmark audio. The only way to know how a model handles your voice, your room, and your vocabulary is to run it on a recording of you. The open Whisper command-line tool makes this a short job in Terminal. It needs Python and ffmpeg installed first. 1. Record one or two minutes of yourself speaking the way you normally dictate, including a few names and technical terms, and save it as an audio file. 2. Install Whisper in Terminal with: pip install -U openai-whisper 3. Transcribe the file with a small model: whisper myclip.m4a --model small --output_format txt 4. Rename the text file it wrote, then run the same command with --model turbo. 5. Read the two transcripts against what you actually said and count the wrong, missing, and extra words in each. 6. Divide the error count by the number of words you spoke. That is your personal word error rate for each model. #### What an ASR model returns besides words Older recognizers returned a flat run of lowercase words. Current open models return more, and knowing what is available tells you what a product built on them can do without extra work. - Punctuation and capitalization. Parakeet TDT 0.6B lists both as built-in features. - Timestamps. Parakeet returns word-level and segment-level timestamps, and Whisper has a word_timestamps option. - Subtitle files. The Whisper command-line tool can write txt, vtt, srt, tsv, and json output. - The language. Whisper and Parakeet TDT 0.6B v3 both detect the spoken language when you do not specify it. - A translation. Whisper’s multilingual models can translate speech in other languages into English text. #### Is ASR accurate? Why one model has many scores There is no single accuracy number for an ASR model. NVIDIA reports an average word error rate of 6.34% for Parakeet TDT 0.6B v3 across the eight English test sets of the Hugging Face Open ASR Leaderboard, but the individual sets range from 1.93% on LibriSpeech test-clean to 11.42% on Earnings-22, which is recordings of company earnings calls. Language matters as much as audio type. The Whisper model card says the models show strong results in about 10 languages, and that accuracy is lower on languages with less training data. When you read a score, check which test set and which language it was measured on before you compare it with another. Questions: - Q: What is the best ASR model for dictation? A: For live dictation, a streaming-capable model such as Parakeet, because words appear as you speak. For the widest language coverage, a model trained on many languages, such as Whisper with 99 or Meta’s Omnilingual ASR with more than 1,600. Most good tools use more than one and pick by language and job. On Apple silicon all of them run locally. - Q: Is ASR the same as speech-to-text? A: ASR is the technical name for the models and the research field. Speech-to-text is the plain-language name for what they do. On a product page the two are interchangeable. - Q: Why does ASR get my names wrong? A: Because a model has only seen a name as often as it appeared in training audio, which for most names and product terms is rarely. It writes the most probable spelling of the sounds. A dictionary of your own terms, and correction against names it can see on screen or in your project, fix this better than any model swap. - Q: What does ASR stand for? A: ASR stands for automatic speech recognition. It is the technology that turns recorded or live speech into text. - Q: Can I run an ASR model on my own computer? A: Yes. Open models such as Whisper and Parakeet can be downloaded and run locally. Whisper installs with pip and runs from Terminal on a Mac. - Q: Do ASR models add punctuation? A: Many current models do. NVIDIA lists automatic punctuation and capitalization as features of Parakeet TDT 0.6B, and Whisper also writes punctuated text. In Flit: Flit runs more than one speech engine, all of them on your Mac, and switches between them by language. (https://flit.fyi/privacy) See it on flit.fyi: - Which languages stream and which do not: https://flit.fyi/#languages - The dictionary that fixes what the engine mangles: https://flit.fyi/features#dictionary - Where the models run: https://flit.fyi/privacy Related terms: whisper, parakeet, word-error-rate, hallucination, speech-to-text ### Whisper URL: https://flit.fyi/glossary/whisper Whisper is a family of open-source speech recognition models released by OpenAI in 2022, trained on 680,000 hours of audio and covering about 99 languages. Because the weights are freely available, Whisper became the engine inside many local transcription and dictation tools. Whisper ships in sizes from tiny to large; bigger models are more accurate and slower. It transcribes in batch rather than streaming, so tools that show words live typically pair it with a streaming model and use Whisper for languages the streaming model does not cover. On Apple silicon, optimized runtimes let even the larger Whisper models run comfortably on-device, which is what made the current generation of private, local Mac dictation tools possible. #### The sizes, and which one you want Tiny and base are fast and rough, fine for a voice memo. Small is the usable floor for dictation. Medium and large are accurate enough for professional work and, on a modern Mac, still fast. Large-v3-turbo is the current sweet spot: a large model with a slimmed-down decoder that keeps most of the accuracy at several times the speed. Most Mac tools that offer a choice are offering these, and the right answer for dictation is the largest model that returns text before you get impatient. #### Batch, not streaming Whisper reads a window of audio and then writes the transcript for it. That makes it accurate, because it sees the whole phrase before deciding, and it makes it batch: nothing appears until you stop. Tools built only on Whisper feel like a pause after every sentence. Tools that stream pair it with a transducer model for live words and keep Whisper for the languages that model cannot handle. #### Known weaknesses Whisper hallucinates on silence and on very short clips, producing plausible phrases nobody said, and it can repeat itself on long audio. Both are well documented and well mitigated: trim silence before the model runs, keep chunks short, and treat any output on non-speech audio as suspect. Numbers and proper nouns follow the usual ASR pattern of being most-probable spellings rather than yours. #### How to run Whisper on a Mac from Terminal OpenAI publishes Whisper as a Python package with a command-line tool. The project says the code is expected to work with Python 3.8 to 3.11, and it needs the ffmpeg tool to read audio files. These steps assume you already have Python and Homebrew. 1. Open Terminal and install ffmpeg with: brew install ffmpeg 2. Install Whisper with: pip install -U openai-whisper 3. Transcribe a file with: whisper audio.mp3 --model turbo 4. Wait while the model downloads on the first run. It is saved in the .cache/whisper folder in your home folder and reused after that. 5. Find the transcript in the current folder. By default Whisper writes txt, vtt, srt, tsv, and json files. 6. Add --language followed by a language name, for example --language Japanese, if you want to skip automatic detection. 7. Run whisper --help to see every option. #### Whisper turbo vs large: the published figures OpenAI’s table lists the large model at 1550 million parameters, about 10 GB of video memory, and a relative speed of 1x. The turbo model is 809 million parameters, about 6 GB, and about 8x. OpenAI describes turbo as an optimized version of large-v3 with a minimal degradation in accuracy, and it is the default model in the command-line tool. Two caveats come with those figures. The speeds were measured on an NVIDIA A100 graphics card transcribing English, and OpenAI says real-world speed may vary significantly with the language, the speaking speed, and the hardware. And turbo is not trained for translation, so it returns the original language even when you ask it to translate. - tiny: 39 million parameters, about 1 GB, about 10x. - base: 74 million parameters, about 1 GB, about 7x. - small: 244 million parameters, about 2 GB, about 4x. - medium: 769 million parameters, about 5 GB, about 2x. - large: 1550 million parameters, about 10 GB, 1x. - turbo: 809 million parameters, about 6 GB, about 8x. #### Can Whisper translate? What it does with other languages Whisper was trained on two jobs: transcribing speech in its own language, and translating speech in other languages into English. Of its 680,000 training hours, about 438,000 are English audio with English transcripts, about 126,000 are non-English audio with English transcripts, and about 117,000 are non-English audio with a transcript in the same language, across 98 languages. That is why translation only goes one way. Add --task translate to get English text from non-English speech, and use the medium or large model for it, as OpenAI recommends. Whisper does not translate English speech into other languages. The tiny, base, small, and medium sizes also come in English-only versions with names ending in .en. OpenAI says these tend to perform better for English, especially tiny.en and base.en, and that the difference becomes less significant for small.en and medium.en. Questions: - Q: Is Whisper free to use? A: Yes. The models are open source under an MIT license and run on your own hardware. Building a usable dictation tool around them is the work, which is what most local Mac dictation apps charge for. - Q: Can Whisper run on a Mac without the cloud? A: Yes, and well. Optimized runtimes for Apple silicon run large models locally at usable speeds. Every local Mac dictation and transcription tool on the market uses Whisper or a model like it this way. - Q: Does Whisper stream text as I speak? A: Not natively. It transcribes a chunk after the chunk ends. Tools that show live words use a streaming model alongside it and reserve Whisper for languages the streaming model does not cover. - Q: How much memory does Whisper need? A: OpenAI lists about 1 GB of video memory for the tiny and base models, about 2 GB for small, about 5 GB for medium, about 6 GB for turbo, and about 10 GB for large. - Q: What is the difference between Whisper base and base.en? A: base.en is trained for English only, and base is multilingual. OpenAI says the .en models tend to perform better for English, especially at the tiny and base sizes. - Q: Can Whisper translate English into other languages? A: No. Whisper’s translate task turns speech in other languages into English text. The turbo model is not trained for translation at all, so use medium or large for that task. In Flit: Flit runs its speech and formatting models on your Mac. Which models are in the stack changes with releases; that they never leave the machine does not. (https://flit.fyi/#languages) See it on flit.fyi: - More than 1,600 languages, 30 of them streaming: https://flit.fyi/#languages - Building it yourself on Whisper, compared: https://flit.fyi/alternatives/whisper-diy - Flit vs MacWhisper: https://flit.fyi/alternatives/macwhisper - The private dictation apps, most of them on Whisper: https://flit.fyi/blog/best-private-dictation-apps Related terms: automatic-speech-recognition, parakeet, hallucination, on-device-processing, streaming-transcription ### Parakeet URL: https://flit.fyi/glossary/parakeet Parakeet is a family of open-source speech recognition models from NVIDIA, built on a transducer architecture, which can emit words as audio arrives. That makes it a common engine inside live dictation tools, where Whisper is built to read a whole chunk first. Parakeet models are trained for the same job as Whisper but with a different shape: instead of reading a whole chunk before writing, a transducer produces tokens continuously, so text can appear while you are still speaking. On Apple silicon the smaller Parakeet variants run in real time with room to spare, which makes them a natural fit for dictation where the words should land at key-up. The trade is coverage. Parakeet models are trained on a smaller set of languages than Whisper, so tools that use one for streaming pair it with a batch model for everything else. #### Transducer versus encoder-decoder An encoder-decoder model like Whisper is a reader and a writer: the encoder digests a window of audio, then the decoder writes the transcript. A transducer joins the two so that every new slice of audio can produce a new token without waiting for the window to close. That is what streaming needs, and it is also why transducers tend to be more stable on long audio: they never get to "reconsider" a whole passage and repeat it. #### Where it fits in a dictation tool Parakeet is the engine you want running while the key is held. It gives the live transcript in the notch or the pill, confirms you were heard, and leaves almost nothing to do at key-up. The final text quality still depends on the formatting stage that follows, and on a dictionary for names, exactly as with any ASR model. #### Parakeet v2 vs v3: which version does what The Parakeet models most people mean today are the two TDT 0.6B releases. Both have 600 million parameters, both add punctuation and capitalization, both return timestamps, and both are released under the CC BY 4.0 license. The difference is language. Version 2 is English only and was trained on about 120,000 hours of English speech. Version 3 extends it to 25 European languages and detects which one is being spoken without being told. NVIDIA reports an average word error rate on the Hugging Face Open ASR Leaderboard of 6.05% for v2 and 6.34% for v3. - Western and northern Europe: Danish, Dutch, English, Finnish, French, German, Swedish. - Southern Europe: Greek, Italian, Maltese, Portuguese, Spanish. - Central Europe: Croatian, Czech, Hungarian, Polish, Slovak, Slovenian. - Baltic: Estonian, Latvian, Lithuanian. - Eastern Europe: Bulgarian, Romanian, Russian, Ukrainian. #### How to run Parakeet on your own audio NVIDIA publishes Parakeet through its NeMo toolkit for Python. The model card lists Linux as the supported operating system and says the model is optimized for NVIDIA graphics cards, so treat a run on a Mac as an experiment. It needs at least 2 GB of memory to load, and more memory lets it take longer audio. 1. Install a recent version of PyTorch, then install the toolkit with: pip install -U nemo_toolkit['asr'] 2. Convert your recording to the format the model expects, 16 kHz mono in wav or flac. With ffmpeg: ffmpeg -i input.m4a -ar 16000 -ac 1 audio.wav 3. Start Python and run: import nemo.collections.asr as nemo_asr 4. Load the model with: asr_model = nemo_asr.models.ASRModel.from_pretrained(model_name="nvidia/parakeet-tdt-0.6b-v3") 5. Transcribe with: output = asr_model.transcribe(['audio.wav']) 6. Print the text with: print(output[0].text) 7. Add timestamps=True to the transcribe call if you want word and segment timestamps. #### Parakeet vs Whisper, by the published facts Both are open speech recognition models you can download, and both write punctuated text. They differ in size, coverage, license, and what they were designed to do. NVIDIA describes Parakeet TDT 0.6B v3 as designed for high-throughput transcription, and its model card also documents a streaming mode that processes audio in chunks of a few seconds. - Languages: Parakeet TDT 0.6B v3 covers 25 European languages. Whisper was trained on English plus 98 other languages. - Size: Parakeet TDT 0.6B has 600 million parameters. Whisper ranges from 39 million (tiny) to 1550 million (large). - License: Parakeet is CC BY 4.0, which requires attribution. Whisper is MIT. - Translation: Whisper can translate other languages into English. The Parakeet TDT cards describe transcription only. - Long audio: Parakeet handles up to 24 minutes in one pass, or up to 3 hours with local attention. Whisper moves a 30-second window across the file. - Input: Parakeet expects 16 kHz mono wav or flac. Whisper uses ffmpeg to read common audio formats. Questions: - Q: What is Parakeet in speech recognition? A: An open-source family of ASR models from NVIDIA built to stream: words come out as audio goes in. It is often paired with Whisper, which covers more languages but works in batch. - Q: Is Parakeet better than Whisper? A: For live dictation in a supported language, it feels better because text appears as you speak. For language coverage and for transcribing files, Whisper is broader. Good tools use each for what it is best at. - Q: What languages does Parakeet support? A: Parakeet TDT 0.6B v2 supports English only. Version 3 supports 25 European languages, including English, Spanish, French, German, Italian, Portuguese, Dutch, Polish, Russian, and Ukrainian, and detects the language automatically. - Q: Is Parakeet free for commercial use? A: Yes. NVIDIA releases Parakeet TDT 0.6B under the CC BY 4.0 license and states that the model is ready for commercial and non-commercial use. The license requires attribution. - Q: How long an audio file can Parakeet transcribe? A: NVIDIA says Parakeet TDT 0.6B v3 handles audio up to 24 minutes long with full attention on an A100 80GB graphics card, or up to 3 hours with local attention. In Flit: Flit runs its speech and formatting models on your Mac. Which models are in the stack changes with releases; that they never leave the machine does not. (https://flit.fyi/#languages) See it on flit.fyi: - The 30 streaming languages: https://flit.fyi/#languages - Why text is finished about 0.7 seconds after key-up: https://flit.fyi/#speed - Where the models run: https://flit.fyi/privacy Related terms: whisper, streaming-transcription, automatic-speech-recognition, neural-engine ### Neural Engine (ANE) URL: https://flit.fyi/glossary/neural-engine The Neural Engine is the part of an Apple silicon chip built to run machine-learning models efficiently, alongside the CPU and GPU. It is why a laptop can run speech recognition continuously without heat, fan noise, or a dead battery. Every M-series Mac has one. It is a set of cores specialized for the matrix arithmetic that neural networks are made of, and it runs that arithmetic at a fraction of the power a CPU or GPU would need. Speech recognition is the kind of steady, always-on workload the Neural Engine was designed for: a model that has to keep up with audio in real time, for as long as you are talking, without making the rest of the machine sluggish. Not every model fits. The Neural Engine has its own constraints on model shape and precision, and a model has to be converted for it. That engineering is one of the real differences between local dictation tools that feel effortless and ones that run the same model on the GPU and warm up your lap. #### Neural Engine, GPU, and CPU The CPU can run anything but is the slowest and hungriest at neural-network math. The GPU is fast and flexible and is where most local language models run, through frameworks like MLX or Metal. The Neural Engine is the most efficient of the three for models that fit its constraints, and speech recognition models fit well. A well-built local dictation tool puts each model where it runs best and leaves the CPU alone for the app you are typing into. #### Why it is a requirement Local dictation on an Intel Mac means running the model on a CPU that was not built for it. It works, slowly, and it is a big part of why on-device tools set Apple silicon as the floor. On an M1 or later the same models run in real time with the machine barely noticing, which is the difference between a tool you leave on all day and one you turn on for special occasions. #### How to check whether your Mac has a Neural Engine On the Mac, the Neural Engine arrived with the M1 chip. So the question is really whether your Mac has Apple silicon, and macOS tells you in one window. 1. Click the Apple menu in the corner of the screen. 2. Choose About This Mac. 3. Read the window that opens. A Mac with Apple silicon shows an item labeled Chip, followed by the name of the chip. 4. If you see an item labeled Processor followed by the name of an Intel processor, the Mac is Intel-based and has no Neural Engine. #### When did Macs get a Neural Engine? The Neural Engine started on the iPhone. Apple says the first generation shipped in the A11 chip in iPhone X in 2017. It came to iPad with the A12 chip and to the Mac with the M1 chip. Apple’s list of Mac computers with Apple silicon includes the MacBook Air with M1 from 2020 and every MacBook Air from 2022 or later, the 13-inch MacBook Pro with M1 from 2020 and every MacBook Pro from 2021 or later, Mac mini from 2020 or later, iMac from 2021 or later, Mac Studio from 2022 or later, and Mac Pro from 2023. Some Intel models were still sold in those years, so check About This Mac instead of going by the year alone. #### Why do some apps use the Neural Engine and others do not? An app does not talk to the Neural Engine directly. It hands a model to Core ML, Apple’s machine-learning framework, and Apple says Core ML blends the CPU, GPU, and Neural Engine into the most effective execution plan for the device. Parts of a model that do not suit the Neural Engine run elsewhere. How the model is written decides how much of it lands on the Neural Engine. In a published example, Apple engineers restructured a text model for the Neural Engine and reported it running up to 10 times faster with 14 times less peak memory on iPhone 13 than the unoptimized version. The same model can be quick in one app and slow in another for exactly this reason. Questions: - Q: What does the Neural Engine do for dictation? A: It runs the speech recognition model in real time at very low power, so a local dictation tool can transcribe continuously without heating the machine or draining the battery. That is why local dictation feels effortless on Apple silicon and sluggish on older Macs. - Q: Do I need Apple silicon for local dictation? A: For it to feel good, yes. The models can technically run on an Intel CPU, but slowly. Most on-device Mac dictation tools require an M1 or later for that reason. - Q: Does my Mac have a Neural Engine? A: It does if it has Apple silicon. Choose Apple menu → About This Mac: a line labeled Chip means Apple silicon, and a line labeled Processor with an Intel name means no Neural Engine. - Q: When was the Apple Neural Engine introduced? A: Apple says the first Neural Engine shipped in 2017 as part of the A11 chip in iPhone X. It reached the Mac with the M1 chip. - Q: Is the Neural Engine the same as the GPU? A: No. They are separate parts of an Apple silicon chip. Core ML can spread one model across the CPU, the GPU, and the Neural Engine. In Flit: Flit’s models are built for Apple silicon, which is why text lands in about 0.7 seconds and why an M1 or later is the requirement. (https://flit.fyi/download#requirements) See it on flit.fyi: - Will it run on your Mac?: https://flit.fyi/download#requirements - The speed that comes from running locally: https://flit.fyi/#speed Related terms: on-device-processing, parakeet, local-llm, latency ### Streaming transcription (real-time transcription) URL: https://flit.fyi/glossary/streaming-transcription Streaming transcription shows words on screen while you are still speaking, rather than after you stop. The alternative, batch transcription, processes the whole utterance at once and shows the text at the end. Streaming requires a model built to emit partial results and revise them as more audio arrives, which is a different engineering problem from transcribing a finished recording. Not every ASR model can stream; products often stream in some languages and fall back to batch in others. For dictation, streaming is mostly about confidence: seeing the words arrive confirms the tool heard you. The final text quality is decided by what happens at the end of the utterance either way. #### Why streaming makes dictation feel instant With batch recognition, the model starts working when you release the key and you wait for the whole recording to be processed. The pause is proportional to how long you spoke. With streaming, the model has been keeping up all along, so at key-up there is only the last fraction of a second of audio to finish and the formatting pass to run. The result is a delay that does not grow with the sentence, which is what people mean when they say a tool is fast. There is a second effect. Watching the live words tells you the tool is hearing you correctly, so you stop glancing at the result afterward to check. That trust is most of what makes dictation a habit rather than an experiment. #### Partial results and revisions A streaming model commits words tentatively and may change its mind as more audio arrives: "the meeting is" can become "the meeting isn’t" a moment later. Good tools show these revisions in a transient display, like a notch overlay or a floating pill, and only insert the settled text into your document. Inserting live partials into the app and then rewriting them is how a tool ends up fighting with your text editor. #### Why not every language streams Streaming needs a transducer-style model trained for the language. The families that stream well today cover a few dozen languages; batch models like Whisper cover about a hundred. A tool that claims 99 languages and live text is usually streaming in the languages its transducer supports and batch-processing the rest, and an honest product page says which is which. #### How to see streaming transcription on your Mac macOS has a built-in example of streaming transcription called Live Captions. It writes captions for audio from your apps, or for sound picked up by the microphone, while the audio is still playing. Apple says the audio is processed on device. Live Captions is available only on Mac computers with Apple silicon, and not in all languages, countries, or regions. 1. Choose Apple menu → System Settings, then click Accessibility in the sidebar. 2. Click Live Captions and turn it on. 3. Play a video or a podcast in any app. 4. Watch the captions window. The text appears while the speaker is still talking, which is streaming. 5. Turn Live Captions off in the same place when you are done. #### Can Whisper and Parakeet transcribe in real time? Not out of the box, in Whisper’s case. OpenAI’s model card says Whisper models cannot be used for real-time transcription out of the box, and that others may be able to build near-real-time applications on top of them. The reference code moves a 30-second window across a finished file. Parakeet is closer. NVIDIA’s model card for Parakeet TDT 0.6B v3 documents a streaming mode, and its example settings show the trade every streaming setup makes: audio is processed in 2-second chunks, with 10 seconds of earlier audio and 2 seconds of later audio as context. The later audio is look-ahead. More of it gives the model more to go on, and it also means the words appear later. #### Streaming or batch: which one do you need? Streaming matters when a person is waiting on the words. Batch is fine, and often simpler, when the audio already exists as a file. If you are choosing a tool, match the mode to the job. - Dictating into a text field: streaming, so you can see that you are being heard. - Captions for a live call or a talk: streaming, because a caption that arrives after the sentence is over is of little use. - Transcribing a recorded meeting, interview, or lecture: batch. Nobody is watching the words arrive. - Making subtitle files: batch. The Whisper command-line tool writes srt and vtt files directly. - A language your streaming model does not cover: batch, with a model that does cover it. Questions: - Q: What is the difference between streaming and batch transcription? A: Streaming shows words as you speak and finishes almost immediately when you stop. Batch waits until you stop, then processes the whole recording, so the pause grows with the length of what you said. Both can be accurate; streaming feels faster and shows you it is listening. - Q: Does streaming make the text less accurate? A: Not in the settled result. A streaming model may revise a word as more audio arrives, but what gets inserted is the final text. The accuracy you experience depends more on the model quality and your microphone than on whether it streamed. - Q: Why does my dictation tool stream in English but not in my other language? A: Because its streaming model does not support that language and it is falling back to a batch model. The text should still be right; it will just arrive after you stop rather than while you speak. - Q: Can Whisper do real-time transcription? A: Not out of the box. OpenAI’s model card says Whisper models cannot be used for real-time transcription as shipped, though others can build near-real-time applications on top of them. - Q: Does a Mac have built-in real-time transcription? A: Yes. Live Captions, under System Settings → Accessibility → Live Captions, shows captions for audio as it plays. It is available only on Mac computers with Apple silicon, and not in every language or region. - Q: Does real-time transcription need an internet connection? A: Not necessarily. Streaming describes when the words appear, not where the audio is processed. Apple says Live Captions on a Mac processes audio on device. In Flit: Flit streams live words in 30 languages and handles every other language with a local batch engine, and the site says which are which. (https://flit.fyi/#languages) See it on flit.fyi: - The 30 languages that stream: https://flit.fyi/#languages - Where the live words appear: the notch or a floating pill: https://flit.fyi/#start - Tools ranked partly on whether they stream: https://flit.fyi/blog/best-superwhisper-alternatives Related terms: parakeet, latency, automatic-speech-recognition, voice-activity-detection ### Language identification (LID) URL: https://flit.fyi/glossary/language-identification Language identification is the step where speech software works out which language is being spoken before, or while, it transcribes. It is what lets a multilingual tool switch languages without the user changing a setting. A recognizer that assumes the wrong language produces confident nonsense: French run through an English model comes out as English-shaped gibberish. Traditional dictation solved this with a setting you had to change by hand, which is fine for someone who works in one language and miserable for the many people who switch mid-conversation. Modern tools run a small detector on the first second or two of audio, pick the most likely language, and route the audio to the right model. The good ones can re-detect on each new dictation, so a Slack reply in German and the next one in English need no intervention. #### How detection works Some ASR models detect the language themselves as a side effect of transcribing; Whisper does this in its first pass. Others rely on a separate, tiny classifier that listens to the opening of an utterance and returns a language with a confidence score. Either way the decision has to be made early, because the wrong model has to be abandoned before it produces much, and it has to be made from very little audio, which is why the first word or two of a dictation is the moment where multilingual tools are most fragile. #### Detection is not translation Identifying the language and transcribing it in that language is the default and the safe behavior. Translation, where you speak one language and the text lands in another, is a separate and optional step, and it should be off unless you turn it on. A tool that translates by surprise has made a decision about your words that was not its to make. #### How to dictate in more than one language on a Mac Apple’s Dictation works with the languages you add to it, and you switch between them yourself while you dictate. Setting it up takes a minute. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation and click the Edit button next to Languages. 3. Select each language and region you want to dictate in. 4. Start Dictation in any text field. 5. To change language, click the language shown next to the cursor. It appears as an abbreviation or as characters that represent the language. 6. Or press the 🌐 key and choose the language you want to use. #### How do Whisper and Parakeet detect the language? Whisper was trained to identify the spoken language as one of its tasks, alongside transcription and translation. When you do not name a language, its transcription tool detects one using up to the first 30 seconds of the audio, then transcribes the whole file as that language. The English-only models skip detection and assume English. Parakeet TDT 0.6B v3 takes a narrower approach. NVIDIA says it automatically detects the language of the audio and transcribes it without additional prompting, across the 25 European languages it supports. A language outside that set is not supported. #### Why does it detect the wrong language, and how to fix it Detection is a guess made from a small amount of audio, and a few situations reliably trip it. Most of them have a simple fix, which is to tell the model the language yourself. - The recording changes language partway through. Whisper’s tool detects once, from the opening of the file, and applies that language to the rest. - The language has little training data. The Whisper model card reports lower accuracy on such languages. - The language is not one the model supports at all, so it can only answer with a language it does support. - The fix for Whisper: add --language and the language name to the command, for example --language German. - The fix for Apple’s Dictation: pick the language yourself by clicking the language next to the cursor. Questions: - Q: Can dictation software detect which language I am speaking? A: Modern tools can, from the first second or two of audio, and route the speech to the right model. Older tools need you to pick the language in a setting before you speak. - Q: What happens if it guesses the wrong language? A: You get confident nonsense in the wrong alphabet or spelling, because the model transcribes sounds as the words it knows. Tools that re-detect on every dictation recover on the next sentence; tools with a fixed setting need the setting changed. - Q: How do I change the dictation language on a Mac? A: Add the language under System Settings → Keyboard → Dictation by clicking Edit next to Languages. While dictating, click the language shown next to the cursor, or press the 🌐 key and choose the language. - Q: How do I stop Whisper from picking the wrong language? A: Name the language in the command, for example whisper audio.mp3 --language German. Whisper then skips automatic detection, which otherwise uses up to the first 30 seconds of the audio. - Q: Can a speech model detect the language automatically? A: Yes. Whisper detects the language when you do not specify one, and NVIDIA’s Parakeet TDT 0.6B v3 detects it automatically across its 25 supported European languages. In Flit: Flit identifies the language in the first seconds and picks the engine for it, so a bilingual conversation never means a trip to Settings. Translation is separate and off by default. (https://flit.fyi/#languages) See it on flit.fyi: - Languages and translation in Flit: https://flit.fyi/#languages - Flit for people who switch languages mid-thought: https://flit.fyi/use-cases/multilingual Related terms: streaming-transcription, whisper, automatic-speech-recognition ### Word error rate (WER) URL: https://flit.fyi/glossary/word-error-rate Word error rate (WER) is the standard measure of speech recognition accuracy: the number of words a model got wrong, divided by the number of words actually spoken. A WER of 5% means about one word in twenty was substituted, dropped, or invented. WER counts three kinds of mistake. A substitution is the wrong word in the right place ("their" for "there"). A deletion is a spoken word that never appeared. An insertion is a word that appeared and was never spoken. Add them up, divide by the number of words in the correct transcript, and you have the rate. It can exceed 100% on bad audio, because insertions are unbounded. The number is only meaningful next to the audio it was measured on. A model that scores 3% on clean, read English from a podcast microphone may score 15% on the same speaker in a café, and worse still on a language it saw less of in training. Published WER figures are almost always measured on standard benchmark sets, which are cleaner than your desk. #### What WER does not capture Every error counts the same, and they are not the same. A dropped "um" is an improvement; a dropped "not" reverses the meaning. A misspelled surname costs you a correction; a misspelled "the" costs you nothing because you never see it. Formatting is invisible to WER entirely: "1030 am" and "10:30 AM" are the same words. So a tool with a slightly worse raw WER and a much better formatting pass can produce text that needs far fewer fixes. That is why the number you actually care about is closer to "corrections per hundred words after formatting", which no benchmark reports and which you can only measure by dictating your own work for a week. #### How to read a vendor’s accuracy claim Ask which benchmark, which language, and which microphone. "99% accurate" with no dataset named is a marketing sentence. Ask whether the figure is before or after any correction step, and whether a personal dictionary was in play. And weigh it against the things WER cannot see: names, numbers, and whether the text arrives as writing or as a transcript. #### Improving your own error rate A better microphone, closer to your mouth, helps more than any model change. Finishing sentences before releasing the key gives the model context to disambiguate. And a personal dictionary for the names and terms you actually say fixes the errors that repeat, which are the ones that matter, because a mistake that happens once a month is a nuisance and one that happens every day is a reason to stop dictating. #### How to measure your own word error rate on a Mac You can measure WER on your own voice and your own microphone with a free open-source tool called jiwer. It computes the minimum edit distance between a correct transcript and what the recognizer produced, which is the same calculation benchmarks use. The result describes your desk, your microphone, and the passage you read. It is a small sample, so treat it as a rough reading and repeat it with different text before you draw conclusions. 1. Pick a passage of a few hundred words and save it as a plain text file named ref.txt, with one sentence per line. This is your reference. 2. Dictate the same passage with the tool you want to test, and save the output as hyp.txt, with the same sentences on the same lines. 3. Decide how you will treat capitals, punctuation, and numbers, then apply the same cleanup to both files. A simple choice is to lowercase everything and remove punctuation in both. 4. Open Terminal and install the tool with: pip install jiwer (it needs Python 3.8 or later). 5. Run: jiwer -r ref.txt -h hyp.txt to print the word error rate. 6. Add -a to the command to print the alignment of each sentence, so you can see which words were substituted, dropped, or inserted. 7. Add -c if you want the character error rate instead. #### Why do two WER numbers for the same model disagree? Usually because the text was cleaned up differently before scoring. WER is based on string edit distance, so it penalizes every difference between the two transcripts, including differences a person would call correct, such as a contraction written out in full or a number written as a digit. The Whisper paper describes this problem and answers it with a text normalizer that is applied to both transcripts before the errors are counted. The authors report that for several datasets this step alone lowered WER by up to 50 percent. A WER figure with no description of the normalization behind it cannot be compared with another one. - The Whisper normalizer removes the words hmm, mm, mhm, mmm, uh, and um before scoring. - It removes any phrase inside brackets or parentheses. - It expands contractions into their full form. - It rewrites spelled-out numbers and currency amounts with digits, so that "ten thousand dollars" becomes "$10000". - It converts British spellings to American spellings and strips most punctuation and symbols. #### WER vs CER: words or characters Character error rate (CER) is the same calculation done on characters instead of words. One wrong letter in a long word is one full error in WER and a small fraction of an error in CER, so CER is the gentler number and the two cannot be compared with each other. Model publishers pick the unit per language. The Whisper project, for example, reports either WER or CER for each language on the Common Voice 15 and Fleurs datasets. When you read an accuracy table, check which of the two a row uses before you compare it with another row. Questions: - Q: What is a good word error rate? A: On clean English benchmark audio, the best current models score under 5%. In real use, with a laptop microphone and ordinary speech, 5 to 10% is typical and still very usable when the tool has a good formatting pass and a dictionary for your names. - Q: How is word error rate calculated? A: Substitutions plus deletions plus insertions, divided by the number of words in the correct transcript. A model that changes one word, drops one, and adds one in a hundred-word passage has a WER of 3%. - Q: Is a lower WER always the better tool? A: No. WER ignores formatting, punctuation, and which words were wrong. A tool that writes times, prices, and names correctly and removes fillers can need fewer corrections than one with a lower raw WER and no formatting pass. - Q: How do I calculate WER myself? A: Save the correct text and the recognizer output as two text files, install the open-source jiwer tool with pip install jiwer, and run jiwer -r ref.txt -h hyp.txt in Terminal. Apply the same cleanup of capitals and punctuation to both files first, or formatting differences will be counted as errors. - Q: What is the difference between WER and CER? A: WER counts wrong words and CER counts wrong characters, each divided by the length of the correct transcript. CER is lower for the same output because a word with one wrong letter is mostly right at the character level. - Q: Does punctuation count in word error rate? A: It depends on the text normalization used. Most benchmark setups strip punctuation and capitals from both transcripts before scoring, as the Whisper paper’s normalizer does, so the published number says nothing about how well a model punctuates. In Flit: We do not publish a WER for Flit, because the benchmark number would not describe your desk. We publish the things you can check instead: delivered speed, and what the text looks like when it lands. (https://flit.fyi/#demo) See it on flit.fyi: - Before and after, on real dictations: https://flit.fyi/#demo - The dictionary that fixes repeat mistakes: https://flit.fyi/features#dictionary - How the speed claims were measured: https://flit.fyi/blog/dictation-vs-typing Related terms: automatic-speech-recognition, hallucination, personal-dictionary, formatting-pass ### Hallucination URL: https://flit.fyi/glossary/hallucination In speech recognition, a hallucination is text the model produces that nobody said: a plausible phrase invented from silence, noise, or a fragment too short to transcribe. It is the insertion error taken to its extreme. Every speech model is trained to produce text, so when given audio that contains no speech it still tries. Batch models like Whisper are the well-known case: fed a few seconds of room tone, they will sometimes emit "Thank you for watching" or a full sentence in another language, because those strings were common in training data at the ends of recordings. The output is fluent and confidently wrong. The same word is used for language models that make up facts, and the mechanism is related: a system optimized to produce likely output will produce it whether or not the input justifies it. In a dictation tool, a hallucination means text you never spoke lands in your document. #### When it happens Silence at the start or end of a recording, which is why key-up capture and voice activity detection matter. Very short utterances, a single word or a cough, where the model has too little to anchor on. Background speech from a video or a colleague, which is real speech but not yours. Long recordings, where some models lose their place and repeat a passage. Each has a known mitigation, and a tool that has not applied them will surprise you eventually. #### How good tools prevent it Run a voice activity detector first and never send non-speech to the recognizer. Trim silence at both ends. Discard results from clips below a minimum length. Watch for the model emitting text with no corresponding audio energy and drop it. On the formatting side, refuse any rewrite that adds a content word the transcript did not contain, so a formatting model cannot hallucinate either. A tool that does all of this will still occasionally mishear you; it should never invent a sentence. #### How to reduce Whisper hallucinations from the command line If you run the open-source Whisper command-line tool on recordings and it invents or repeats text, the tool has documented options aimed at exactly this. They do not remove the problem, but they address the two common forms: text invented over silence, and a passage repeated in a loop. The Whisper authors describe these checks as a workaround for noisy predictions, not a cure, so read the result against the audio after you change them. 1. Open Terminal and run the file once with default settings, for example: whisper recording.wav --model turbo. Note where the invented or repeated text appears. 2. If a passage repeats, run it again with --condition_on_previous_text False. The documentation says this can make text less consistent between windows, but the model becomes less prone to getting stuck in a failure loop. 3. If text appears over silence, add --word_timestamps True --hallucination_silence_threshold 2. With word timestamps on, this skips silent periods longer than the given number of seconds when a possible hallucination is detected. 4. Leave --no_speech_threshold at its default of 0.6 and --logprob_threshold at its default of -1.0 unless you have a reason to change them. Together they decide when a segment is treated as silent. 5. Trim long silences from the start and end of the recording in any audio editor and run it again, so the model has less non-speech to work with. 6. Compare the new transcript with the audio at the places you noted in the first step. #### Hallucination vs mishearing: how to tell them apart A mishearing is a wrong word in the place where you said a different word. It sounds like what you said, and you can see why the model chose it. A hallucination has no source in the audio at all. The Whisper paper describes the extreme case as a transcript entirely unrelated to the actual audio. The difference matters because the fixes are different. A mishearing is improved by a better microphone, clearer speech, or a dictionary entry. A hallucination is prevented by keeping silence and noise away from the model, or by rejecting output the model itself was unsure about. - Text that appears where you were silent or had stopped speaking is a hallucination. - The same phrase or sentence repeated several times in a row is a repeat loop, which the Whisper paper lists as a separate failure from hallucination. - A wrong word that sounds like the right one is a mishearing. - The first or last few words of a recording going missing is a third failure named in the Whisper paper, and it is a deletion, not a hallucination. #### How Whisper decides that its own output has failed The open-source Whisper code checks every 30-second window after decoding it. If the average log probability of the generated tokens is below -1.0, or the gzip compression ratio of the text is above 2.4, the window is treated as failed and decoded again at a higher temperature. Highly repetitive text compresses well, which is why the compression ratio catches loops. For silence, the paper notes that the probability of the no-speech token alone is not sufficient to detect a window with no speech. Combining a no-speech threshold of 0.6 with the log probability threshold of -1 makes the detection more reliable. The model card still lists hallucination and repetitive text as known limitations. Questions: - Q: Why did my dictation app write something I never said? A: Most likely the recognizer was given audio with no speech in it, such as silence at the end of a recording, or a very short clip, and produced a plausible phrase anyway. This is a known behavior of batch models and it is preventable with voice activity detection and silence trimming. - Q: Can the formatting step hallucinate too? A: Yes, if it is a language model rewriting your transcript freely. The safeguard is a rule that the output may remove fillers and fix punctuation but may never add a content word you did not say, and that a rewrite which breaks that rule is thrown away. - Q: Why does Whisper repeat the same sentence over and over? A: The Whisper model card says its sequence-to-sequence architecture makes it prone to generating repetitive text. In the open-source command-line tool, setting --condition_on_previous_text False makes the model less prone to getting stuck in such a loop. - Q: Does a bigger Whisper model stop hallucinations? A: Not reliably. The Whisper paper says larger models made steady progress on perception errors such as confusing similar-sounding words, but that repeat loops and hallucination are more stubborn and are a decoding problem. - Q: Is a hallucination the same as a transcription error? A: It is one kind of transcription error. Most errors are mishearings of something you did say, while a hallucination is text with no spoken source at all, and it counts as an insertion when word error rate is calculated. In Flit: Flit checks the model’s rewrite against the transcript: if a content word you said is missing from it, the rewrite is discarded and the plain transcript lands instead. Silence and non-speech never reach the recognizer in the first place. (https://flit.fyi/#exact) See it on flit.fyi: - What is decided by rule and what by model: https://flit.fyi/#exact - “Every word you said survives”: https://flit.fyi/features#writes Related terms: word-error-rate, whisper, voice-activity-detection, formatting-pass ### Latency (key-up latency) URL: https://flit.fyi/glossary/latency In dictation, latency is the time between the end of your speech and finished text appearing at the cursor. It is the number that decides whether a tool feels instant or feels like waiting. Measure it from the moment you release the key, because that is the moment you are waiting from. Everything that happens after key-up counts: finishing recognition of the last few hundred milliseconds of audio, the formatting pass, and the insertion into the app. A tool that streams recognition has only the tail to finish; a tool that batches has the whole utterance still to process, so its latency grows with every second you spoke. Under about a second, most people stop noticing. Around two or three seconds, they start glancing at the screen and then at the tool. Beyond that, dictation stops being a way to write and becomes a way to make a recording you then wait for. #### Where the time goes Recognition tail: how much audio remains to be processed at key-up, which streaming reduces to almost nothing. Formatting: a language-model pass over the transcript, typically the largest single cost in a local tool, and the reason a fast tool keeps that model small and gives it a strict time budget. Insertion: typing the result into the focused app through the accessibility layer, usually tens of milliseconds. Network, for cloud tools: a round trip plus queueing on a server you do not control, which is why cloud latency varies from fast to unusable depending on the hour. #### Latency versus quality is a real trade A formatting pass costs time, and skipping it is the fastest possible tool. It is also a tool that hands you a transcript to fix by hand, which costs more time than it saved. The right design offers the choice: a verbatim or lightly cleaned mode that lands almost immediately, and a fully formatted mode that takes a little longer and needs no fixing. What a tool should never do is let a slow model hold the text hostage; if the pass runs late, the plain transcript should land anyway. #### How to measure dictation latency on your Mac You do not need special tools. A screen recording that also captures your microphone puts your voice and the screen on the same timeline, so you can read the delay straight from the playback. Measure several dictations of different lengths. One reading tells you little, and the difference between a short and a long dictation tells you whether the wait grows with how long you speak. 1. Open the app you normally dictate into and place the cursor in a text field. 2. Press Shift-⌘-5 to open Screenshot, and choose to record the entire screen or a portion of it. 3. Click Options and choose your microphone, so the recording includes your voice. 4. Start the recording, then dictate one short sentence and one long one, leaving a few seconds between them. 5. Click the stop button in the menu bar. The recording is saved as a .mov file whose name starts with Screen Recording. 6. Play the file and note two moments for each dictation: when your voice stops, and when the finished text is on screen. The gap between them is the latency. 7. Compare the short and the long dictation. If the long one waits noticeably longer, the tool is processing in batch. #### Why a bigger model means a longer wait Model size is the most direct trade between accuracy and speed. The Whisper project publishes relative speeds for its own models: against the large model as 1x, tiny runs at about 10x, base at about 7x, small at about 4x, medium at about 2x, and turbo at about 8x. Those figures are relative to each other, and real speed depends on your hardware. Speed alone does not make a model live. Whisper reads audio through a sliding 30-second window, and its model card states that the models cannot be used for real-time transcription out of the box. A dictation tool built on such a model has to add its own streaming or chunking, and how well it does that decides the delay you feel. #### How to time a speech model in Terminal If you want to see the raw cost of recognition without any app around it, you can time the open-source Whisper command-line tool on a recording of your own voice. This measures the model on your Mac and nothing else: no formatting, no insertion, no network. The number is not the latency of any dictation app, because apps process audio while you speak. It is useful for comparing model sizes on your own hardware. 1. Install the audio dependency with: brew install ffmpeg 2. Install Whisper with: pip install -U openai-whisper 3. Record yourself speaking for about ten seconds and save the file, for example as clip.wav. 4. Run: time whisper clip.wav --model tiny and note the real time that Terminal prints. 5. Run: time whisper clip.wav --model turbo and note the time again. 6. Run each command a second time and compare those timings, so that one-time setup is not counted. Questions: - Q: What is a good latency for dictation? A: Under a second from key-up to finished text feels instant. One to two seconds is acceptable. Beyond that, the wait becomes the experience. Streaming recognition and a small, time-boxed formatting model are how local tools get under a second. - Q: Why does my dictation tool get slower the longer I talk? A: Because it recognizes in batch: nothing is processed until you stop, so a longer recording means a longer wait. Streaming tools process as you speak and keep the delay constant. - Q: Is cloud dictation slower than local? A: Often, and less predictably. A round trip to a server adds time on every dictation and varies with your connection and their load. Local tools have a fixed cost that depends only on your Mac. - Q: How do I test how fast my dictation app is? A: Record your screen with the microphone on using Shift-⌘-5, dictate a sentence, and play the recording back. The time between your voice stopping and the finished text appearing is the latency. - Q: Can Whisper transcribe in real time? A: Not by itself. The Whisper model card says the models cannot be used for real-time transcription out of the box, and the standard code processes audio in 30-second windows. Apps that feel live with Whisper have added their own streaming layer. - Q: Does a smaller speech model make dictation faster? A: Yes, at a cost in accuracy. The Whisper project lists its tiny model as about 10 times faster than its large model, and its turbo model as about 8 times faster. In Flit: Flit lands finished, formatted text about 0.7 seconds after key-up, or about 0.15 seconds with formatting set to Clean, and a late model never delays the transcript. (https://flit.fyi/#speed) See it on flit.fyi: - Speed, measured on real dictations: https://flit.fyi/#speed - What happens when the model runs late: https://flit.fyi/#exact - Delivered words per minute, including the wait: https://flit.fyi/blog/dictation-vs-typing Related terms: streaming-transcription, formatting-pass, neural-engine, push-to-talk ### Speaker diarization (speaker labels) URL: https://flit.fyi/glossary/speaker-diarization Speaker diarization is the process of working out who spoke when in a recording, so a transcript can be labeled by speaker. It answers "who said this?" rather than "what was said?". Diarization is a separate model from speech recognition. It listens for changes in voice, clusters the segments that sound like the same person, and assigns each a label, typically Speaker 1 and Speaker 2 until a name is attached. Combined with a transcript it produces the familiar meeting-notes layout, with each line attributed. It is a transcription feature, not a dictation one. When you dictate, there is one speaker and it is you, so a dictation tool has no use for diarization and should not be judged for lacking it. When you transcribe a meeting, an interview, or a podcast, it is the difference between a transcript and a wall of text. #### Why it is hard Two people with similar voices, one person whose voice changes with a cold, overlapping speech, and a laptop microphone that flattens everyone into the same room tone all defeat the clustering. Speaker counts are usually guessed rather than known. The best systems are good on clean, turn-taking audio like a podcast and much worse on a real meeting where three people talk over each other. Expect to fix labels by hand. #### Which tools have it Meeting transcribers like Otter.ai build their product around it. File transcribers such as MacWhisper offer speaker labels on recordings. Dictation tools generally do not, because the job never has two speakers. If diarization is on your list of requirements, you are shopping for a transcriber. #### How to add speaker labels to a recording on your Mac The open-source toolkit pyannote.audio runs speaker diarization locally on your computer. It tells you who spoke when, with start and end times for each turn. It does not produce the words, so you pair its output with a transcript from a speech recognizer. The steps below use the community pipeline, which is released under a CC-BY-4.0 license and needs a free Hugging Face account to download. 1. Make sure ffmpeg is installed, for example with: brew install ffmpeg 2. Install the toolkit in Terminal with: pip install pyannote.audio 3. Sign in to Hugging Face and accept the user conditions on the pyannote/speaker-diarization-community-1 page. 4. Create a Hugging Face access token at hf.co/settings/tokens. 5. In Python, load the pipeline with Pipeline.from_pretrained, passing the pipeline name and your token, then call it on your audio file. 6. Loop over the speaker_diarization field of the output to read each turn with its start time, end time, and speaker label. 7. If you know how many people spoke, pass num_speakers when you call the pipeline. If you only know a range, pass min_speakers and max_speakers. #### Can Whisper or Parakeet tell speakers apart? No. Both are speech recognition models, and neither is built to label speakers. The Whisper model card says the models have not been robustly evaluated on speaker classification or speaker diarization. NVIDIA’s model card for Parakeet TDT 0.6B v2 does not list speaker diarization among its capabilities. The Whisper paper records a related quirk. Early models tended to write plausible but almost always incorrect guesses for the names of speakers, because many training transcripts included the name of the person speaking. The authors removed the behavior by fine-tuning on transcripts without speaker annotations. If a transcript from a recognition model alone contains speaker names, do not trust them. #### How diarization accuracy is measured The standard metric is diarization error rate (DER), given as a percentage where lower is better. It adds up three kinds of mistake: false alarm, where speech is reported that was not there, missed detection, where real speech is not attributed to anyone, and speaker confusion, where speech is attributed to the wrong person. As with word error rate, the scoring rules change the number. Some evaluations forgive a short collar of time around each speaker change, and some skip overlapping speech. The pyannote benchmarks state that they use no forgiveness collar and do not skip overlapping speech, which is the stricter setup. Check these two choices before you compare DER figures from different sources. Questions: - Q: What does diarization mean in transcription? A: Labeling a transcript by who was speaking. A diarization model detects voice changes and groups segments by speaker, so a meeting transcript reads as a conversation rather than a single block. - Q: Does dictation software do speaker labeling? A: Almost never, because dictation has one speaker. Speaker labels are a feature of meeting and file transcription tools. If you need them, choose a transcriber; if you need live dictation, a dictation app. - Q: Does Whisper do speaker diarization? A: No. Whisper transcribes words, and its model card says it has not been robustly evaluated on speaker diarization. To get speaker labels you run a separate diarization model, such as the open-source pyannote.audio pipeline, and combine its timestamps with the transcript. - Q: What is diarization error rate? A: Diarization error rate (DER) is the standard accuracy measure for speaker labeling, where lower is better. It combines false alarms, missed speech, and speech attributed to the wrong speaker. - Q: Can speaker diarization run offline on a Mac? A: Yes. The open-source pyannote.audio community pipeline runs locally on your computer, and once it has been downloaded it can be used without an internet connection. In Flit: Flit does not diarize yet. A recording you transcribe in Flit comes back as one flow of text, and speaker labels are planned. For meetings today, use a transcriber. (https://flit.fyi/alternatives/otter-ai) See it on flit.fyi: - When a meeting transcriber is the right tool: https://flit.fyi/alternatives/otter-ai - Speaker labels on recorded files: https://flit.fyi/alternatives/macwhisper - Where Flit’s focus ends: https://flit.fyi/features#not-yet Related terms: transcription, speech-to-text, automatic-speech-recognition ### Disfluency (fillers) URL: https://flit.fyi/glossary/disfluency A disfluency is any break in fluent speech: fillers like "um" and "uh", repeated words, false starts, and self-corrections. Natural speech is full of them; finished writing has none. Removing disfluencies is the first and most visible part of turning a transcript into writing. The subtle part is telling filler from meaning: a repeated word can be a stumble or emphasis, and a false start can be noise or a deliberate correction ("three hundred, scratch that, four hundred dollars"). Tools handle this anywhere from not at all (raw transcripts) to rule-based filler removal to a full language-model cleanup pass, with the risk at the far end being rewording things you meant. #### The kinds of disfluency Filled pauses: um, uh, er, and their equivalents in every language. Discourse markers: like, you know, sort of, when they carry no meaning. Repetitions: "the the meeting", "I I think". False starts: a phrase abandoned and restarted. Self-corrections: a value replaced by another, signaled by "no wait", "scratch that", "I mean". Each needs different handling, and the last one is the interesting case, because the right output is not the removal of words but the survival of the correction. #### Rules, models, or both Filled pauses and repetitions are safe to remove by rule: "um" is never a word you meant. Discourse markers are trickier, because "like" is sometimes a verb. Self-corrections are the strongest case for a narrow rule: when both sides of "no wait" are the same kind of thing, a time or a number or a day, replacing the first with the second is certain, and a model is not needed. Everything looser belongs to a language model with a strict brief: remove what was not meant, never rewrite what was. #### The failure mode to watch for Overcleaning. A model told to make speech read well will, given the chance, tighten your sentences, swap your words for its own, and drop the hedge that was the whole point. The test is simple: dictate something with an unusual word choice and see whether it survives. A tool that removes your fillers and keeps your words is doing the job; one that improves your prose is doing a different job you did not ask for. #### How to dictate with fewer fillers The cleanest way to deal with fillers is to say fewer of them. A filler is usually a pause filled with sound, and the habit comes from conversation, where silence invites someone else to start talking. A dictation tool does not interrupt, so you can pause in silence instead. With Apple’s built-in Dictation you have plenty of room to do that. Apple states that Dictation stops automatically only when no speech is detected for 30 seconds. 1. Decide what the sentence is before you start the dictation, even if only roughly. 2. Speak one sentence at a time and stop at the end of it. 3. When you need to think, close your mouth and say nothing. A silent pause leaves nothing in the text. 4. Slow down slightly. Fillers appear when your mouth runs ahead of the next thought. 5. If you make a mistake, finish the sentence and correct it afterward instead of restarting it halfway. 6. Read back what you dictated and note which fillers you use most, so you can listen for them next time. #### Why is "um" not counted in accuracy benchmarks? Because many benchmark setups delete fillers from both transcripts before they count errors. The text normalizer published with the Whisper paper removes the words hmm, mm, mhm, mmm, uh, and um from the reference and from the model output before word error rate is calculated. The practical result is that a published accuracy figure tells you nothing about fillers. A model that writes every um and a model that writes none of them can receive the same score. Whether fillers reach your document is a property of the tool around the model, and you have to test it yourself. #### How to test how a dictation tool handles disfluencies A short test separates the tools that remove fillers by careful rule from the ones that remove too much or nothing at all. Dictate each line below as written, hesitations included, and compare the output with what you meant. 1. Dictate a sentence with filled pauses, such as: "the report is um almost uh finished". Both fillers should be gone and every other word should remain. 2. Dictate a stumble, such as: "I I think the the meeting moved". The doubled words should appear once. 3. Dictate a deliberate repeat, such as: "that was a very very long day". Both words should survive, because the repeat is emphasis. 4. Dictate "like" used as a real word, such as: "I like the second option". The word must not be removed. 5. Dictate a self-correction, such as: "send it Monday no wait Tuesday". Check whether you get the corrected value, both values, or the wrong one. 6. Dictate a hedge you mean, such as: "I think we should probably wait". If "I think" or "probably" is missing, the tool is rewriting, not cleaning. Questions: - Q: Does dictation software remove ums and uhs? A: The good ones do, automatically, along with repeated words and false starts. Raw transcribers leave them in. Built-in dictation on most systems leaves most of them in too. - Q: What happens if I correct myself while dictating? A: With a tool that handles self-corrections, saying "Monday, no wait, Tuesday" produces Tuesday. The safest implementations do this by rule when both values are the same kind of thing, and leave looser cases to the formatting model. - Q: Can filler removal change what I meant? A: Rule-based removal of ums and repeats cannot. A free-rewriting language model can, by tightening or rewording. Look for a tool that promises your content words survive, and test it with a sentence you would notice being changed. - Q: How do I stop saying um when I dictate? A: Replace the sound with silence. Decide the sentence before you start, speak one sentence at a time, and pause without speaking when you need to think. A silent pause adds nothing to the text. - Q: Do filler words count as errors in word error rate? A: Often they do not. The Whisper paper’s text normalizer, for example, removes hmm, mm, mhm, mmm, uh, and um from both transcripts before scoring, so fillers neither help nor hurt the published number. - Q: What is the difference between a filler and a false start? A: A filler is a sound or word that fills a pause, such as um or uh. A false start is a phrase you begin, abandon, and say again differently. Both are disfluencies, but a false start contains real words, so removing it takes more care. In Flit: Flit removes fillers and false starts, honors "scratch that" and "no wait" by rule, and throws away any rewrite that loses a word you said. (https://flit.fyi/#writes) See it on flit.fyi: - “the budget is uh three hundred dollars scratch that four hundred”: https://flit.fyi/#demo - Correcting yourself, and the phrases Flit listens for: https://flit.fyi/commands#corrections - Why delivered WPM is lower than speaking rate: https://flit.fyi/blog/dictation-vs-typing Related terms: formatting-pass, verbatim-mode, words-per-minute, dictation ### Formatting pass (post-processing) URL: https://flit.fyi/glossary/formatting-pass The formatting pass is the step after speech recognition that turns a raw transcript into text a person would have typed: punctuation, capitalization, numbers as digits, fillers removed, and a tone suited to where the text is going. It is where most of the difference between dictation tools lives. A recognizer produces words. It does not know that "ten thirty am" should be 10:30 AM, that a Slack message is lowercase and a client email is not, or that "bob at acme dot com" is an address. The formatting pass knows, or is supposed to. Early tools did it with rules; the current generation runs a small language model over the transcript; the best do both, with rules for anything that has one right answer and the model for judgment. It is also the step with the most ways to go wrong. A model asked to "clean this up" will reword, tighten, and occasionally invent, and a slow model turns a fast tool into a waiting room. How a product constrains its formatting pass tells you more about its quality than which recognizer it uses. #### What a good pass does Removes fillers and false starts. Punctuates and capitalizes. Writes numbers, times, dates, prices, units, emails, and URLs in their proper form. Applies spoken commands like "new paragraph". Inserts snippets and dictionary terms exactly as saved. Matches the tone of the destination app, casual or professional, and can be pinned per app. Leaves terminals and code editors untouched. Everything on that list is either a rule or a narrowly briefed model, and the order matters: rules first, so the model never gets to second-guess a clock time. #### Rules versus the model A language model is good at judgment and unreliable at precision. Given "ten thirty AM" it usually writes 10:30 AM, and "usually" is the problem, because you will not notice the time it wrote 10:30 PM until the meeting is missed. Anything with one correct answer, clock times, spoken punctuation, dictionary words, snippets, symbol names, should be handled by a deterministic rule that produces the same output every time and can be tested. The model gets what is left: sentence flow, paragraph breaks, which fillers were filler, and tone. #### Guardrails Three are worth insisting on. A content check: if the model’s output loses a word you said or adds one you did not, discard it and keep the rule-formatted transcript. A time budget: if the model has not answered within a fixed window, the transcript lands without it. And a verbatim escape: apps where any rewrite is a bug get no model at all. With those in place, a formatting pass can only improve the text or leave it alone; it can never break a dictation. #### How to format text yourself with Apple Dictation Apple’s built-in Dictation does a small part of this work automatically and leaves the rest to spoken commands. Knowing where that line sits tells you what a fuller formatting pass would be doing for you. The automatic part is auto-punctuation, which in supported languages inserts commas, periods, and question marks as you speak. Paragraphs, capitals beyond the defaults, and symbols are things you say out loud. 1. Choose Apple menu → System Settings, click Keyboard in the sidebar, and go to Dictation. 2. Turn Dictation on, and check that auto-punctuation is on if you want commas, periods, and question marks inserted for you. 3. Start dictating, and say "new line" or "new paragraph" where you want a break. 4. Say "numeral" before a number to format the next phrase as a number. 5. Say "caps on" to write the words that follow in title case and "caps off" to return to normal. Say "all caps" to capitalize the next word. 6. Say the name of a symbol to insert it, such as "at sign", "percent sign", "hashtag", or "dollar sign". 7. Read the result back. Anything you still had to fix by hand is work a formatting pass would have to do. #### Formatting pass vs inverse text normalization Inverse text normalization is the older, narrower name for part of this job: turning spoken forms into written ones, such as "ten thirty" into 10:30. The Whisper paper notes that many speech recognition systems include some level of it, and that it is often simple or rule-based and detectable from what it leaves unhandled, such as never including commas. Newer open models moved some of that work into recognition itself. Whisper was trained to predict the text of transcripts as written, which its authors say removes the need for a separate inverse text normalization step. NVIDIA describes Parakeet TDT 0.6B v2 as supporting punctuation and capitalization directly. A formatting pass is everything that still happens after that. Punctuation from the recognizer is a starting point. Removing fillers, applying your own terms, and matching the destination app are separate steps that a recognizer does not perform. #### How to add your own formatting rules in macOS macOS has a small rule-based formatting layer of its own, which is a useful picture of what a deterministic rule is: the same input always gives the same output. Text replacements swap one string for another, and smart quotes and dashes change punctuation as text is entered. Apple notes that text replacements, smart quotes and dashes, and smart links are not available in all apps. If text arrives with curly quotes or a substitution you did not expect, check these settings before you look anywhere else. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Click Text Replacements. 3. Click the Add button, type the text to replace and its replacement, then click Add. 4. To change quote and dash behavior, go to Text Input in Keyboard settings and click Edit. 5. Turn "Use smart quotes and dashes" on or off, choose quote styles if you want them, then click Done. 6. In an individual app, choose Edit → Substitutions to turn Text Replacement on or off for that app. Questions: - Q: What is post-processing in dictation software? A: Everything that happens to the transcript after the speech model produces it: punctuation, capitalization, number formatting, filler removal, spoken commands, and tone. It is the stage that turns a transcript into writing, and the stage where tools differ most. - Q: Does the formatting use AI? A: In most current tools, partly. A small language model handles judgment calls like tone and sentence flow. The better tools handle anything with one right answer, such as times and punctuation, by fixed rule instead, because a model is only usually right. - Q: Can I turn formatting off? A: A good tool lets you choose the level, from verbatim to fully polished, and switches to verbatim automatically in terminals and code editors. Less formatting also means faster text, which is a reasonable trade for a quick reply. - Q: Does Apple Dictation add punctuation automatically? A: Yes, in supported languages. Apple’s Dictation automatically inserts commas, periods, and question marks, and you can turn this off in System Settings under Keyboard, in the Dictation section. Other punctuation is entered by saying its name. - Q: What is inverse text normalization? A: It is the step that turns spoken forms into written forms, such as spoken numbers into digits. It is one part of a formatting pass, and some newer speech models produce punctuation and capitalization directly so that less of it is needed. - Q: How do I make dictation write numbers as digits? A: In Apple’s Dictation, say "numeral" before the number, which formats the next phrase as a number. Tools with a fuller formatting pass convert numbers, times, and prices without a command. In Flit: Flit’s pass runs on your Mac: fixed rules for everything exact, a local model for judgment, a content check on its output, and a time budget so a late model never holds the text. (https://flit.fyi/#exact) See it on flit.fyi: - Fixed rules on one side, the model on the other: https://flit.fyi/#exact - One sentence, three apps, three results: https://flit.fyi/#writes - Finished text, not a transcript: https://flit.fyi/features#writes - Before and after: https://flit.fyi/#demo Related terms: disfluency, tone-matching, spoken-punctuation, local-llm, verbatim-mode, latency ### Verbatim mode URL: https://flit.fyi/glossary/verbatim-mode Verbatim mode is a dictation setting where the output is exactly what you said, with no rewording, tone adjustment, or cleanup beyond basic recognition. It exists because some text must survive intact. The classic cases are shell commands, code, legal quotes, and anything where a helpful rewrite is a bug: "git log dash dash oneline" must become git log --oneline and nothing else. Dictation tools with a formatting layer need a way to switch it off, ideally automatically by context. The strongest implementations bind verbatim behavior to the application: terminals and editors always verbatim, prose apps formatted, without the user toggling anything. #### What verbatim still does Verbatim is not raw. Spoken punctuation and symbol commands still apply, because "dash dash force" is how you say --force, and without commands a terminal is unusable by voice. Dictionary terms and file names from your project still resolve, because a misheard file name is a recognition error, not a rewrite. What verbatim switches off is judgment: no capitalizing the first word, no smart quotes, no trailing period, no filler removal that might eat a flag, and no tone. #### Automatic by app, or a toggle A toggle you have to remember is a toggle you will forget, and the cost of forgetting is a broken command. Binding the mode to the frontmost app removes the decision: Terminal, iTerm, Warp, Xcode, VS Code, Cursor, and their kin are verbatim; Mail and Messages are not. The remaining edge is a prose field inside a code editor, like a commit message or a PR description, where some people want formatting back, and a per-app override is the right answer there. #### Testing a tool for it On day one of any trial, dictate a command with a double dash, a quoted string, and a path into a terminal, and read it back character by character. Then dictate the same thing into a notes app and confirm it was punctuated. A tool that passes both is treating context as the signal. A tool that capitalizes your command has no verbatim mode worth the name, whatever the settings screen says. #### How to dictate a command with Apple Dictation Apple’s built-in Dictation has no verbatim switch, but you can get close by turning off its one automatic behavior and speaking every symbol. This is slow, and it shows clearly what an automatic verbatim mode saves you. One detail matters for commands. In Apple’s command list, "hyphen" inserts the short hyphen character that flags use, while "dash" inserts a longer dash character that a shell will not accept as a flag. 1. Choose Apple menu → System Settings, click Keyboard in the sidebar, and go to Dictation. 2. Turn off auto-punctuation, so that no commas, periods, or question marks are inserted for you. 3. Click into your terminal or editor and start dictating. 4. Say "hyphen" for each hyphen in a flag. For a double hyphen, say it twice. 5. Say "no space on" before a run of text that must not contain spaces, such as a path, and "no space off" after it. 6. Say symbol names where you need them: "forward slash", "underscore", "vertical bar", "backslash", "asterisk". 7. Read the line back character by character before you press Return. #### Why do straight quotes turn into curly quotes? Because of a macOS text setting, not necessarily your dictation tool. macOS has a setting called "Use smart quotes and dashes" that swaps straight quotes for typographic ones as text is entered. That is correct in prose and wrong in a command or a string in code. The setting is at System Settings → Keyboard → Text Input. Click Edit, then turn off "Use smart quotes and dashes". Apple notes that smart quotes and dashes are not available in all apps, so the effect varies from one app to the next. Many apps also have an Edit → Substitutions menu where you can see which substitutions are active in that app. #### Verbatim mode vs Spelling mode in Voice Control Apple’s Voice Control has a mode for text that must be exact, and it works differently from verbatim mode. Verbatim mode still takes whole words and leaves them alone. Spelling mode takes one character at a time. Apple describes it as helpful when you need to enter a password, a web address, or another sequence of characters. Voice Control is at System Settings → Accessibility → Voice Control. With it on, say "Spelling mode" to switch, and "Dictation mode" to return to word-by-word dictation. Standard Dictation is not available while Voice Control is on. Use Spelling mode for a short exact string such as an identifier or an address. For a full command or a line of prose it is too slow, and word-level verbatim dictation is the better fit. Questions: - Q: What does verbatim mean in a dictation app? A: The output is what you said, character for character, with no rewording, capitalization, or punctuation added by a model. Spoken symbol commands still apply, because that is how you say a dash or a bracket. - Q: Why do dictation apps break shell commands? A: Because their formatting pass treats everything as prose: it capitalizes the first word, turns quotes into smart quotes, and adds a period. Each of those breaks a command. Verbatim mode, ideally automatic in terminals, is the fix. - Q: Can I dictate into a code editor? A: Yes, with a tool that goes verbatim in editors and knows the file names in your project. The prose around code, such as commit messages and comments, dictates well; dense syntax is usually still faster to type. - Q: How do I dictate a hyphen or double dash on a Mac? A: With Apple’s Dictation, say "hyphen" to insert a hyphen, and say it twice for a double hyphen. Do not say "dash", because in Apple’s command list that inserts a longer dash character. - Q: How do I stop my Mac from changing quotes to curly quotes? A: Go to System Settings → Keyboard → Text Input, click Edit, and turn off "Use smart quotes and dashes". The feature is not available in all apps, so some apps are unaffected either way. - Q: Can I turn off automatic punctuation in Mac dictation? A: Yes. Auto-punctuation for Apple’s Dictation can be turned off in System Settings under Keyboard, in the Dictation section. After that, punctuation appears only when you say its name. In Flit: Flit is always verbatim in terminals and editors, automatically, and never rewords a command. (https://flit.fyi/use-cases/developers) See it on flit.fyi: - Work with your agents, on the homepage: https://flit.fyi/#agents - The editors and terminals Flit recognizes: https://flit.fyi/features#code - The commands that still work in verbatim: https://flit.fyi/commands - How to dictate code, honestly: https://flit.fyi/blog/how-to-dictate-code Related terms: formatting-pass, spoken-punctuation, disfluency, personal-dictionary ### Spoken punctuation (voice commands) URL: https://flit.fyi/glossary/spoken-punctuation Spoken punctuation is saying the name of a mark or symbol, like "comma", "open paren", or "new paragraph", and having the dictation tool type the symbol instead of the word. The broader set of such phrases is usually called spoken commands. Speech has no punctuation, so a dictation tool has to get it from somewhere. For prose, a formatting pass can infer most of it from the rhythm of the sentence. For anything exact, a terminal, a spreadsheet formula, a bracketed aside, inference is not good enough, and the only reliable way to get a specific character is to ask for it by name. The set of commands a tool understands, and whether each one is resolved by a fixed rule or by a model guessing, is the difference between a tool you can use in a terminal and one you cannot. #### What a command set covers Punctuation: period, comma, question mark, colon. Symbols: dash, underscore, dollar sign, at sign, hash. Pairs: open and close paren, bracket, brace, and quotes, ideally closing themselves. Structure: new line, new paragraph, tab. Case: camel case, snake case, all caps, for the next word or phrase. Emoji by name. And an escape hatch for saying the word itself, "literally comma", when you mean the word and not the mark. #### Rules, not a model A command should map to exactly one output, every time, on every machine. That means a lookup, not an inference: "dash dash force" is --force by rule, and a model never sees it. Rule-based commands can be published in full, tested, and trusted in a terminal. Model-resolved commands are usually right, and "usually" is one broken flag away from a bad afternoon. #### When commands should be on Always, in terminals and code editors, where there is no other way to type a symbol by voice. Optionally, in prose, where a formatting pass punctuates for you and a spoken "comma" in an ordinary sentence would otherwise become a mark by accident. Commands are usually one language, most often English, because a half-understood command in the middle of a sentence is worse than none. #### How to dictate punctuation with Apple Dictation on a Mac The built-in Dictation feature on a Mac understands a published list of spoken commands. You can try them in any text field before you look at anything else. 1. Choose Apple menu → System Settings, click Keyboard in the sidebar, go to Dictation, and turn it on. 2. Click in a text field, then press the Microphone key if your keyboard has one, press your dictation shortcut, or choose Edit → Start Dictation. 3. Say a sentence and name the marks as you go, for example "thanks comma see you at noon period". 4. Say "new line" or "new paragraph" to break the text, and "open parenthesis" and "close parenthesis" to wrap an aside. 5. Say "all caps" before a word to capitalize all of it, or "numeral" before a number to get digits. 6. Press Escape, the Microphone key, or your shortcut again to stop. #### What Apple’s command list includes Apple publishes its dictation commands in the Mac User Guide under "Commands for dictating text on Mac". The list is grouped by type, and it is wider than most people expect. In some languages you can also insert emoji by name, such as "heart emoji". - Punctuation: comma, period, question mark, colon, semicolon, apostrophe, ellipsis, hyphen, dash, quote and end quote. - Pairs: open and close parenthesis, open and close square bracket, open and close brace, open and close angle bracket. - Typography: ampersand, asterisk, at sign, backslash, forward slash, caret, underscore, vertical bar, percent sign, hashtag or pound sign. - Formatting: new line, new paragraph, tab key, numeral, roman numeral, no space on and no space off. - Capitalization: caps on and caps off for title case, all caps for the next word, all caps on and all caps off for a phrase. - Math and currency: equal sign, plus sign, minus sign, greater than sign, less than sign, dollar sign, euro sign, yen sign. #### Auto-punctuation vs spoken punctuation There are two ways punctuation gets into dictated text. Auto-punctuation adds it for you from the shape of the sentence. Spoken punctuation adds only what you name. In supported languages, Apple Dictation inserts commas, periods, and question marks automatically, and you can turn that off under System Settings → Keyboard → Dictation. Some open speech models also punctuate on their own. The model card for NVIDIA Parakeet TDT 0.6B v2 lists automatic punctuation and capitalization as a feature of the model itself. That is useful for prose, and it is the reason a tool needs a separate rule layer when you want an exact character. A practical rule: leave auto-punctuation on for messages and email, and name the mark whenever the character matters. If the automatic marks keep landing in the wrong place for the way you speak, turn the feature off and say every mark yourself. Questions: - Q: How do I dictate punctuation? A: Say its name: "comma", "period", "question mark", "new paragraph". In prose, a good tool also infers most punctuation from the sentence, so you only need to say it when you want something specific. - Q: How do I say a symbol like a dash or a bracket? A: By name: "dash", "underscore", "open paren", "close bracket", "dollar sign". In a tool built for it, pairs close themselves and "dash dash" produces the double dash used in command flags. - Q: What if I want the word “comma” and not the mark? A: Say "literally comma", or the tool’s equivalent escape phrase. In prose apps where commands are optional, the plain word usually goes through as a word anyway. - Q: How do I turn off auto-punctuation on Mac? A: Choose Apple menu → System Settings → Keyboard → Dictation, then turn off Auto-punctuation. After that, Dictation only types the marks you say by name. - Q: How do I dictate in all caps on a Mac? A: With Apple Dictation, say "all caps" before a single word, or say "all caps on", speak the phrase, then say "all caps off". Saying "caps on" gives you title case instead. - Q: Can I dictate emoji on a Mac? A: Yes, in some languages. With Apple Dictation you say the name followed by the word emoji, for example "heart emoji" or "car emoji". In Flit: Flit ships more than 60 commands and 115 spoken emoji names, every one resolved by rule, published in full, always on in terminals and optional everywhere else. (https://flit.fyi/commands) See it on flit.fyi: - Every spoken command, in one place: https://flit.fyi/commands - Pairs and quotes that close themselves: https://flit.fyi/commands#pairs - Saying the word instead of the mark: https://flit.fyi/commands#escape - 60 commands, and not one of them is a guess: https://flit.fyi/features#commands Related terms: verbatim-mode, formatting-pass, text-expansion, disfluency ### Text expansion (snippets) URL: https://flit.fyi/glossary/text-expansion Text expansion replaces a short trigger with a longer saved text: an address, a signature, a canned reply. In voice tools, the trigger is spoken ("insert my scheduling link") and the full text is typed for you. Expansion predates dictation (keyboard tools like TextExpander built the category), but it is a natural fit for voice: saying one phrase and getting a paragraph is the highest-leverage sentence you can speak. The details that matter: triggers should tolerate natural variations in how you say them, and the expanded text should be inserted exactly as saved, never passed through a rewriting model. #### What people keep in snippets Sign-offs and signatures. Scheduling links and addresses. The three canned replies that cover half of a support inbox. Legal boilerplate. The paragraph that explains a refund policy. A commit message prefix. Anything you have typed more than five times is a candidate, and the more of it you say by voice, the more the savings compound. #### Spoken triggers are fuzzy A keyboard trigger is exact: type the abbreviation and the expansion fires. A spoken trigger goes through a speech model first, so "insert my scheduling link" may arrive as "insert my schedule link" or "insert scheduling link". A voice tool has to match the trigger loosely, against the raw transcript before any cleanup, so that it fires even when the trigger itself was slightly misheard. The expansion, on the other hand, must be exact: what lands is what you saved, character for character. #### Snippets versus the dictionary They are mirror images. A dictionary entry is one word the engine tends to get wrong, spelled your way. A snippet is one short phrase that stands for a lot of text. The dictionary fixes recognition; snippets replace typing. Both should be local to your machine, and neither should ever be reworded by a formatting model. #### How to create a text replacement on a Mac macOS has a basic text expansion feature built in, called text replacements. It is typed, not spoken, but it is the quickest way to try the idea, and it costs nothing. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Click Text Replacements. 3. Click the Add button, type the short text you want to replace, then type the full text that should replace it. 4. Click Add, then click Done. 5. Open a document or a message, type the short text, and check that it expands. #### Why does my text replacement not work in some apps? Apple states the limit plainly in the Mac User Guide: text replacements are not available in all apps. An app has to use the system text features for the replacement to fire, and some apps do not. This is the main practical difference between the built-in feature and a dedicated expansion tool. Before you build a long list, test one replacement in the apps you write in most. In apps that support it, you can also reach the same settings from Edit → Substitutions → Show Substitutions, then Text Settings. #### How to back up and sync your replacements A snippet list becomes valuable quickly, so it is worth knowing where yours lives. Apple text replacements stay up to date across your Apple devices through iCloud when iCloud Drive is set up and you are signed in to the same Apple Account on each device. You can also export them as a file to keep a copy or share it with another Mac user. 1. Choose Apple menu → System Settings, click Keyboard, then click Text Replacements. 2. Select the replacements you want, or choose Edit → Select All. 3. Drag the selection from the Text Replacements window to the desktop. This creates a file named property list.plist. 4. To restore or share the list, drag that file into the Text Replacements window on the other Mac, then click Done. Questions: - Q: What is a snippet in a dictation app? A: A saved block of text with a spoken trigger. Say the trigger and the whole block is typed for you, exactly as you saved it. - Q: Do snippets work if the trigger is misheard? A: In a well-built tool, yes. Triggers are matched loosely against the raw transcript so small recognition errors still fire them. The expansion itself is always exact. - Q: Is a snippet the same as a dictionary entry? A: No. A dictionary entry teaches the recognizer how to spell a word you say. A snippet replaces a short phrase with a long saved text. Most tools have both. - Q: Does Mac have built-in text expansion? A: Yes. It is called text replacements, and you set it up in System Settings → Keyboard → Text Replacements. Apple notes that it is not available in all apps. - Q: Do Mac text replacements sync to iPhone? A: Yes. With iCloud Drive set up and the same Apple Account on each device, a text replacement you add on your Mac works on your iPhone and iPad, and the reverse. - Q: How do I export text replacements from a Mac? A: Open System Settings → Keyboard → Text Replacements, select the entries, and drag them to the desktop. macOS creates a file named property list.plist that you can back up or drag into the same window on another Mac. In Flit: Flit’s snippets expand from spoken triggers, matched fuzzily on the raw transcript, and are inserted exactly as you wrote them. (https://flit.fyi/#snippets) See it on flit.fyi: - “Insert my scheduling link”: https://flit.fyi/#snippets - Your words: dictionary and snippets: https://flit.fyi/features#words - Canned replies, spoken: https://flit.fyi/use-cases/customer-support Related terms: personal-dictionary, verbatim-mode, spoken-punctuation, dictation ### Personal dictionary (custom vocabulary) URL: https://flit.fyi/glossary/personal-dictionary A personal dictionary is the list of words a dictation tool has been taught to spell your way: names, products, jargon, and anything the speech model would otherwise write as the most probable English. It is the single most effective fix for the errors that repeat. A speech model has seen "Kubernetes" far less often than "kuber netties", and it has never seen your colleague’s surname at all. Left alone it writes the likeliest spelling of the sounds, and it will do so every single time, which is why the mistakes that matter in dictation are not the random ones but the ones that recur. A dictionary tells the tool, once, what you meant. Implementations range from a plain word list you edit by hand to tools that notice when you correct a word and offer to remember it. The second kind is the one that gets used, because nobody maintains a list. #### How the dictionary is applied Two ways, and the good tools use both. Biasing: the recognizer is told these words are likely, so it prefers them when the audio is ambiguous. Repair: after recognition, the output is checked for the known mishearings of each term ("kuber netties", "cube or netties") and they are replaced with the correct spelling. Repair is what makes a dictionary work for the words a model cannot be biased toward, and it is what lets a tool fix a name it has mangled the same way for the tenth time. #### Learning, with permission The best signal that a word belongs in the dictionary is you fixing it. A tool that watches for that edit and offers to remember the word builds a vocabulary you never had to type. The line to hold is consent: nothing should enter the dictionary without you saying yes, because a tool that silently learns from every edit will learn your typos too, and because a list of everything you have corrected is personal data. #### Beyond a word list A dictionary is a fixed list. Some vocabulary is better read from context: the names in the thread you are replying to, the file and folder names in the project you have open. A tool that reads those locally, for the duration of one dictation, extends the dictionary to whatever you are looking at without you saving anything. That is the developer-mode idea, applied to writing. #### How to add custom vocabulary on a Mac with Voice Control The custom vocabulary that Apple documents for the Mac belongs to Voice Control, the accessibility feature for controlling a Mac by voice. You can add up to 1000 terms and record how you say each one. Keep in mind that when Voice Control is on, standard Dictation is not available. 1. Choose Apple menu → System Settings, click Accessibility in the sidebar, then click Voice Control. 2. Click Vocabulary. If it is dimmed, your Voice Control language does not support a custom vocabulary. 3. Click the Add button, type the word or phrase exactly as you want it entered, then click Next. 4. Click the Record button, say the phrase, then click Save. You can also save without recording and record the pronunciation later. 5. Repeat for each term, then click Done. #### What to put in a dictionary first A short list of the right words beats a long list of every word you know. Start with the terms that come up every day and that a speech model has no reason to spell your way. Add the rest only when you see a word come out wrong twice. - Names of the people you write to most, including surnames with unusual spellings. - Your company, product, and project names, with their exact capitalization. - Acronyms and initialisms that are spoken as letters. - Technical terms from your field: tools, libraries, drugs, statutes, part numbers. - Place names and street names you use in addresses. - Words borrowed from another language that you say in the middle of English sentences. #### How to move a vocabulary list between tools Keep your master list as a plain text file with one term on each line. It is the format most tools can read, and it survives a change of app. Voice Control on the Mac imports exactly that: a text file with each term on a separate line, or a file that Voice Control exported earlier. In Voice Control, the Vocabulary settings have an Advanced Actions menu with Import Vocabulary and Export Vocabulary, and you choose the language each time. If you use Voice Control in more than one supported language, each language has its own list. Export before you reset a Mac, because a dictionary takes weeks to build and seconds to lose. Questions: - Q: How do I teach my dictation app a name? A: Add it to the personal dictionary, or in a tool that learns, fix the word once and accept the offer to remember it. From then on the tool prefers your spelling and repairs its usual mishearing of the word. - Q: Does the dictionary sync between devices? A: In cloud tools with an account, usually yes. In local tools, it lives on the Mac and travels with a backup or an export. That is the trade for the list never leaving your machine. - Q: Why does my tool keep getting the same word wrong? A: Because a speech model writes the most probable spelling of the sounds, and for your names and jargon the probable spelling is wrong every time. That is exactly what a personal dictionary fixes. - Q: How many words can I add to Voice Control vocabulary on Mac? A: Apple documents a limit of 1000 vocabulary terms. Each term can be a word or a phrase, and you can record how you pronounce it. - Q: Can I import a list of custom words on a Mac? A: Yes, in Voice Control. Open System Settings → Accessibility → Voice Control → Vocabulary, use the Advanced Actions menu, and choose Import Vocabulary with a text file that has one term per line. - Q: Is a text replacement the same as a custom vocabulary word? A: No. A text replacement swaps something you typed for other text. A vocabulary term teaches the speech recognizer a word so it is written correctly when you say it. In Flit: Flit notices when you fix a word by hand and offers to remember it, keeps usage counts so the terms you actually say win, and repairs the engine’s known mishearings of each term. (https://flit.fyi/features#dictionary) See it on flit.fyi: - Your words, in Flit: https://flit.fyi/features#dictionary - “kuber netties” to Kubernetes: https://flit.fyi/#dictionary - The same idea, applied to your codebase: https://flit.fyi/#developers - Building a dictionary in a week: https://flit.fyi/blog/how-to-dictate-code Related terms: word-error-rate, text-expansion, automatic-speech-recognition, verbatim-mode ### Tone matching (register) URL: https://flit.fyi/glossary/tone-matching Tone matching is a formatting feature that writes the same dictated words differently depending on where they are going: lowercase and loose in a chat app, punctuated and complete in an email, untouched in a terminal. The linguistic term for the level of formality is register. You do not write the way you speak, and you do not write the same way everywhere. "shipping this today will ping you when it’s live" is right in Slack and wrong in a client email, and the words you said were the same both times. A dictation tool that produces one house style for every destination is right in one app and slightly off in all the others. Tone matching reads the app you are dictating into and applies a style rule for it. The rule can be inferred from the app’s category, chat versus mail versus notes versus code, and it should be something you can pin per app when the default is wrong for you. #### What changes between registers Casual: sentence case optional, fragments allowed, contractions kept, an emoji if you said one. Professional: full sentences, proper capitalization, contractions expanded where they read as slack, a greeting and sign-off preserved if you dictated them. Neutral: clean and complete without the polish. Verbatim: nothing changes. What never changes, in any register, is your content: the words you said are the words that land, arranged for the room. #### Context, read locally The app is the first signal. The second is the text already on screen: a reply to a thread should pick up the names in it and match its level of formality. Reading that field is a real privacy question, and the difference between tools is stark. Some send a screenshot to a server. A careful tool reads the focused field on the machine, keeps it in memory for one dictation, and stores nothing. #### When to pin a rule When the category guess is wrong for you. A founder who writes to investors in Messages wants professional there; a developer who dictates commit messages in a terminal may want neutral rather than verbatim for that one field. The rule should be one setting per app, visible, and reversible, and the defaults should be good enough that most people never open it. #### How to test tone matching in any dictation app You can check what a tool does to your tone in five minutes. The test is the same sentence, spoken the same way, in three different apps. What you compare is the form of the result and whether any word changed. 1. Pick one sentence of about twenty words that includes a name, a number, and a contraction. 2. Dictate it into a chat app, such as Messages. 3. Dictate the same sentence into a new email in Mail. 4. Dictate it a third time into a plain text editor or a terminal. 5. Put the three results side by side and compare capitalization, punctuation, and contractions. 6. Check that the name, the number, and every content word are identical in all three. If a word was added, dropped, or swapped, the tool is rewriting, not matching tone. #### Tone matching vs AI rewriting The two are easy to confuse because both use a language model and both change the text. Tone matching changes form: case, punctuation, contractions, and whether fragments are completed. Rewriting changes content: it picks different words, reorders ideas, and may add or remove a point. The practical difference is trust. After tone matching you can send the text without reading it against what you said, because the words are yours. After a rewrite you have to proofread for meaning. If you want a rewrite, ask for one as a separate step, so you always know which kind of change was made. #### How to see which apps can read what is on your screen A tool that matches tone to the thread you are replying to has to read text from other apps. On a Mac, that goes through the Accessibility permission, and you can review the list of apps that hold it at any time. 1. Choose Apple menu → System Settings, then click Privacy & Security in the sidebar. 2. Click Accessibility. 3. Read the list of apps. Each one that is turned on is allowed to control your Mac through the accessibility features. 4. Turn off any app you do not recognize or no longer use. 5. For each dictation tool that stays on, read its privacy statement to see whether the text it reads stays on your Mac. Questions: - Q: What does per-app tone mean in a dictation app? A: The tool writes the same words differently depending on the app you are in: casual in a chat app, professional in email, verbatim in a terminal. You can usually override the rule per app. - Q: Does tone matching change what I said? A: It changes the form, not the content: capitalization, punctuation, contractions, sentence completeness. Your words should survive in every register, and a tool that keeps a content check on its output guarantees that. - Q: How does the tool know which app I am in? A: From the frontmost application, which macOS reports through the accessibility layer. Some tools also read the focused text field for the names and formality of the thread; ask whether that reading happens locally or on a server. - Q: Is tone matching the same as rewriting with AI? A: No. Tone matching changes capitalization, punctuation, and contractions and keeps your words. Rewriting chooses new words and can change the meaning, so it needs proofreading. - Q: Does Apple Dictation change punctuation for different apps? A: Apple documents auto-punctuation as a single setting under System Settings → Keyboard → Dictation. It is either on or off for Dictation as a whole, and it inserts commas, periods, and question marks in supported languages. - Q: What is register in writing? A: Register is the level of formality of a piece of writing. A text to a friend and an email to a client can say the same thing in two registers, and tone matching is the feature that picks the register for you. In Flit: Flit matches tone to the app, casual in Messages, professional in Mail, verbatim in Terminal, lets you set the rule per app, and reads context locally and in memory only. (https://flit.fyi/#writes) See it on flit.fyi: - The tone cycler: Messages, Mail, Terminal: https://flit.fyi/#writes - One sentence, two apps, two results: https://flit.fyi/features#writes - Clearing an inbox by talking: https://flit.fyi/use-cases/founders - The screen-context difference, compared: https://flit.fyi/alternatives/wispr-flow Related terms: formatting-pass, verbatim-mode, accessibility-permission, disfluency ### Words per minute (WPM) URL: https://flit.fyi/glossary/words-per-minute Words per minute (WPM) measures text production speed. Conversational speech runs around 120 to 160 WPM, while composing original text at a keyboard averages roughly 31 to 40 WPM, which is the gap dictation exploits. The number that matters for writing is composition speed (writing while thinking), not transcription typing speed (copying existing text), which is why keyboard WPM figures from typing tests overstate real writing speed. Stanford’s 2016 text-entry study found speech input about three times faster than typing on a phone’s touchscreen keyboard. Delivered dictation WPM is lower than raw speaking rate because it includes pauses and thinking. A realistic measured rate for real work is 120 to 130 WPM, still around four times keyboard composition. #### Three different numbers called WPM Speaking rate: how fast words leave your mouth in conversation, 120 to 160 for most people. Typing-test speed: copying text you can see, 40 to 60 for a competent typist, higher for a fast one. Composition speed: writing something new at a keyboard, about 31 to 40, because most of the time is spent deciding what to say. Comparisons that put a speaking rate against a typing-test score flatter dictation; the honest comparison is delivered dictation against composition typing, and it still comes out around four to one. #### What “delivered” means Delivered WPM is the number of finished words that landed in your document, divided by the wall-clock time from the first key-down to the last text insertion, including the pauses where you thought, the corrections you spoke, and the tool’s own latency. It is the only WPM that predicts how much time you get back, and it is the one almost nobody publishes because it has to be measured on real dictations rather than a benchmark. #### Turning WPM into hours At 2,000 words a day, moving from 31 WPM composition to 124 WPM delivered dictation returns about 48 minutes a day, which is roughly four hours a week and five working weeks a year. The arithmetic is simple and the assumption that matters is the 2,000: a support agent or a founder clearing an inbox is above it, a developer writing mostly code is below it. Slide the number to your day before you believe the result. #### How to measure your own dictation speed on a Mac Published averages tell you little about your own day. A stopwatch and one terminal command give you a number you can trust. Measure a real task, such as replying to an email, not a passage you read aloud. 1. Choose a real piece of writing you need to do, and open the app you would normally write it in. 2. Start a timer at the moment you begin dictating, and do not pause it while you think. 3. Dictate the whole piece, fix any errors the way you normally would, and stop the timer when the text is ready to send. 4. Select the finished text and copy it with ⌘C. 5. Open Terminal and run: pbpaste | wc -w 6. Divide the word count by the minutes on the timer. For example, 300 words in 2.5 minutes is 120 WPM. 7. Repeat the same kind of task by typing, and compare the two numbers. #### Is speech really three times faster than typing? The figure comes from a 2016 study by researchers at Stanford University, the University of Washington, and Baidu. It is worth knowing exactly what was measured. The comparison was speech recognition against a touchscreen smartphone keyboard, not a desktop keyboard. In that setting, English text entry by speech was 3.0 times faster than the keyboard, and Mandarin Chinese was 2.8 times faster. The error rate with speech was also lower: 20.4 percent lower for English and 63.4 percent lower for Mandarin. The result supports a clear claim about phones. It does not, on its own, tell you the ratio at a full-size keyboard, which is why measuring your own speed is the better evidence for your own desk. #### Why your dictation WPM is lower than you expect If your measured number is well below your speaking rate, the cause is almost always time that is not speech. These are the usual places it goes. - Thinking time. Pauses between sentences count against the clock, exactly as they do when you type. - Corrections. Every word you fix by hand costs more than the word took to say. - Waiting. The gap between the end of your speech and the text appearing is part of the total. - Restarts. Apple Dictation stops automatically when no speech is detected for 30 seconds, so a long pause means starting again. - Short tasks. A one-line reply is dominated by the start and stop, so measure longer pieces as well. Questions: - Q: How many words per minute can you dictate? A: Speaking runs 120 to 160 WPM. Delivered dictation, counting pauses, corrections, and the tool’s latency, is closer to 120 to 130 in real work. Against 31 to 40 WPM for composing at a keyboard, that is about four times faster. - Q: Is 40 WPM typing slow? A: For copying text, it is average. For composing original writing it is fast: most people write new text at 31 to 40 WPM because the thinking, not the fingers, sets the pace. - Q: How much time does dictation save? A: At 2,000 words a day, roughly 48 minutes a day and around five working weeks a year, using 124 WPM delivered dictation against 31 WPM composition typing. Adjust the words-per-day figure to your own and the answer scales with it. - Q: How do I calculate words per minute? A: Divide the number of finished words by the number of minutes it took to produce them. For example, 250 words in 5 minutes is 50 WPM. - Q: Is dictation faster than typing on a phone? A: Yes, according to a 2016 Stanford, University of Washington, and Baidu study. It found speech input 3.0 times faster than a smartphone keyboard for English and 2.8 times faster for Mandarin Chinese. - Q: How do I count the words in text I copied on a Mac? A: Copy the text, open Terminal, and run pbpaste | wc -w. The command prints the number of words on the clipboard. In Flit: Flit’s own measure across 500+ real dictations is 124 WPM delivered, the number behind the 4× claim, including the wait. (https://flit.fyi/#speed) See it on flit.fyi: - The same 31-word reply, typed and spoken: https://flit.fyi/#speed - Dictation vs typing: what the studies measured: https://flit.fyi/blog/dictation-vs-typing - The comparison: https://flit.fyi/best-dictation-app-for-mac Related terms: dictation, disfluency, latency ### On-device processing (local processing) URL: https://flit.fyi/glossary/on-device-processing On-device processing means the speech recognition model runs on your own computer, so audio is transcribed locally and never uploaded. The alternative is cloud processing, where audio is sent to a server to be transcribed. The distinction decides three things at once: privacy (nothing to intercept, subpoena, or breach on a server), offline behavior (local tools keep working without a network), and often latency (no round trip). The trade is that the model must fit and run on your hardware, which is why on-device tools state minimum requirements. For confidential work, on-device is the difference between a tool your compliance review waves through and one that needs a vendor agreement: there is no processor of the audio other than your own machine. #### The whole pipeline, or part of it "Local" on a product page can mean several things. Recognition on the device but formatting in the cloud. Recognition local by default but cloud models offered as an option. Everything local except the screen context, which is uploaded for personalization. Or genuinely everything: capture, recognition, formatting, dictionary, history, and any context reading, all on the machine. Only the last one lets you turn off the network and lose nothing, and only the last one means there is no copy of your words anywhere but your own disk. #### How to verify it Three tests, in order of effort. Turn off wifi and dictate; a fully on-device tool works exactly the same. Open a network monitor such as Little Snitch, filter to the app, and dictate for an hour; you should see nothing but the connections the vendor has documented, such as model downloads you started and an update check. Read the privacy statement and see whether it names every connection the app makes and why. A tool that passes all three has earned the word. A policy page describes intentions; the network monitor describes behavior. #### Why it became possible Two things arrived together. Open speech models good enough to compete with cloud engines, and consumer chips with dedicated machine-learning hardware to run them. On Apple silicon, a large recognition model and a small language model run side by side in real time on a laptop. The compromise that used to define local tools, slower and less accurate, has mostly gone, which leaves cloud processing with cross-device sync and not much else to argue. #### How to check whether Apple Dictation is processed on your Mac Apple Dictation can run on the device or on Apple servers, depending on your Mac and your language. You do not have to guess which one applies to you, because macOS tells you in settings. 1. Choose Apple menu → System Settings, then click Keyboard in the sidebar. 2. Go to Dictation. 3. Read the text under the Dictation switch. It says whether your voice input is processed on your device. 4. If you dictate in more than one language, change the Dictation language and read the text again, because availability varies by language and Mac model. 5. If it does not say the processing is on the device, assume what you dictate is sent to Apple to be processed. #### How to watch network activity while you dictate Activity Monitor is included with every Mac and gives you a first check without installing anything. It will not show you where data goes, but it shows whether data is moving while you speak. 1. Quit apps that use the network heavily, such as browsers and sync tools, so the reading is quiet. 2. Open the Activity Monitor app and click Network. 3. Look at the bottom of the window and note Data sent and Data sent/sec. 4. Dictate for one full minute into a text editor. 5. Compare the numbers. Audio leaving your Mac for a minute shows up as a clear rise in data sent. A flat line is what on-device processing looks like. 6. Repeat with Wi-Fi turned off in System Settings → Wi-Fi. A tool that still works is not depending on a server for recognition. #### Does Apple Dictation send audio to Apple? It depends on the device and the language. Apple’s privacy notice for Siri and Dictation says your device indicates in Keyboard settings if your audio and transcripts are processed on the device and not sent to Apple servers. When that is not the case, the same notice says the things you dictate are sent to and processed on the server, and are not stored unless you opt in to Improve Siri and Dictation. Apple also describes collecting related request data, such as device specifications and performance statistics. This is the general pattern to look for in any tool: a clear statement of when audio stays local, when it does not, and what else is sent. A statement that only covers the best case is not a complete answer. Questions: - Q: What does on-device mean for a dictation app? A: The speech model, and ideally every other step, runs on your own computer. Your audio and text are never uploaded, the tool works offline, and there is no server holding a copy of anything you said. - Q: How can I check whether a dictation app is really on-device? A: Turn off wifi and dictate; it should work unchanged. Then watch it with a network monitor for a while and compare what you see with what the privacy statement says the app connects to. Behavior beats a policy page. - Q: Is on-device dictation less accurate than cloud? A: Not meaningfully, on Apple silicon with current open models. Your microphone, room, and vocabulary matter more than where the model runs. The remaining cloud advantage is syncing a dictionary across devices, not accuracy. - Q: How do I know if dictation on my Mac is on-device? A: Open System Settings → Keyboard and look under the Dictation switch. macOS states there whether your voice input is processed on your device. - Q: How do I see which apps can use my microphone on a Mac? A: Choose Apple menu → System Settings, click Privacy & Security, then click Microphone. You can turn access on or off for each app in the list. - Q: How can I tell if my Mac microphone is being used? A: macOS shows a recording indicator in Control Center when the microphone is in use or has been used recently. In Flit: Flit is on-device for every step, verifiably: turn off wifi and everything still works, and the privacy page lists every connection the app ever makes. (https://flit.fyi/#privacy) See it on flit.fyi: - Private, provably: https://flit.fyi/#privacy - The privacy statement, with the network-monitor recipe: https://flit.fyi/privacy - Private dictation apps, ranked, with three levels of testing: https://flit.fyi/blog/best-private-dictation-apps - Privileged notes that never upload: https://flit.fyi/use-cases/lawyers Related terms: speech-to-text, automatic-speech-recognition, neural-engine, local-llm, dictation ### Local LLM (on-device language model) URL: https://flit.fyi/glossary/local-llm A local LLM is a language model that runs on your own computer rather than on a vendor’s server. In dictation tools it is the model behind the formatting pass: fixing punctuation, removing fillers, and matching tone, without the transcript ever leaving the machine. The formatting pass is the step most likely to be outsourced to the cloud, because it is the step that wants a language model, and language models were until recently too big to run on a laptop. Small models changed that. A small model, run through the on-device frameworks Apple silicon provides, can clean up a sentence in a few hundred milliseconds. Local is the right home for this step for the same reason it is the right home for recognition: the model sees your words, and your words should not become a vendor’s data on the way to being punctuated. #### What the model is asked to do Less than you would think, in a well-designed tool. Not to rewrite, not to improve, not to decide clock times or spellings, all of which are rules. It is asked for judgment: which fillers were filler, where the sentences break, what tone the destination wants. A narrow brief is what lets a small model do the job well, and it is also what makes the output checkable, because a model that was never allowed to add or remove content words can be tested for having done so. #### Speed, memory, and heat A local model competes with your other apps for memory and with your battery for power. The engineering that matters is keeping it small, keeping it resident so there is no cold start, capping how much memory its runtime may hold, and giving it a time budget so a slow answer is dropped rather than waited for. On an M-series Mac the whole pass fits inside a second with room to spare. #### Cloud LLM formatting, for comparison Sending the transcript to a hosted model buys a bigger model and costs a round trip, a subscription to pay for the inference, and a copy of every sentence you dictate on someone else’s server. For prose formatting the bigger model is not a meaningful advantage; the task is small. The round trip and the copy are real costs, every time. #### How to check whether your Mac can run a local model Two things decide it: the chip and the free memory. Both take a minute to check, and neither needs any extra software. 1. Choose Apple menu → About This Mac. 2. Look for an item labeled Chip. On a Mac with Apple silicon it is followed by the name of the chip. Note the Memory figure as well. 3. Open the Activity Monitor app and click Memory. 4. With your usual apps open, look at the Memory Pressure graph at the bottom of the window. It shows how efficiently memory is serving what you are running. 5. Check Swap Used. This is the space on your startup disk being used to swap unused files to and from RAM. 6. If pressure is already high and swap is in use before you add a model, a local model will compete with your apps for memory. #### Why does a local LLM use so much memory? A language model is a large file of numbers, and all of it has to be held in memory while the model runs. Nothing is fetched from a server on demand. The memory a model needs grows with the size of the model, which is why small models are the practical choice for a task that runs many times a day. Apple silicon has one feature that helps here. MLX, the machine learning framework from Apple machine learning research, describes a unified memory model in which arrays live in shared memory and operations can run on any supported device type without transferring data. In plain terms, the model does not have to be copied between separate pools of memory for the processor and the graphics hardware. #### Local LLM vs speech model: what is the difference? They do different jobs, and a dictation tool that formats text uses both. A speech model turns audio into words. A language model takes those words and decides how they should be written. Whisper and Parakeet are speech models, not LLMs. The sizes are public. OpenAI lists Whisper from tiny at 39 million parameters to large at 1550 million, with a turbo model at 809 million that it describes as an optimized version of large-v3. NVIDIA lists Parakeet TDT 0.6B v2 at 600 million parameters, for English. The two kinds of model overlap a little. The Parakeet model card includes automatic punctuation and capitalization in the transcript, so some formatting arrives before any language model sees the text. The language model is only needed for the judgment that is left. Questions: - Q: Does dictation software need an LLM? A: For raw transcription, no. For turning a transcript into finished, tone-appropriate writing, a small language model helps a great deal, and it can run locally on Apple silicon. The best tools give it only the judgment calls and handle everything exact by rule. - Q: Can a language model run on my Mac without the cloud? A: Yes. Small models run on Apple silicon through on-device frameworks, and recent versions of macOS include models that apps can call locally. Both are fast enough to format a sentence in well under a second. - Q: Will a local LLM slow down my Mac? A: A well-behaved one will not. It should stay small, cap its memory, and run on the GPU while the rest of the machine carries on. Badly managed runtimes that let their cache grow without limit are the ones people notice. - Q: How do I know if my Mac has Apple silicon? A: Choose Apple menu → About This Mac. A Mac with Apple silicon shows an item labeled Chip, followed by the name of the chip. - Q: Is Whisper an LLM? A: No. Whisper is a speech recognition model that turns audio into text. An LLM works on text that already exists, which in dictation means cleaning up the transcript. - Q: What is MLX on a Mac? A: MLX is an array framework for machine learning on Apple silicon, published by Apple machine learning research. It is one of the ways software runs models locally on a Mac. In Flit: Flit’s formatting model runs on your Mac, with a strict brief, a content check on its output, and a time budget. (https://flit.fyi/features#writes) See it on flit.fyi: - The pass that runs on your Mac: https://flit.fyi/features#writes - What the model is allowed to decide: https://flit.fyi/#exact Related terms: formatting-pass, on-device-processing, neural-engine, hallucination ### Whisper large-v3 turbo (Whisper turbo, large-v3-turbo) URL: https://flit.fyi/glossary/whisper-large-v3-turbo Whisper large-v3 turbo is a speech recognition model from OpenAI, released on 1 October 2024. It is Whisper large-v3 with the decoder cut from 32 layers to 4, which makes it much faster at a small cost in accuracy. Whisper comes in sizes, from tiny to large, and for years the choice was between a small model that was quick and a large one that was accurate. Turbo changed that. OpenAI kept the large encoder, the part that listens, and replaced the large decoder, the part that writes, with one as shallow as the tiny model has. The result has 809 million parameters against 1,550 million for large-v3. OpenAI describes it as offering faster transcription speed with a minimal degradation in accuracy. #### What changed from large-v3 OpenAI’s announcement puts it in one sentence: turbo is an optimized version of large-v3 with only 4 decoder layers, the same as the tiny model, down from the 32 in the large series. The model card calls it a fine-tuned version of a pruned large-v3. After the decoder was cut down, the model was trained for two more epochs on the same multilingual transcription data as large-v3. Translation data was left out, which matters for the next section. #### How fast and how accurate it is OpenAI’s own table lists turbo at about 8 times the speed of the large model, and about 6 GB of video memory against about 10 GB. Those figures were measured transcribing English speech on an A100 graphics card, so they describe the ratio, not what a particular Mac will do. On accuracy, OpenAI says that across languages turbo performs similarly to large-v2, with larger degradation on some languages, and it names Thai and Cantonese. #### What it cannot do Turbo shares the limits of every Whisper model, and adds one of its own. - It does not translate. OpenAI states that turbo is not trained for translation tasks and returns the original language even when asked to translate into English. - It can hallucinate. The model card says predictions may include text that was not actually spoken. - It can repeat itself. The card says the architecture makes it prone to generating repetitive text. - It is uneven across languages, in the card’s own words. - It is not built for live use. The card says Whisper models cannot be used for real-time transcription out of the box. #### Size, languages and license The published weights are about 1.6 GB in 16-bit form. Smaller builds exist because the weights can be quantized, which trades a little accuracy for a smaller download. The model card lists 99 languages. The license is MIT for both the code and the weights, so it can be used and shipped freely. #### Running it on a Mac There is no single way. OpenAI publishes the model for Python. Other projects convert the same weights to run on Apple silicon, either on the graphics chip or through Apple’s own frameworks, and those are what Mac apps build on. Because Whisper reads a chunk of audio before it writes, it suits finished recordings and dictation that appears when you stop. Tools that show words while you are still speaking chunk the audio themselves, or use a streaming model for that part. Questions: - Q: Is Whisper large-v3 turbo better than large-v3? A: It is faster, not more accurate. OpenAI describes it as offering faster transcription with a minimal degradation in accuracy, and lists it at about 8 times the speed of the large model. - Q: How big is Whisper large-v3 turbo? A: It has 809 million parameters. The published 16-bit weights are about 1.6 GB, and quantized builds are smaller. - Q: Can Whisper turbo translate? A: No. OpenAI states that the turbo model is not trained for translation tasks and returns the original language even if translation is requested. - Q: How many languages does Whisper large-v3 turbo support? A: Its model card lists 99 languages. OpenAI notes that accuracy is uneven across them. - Q: Is Whisper large-v3 turbo free to use? A: Yes. OpenAI releases Whisper’s code and model weights under the MIT License. - Q: Does Whisper turbo work offline? A: Yes. The weights are a file. Once it is downloaded, the model runs on your own machine with no connection. In Flit: Whisper large-v3 turbo is one of the voice models you can choose in Flit. It is a download, and it runs on your Mac. (https://flit.fyi/#languages) Related terms: whisper, parakeet, quantization, hallucination, automatic-speech-recognition ### Cohere Transcribe (cohere-transcribe-03-2026) URL: https://flit.fyi/glossary/cohere-transcribe Cohere Transcribe is an open-source speech recognition model from Cohere, released in March 2026. It has 2 billion parameters, takes audio in and gives text out, and supports 14 languages. The full name on its model card is cohere-transcribe-03-2026. Cohere describes it as a dedicated audio-in, text-out model: it transcribes, and it does nothing else. There is no translation, no chat and no summary. It is released under the Apache 2.0 license, so the weights can be downloaded and run on your own hardware, and built into other software. #### The 14 languages The model card groups them by region. - European: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch and Polish. - Asia-Pacific: Chinese (Mandarin), Japanese, Korean and Vietnamese. - Middle East and North Africa: Arabic. #### How it is built Cohere describes a Conformer-based encoder-decoder: a large Conformer encoder extracts the acoustic representation, and a lightweight Transformer decoder generates the text. It was trained from scratch, not adapted from an existing model. #### How accurate it is The model card reports an average word error rate of 5.42 percent on the English sets of the Open ASR Leaderboard, as of 26 March 2026. Cohere said at release that this placed it first on that leaderboard for accuracy. Leaderboards move, so read that as a claim about that day. The figure is for English. The card does not give one number for the other 13 languages. #### What it does not do The model card is direct about three limits. - No automatic language detection. You tell it which language is being spoken, and the card says it performs inconsistently when a speaker switches language mid-sentence. - No timestamps and no speaker labels. It returns text only. - It is eager to transcribe. In Cohere’s words, like most models of its kind it will transcribe even non-speech sounds, so it benefits from a noise gate or voice activity detection in front of it. #### Running it on your own machine Cohere offers the model through its own service, and also publishes the weights. Converted and quantized builds of those weights run on Apple silicon, which is how a 2 billion parameter model fits in a laptop’s memory next to everything else. Questions: - Q: Is Cohere Transcribe open source? A: Yes. Cohere released the model under the Apache 2.0 license, and the weights are published. - Q: Which languages does Cohere Transcribe support? A: Fourteen: English, French, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Mandarin Chinese, Japanese, Korean, Vietnamese and Arabic. - Q: Does Cohere Transcribe detect the language automatically? A: No. Its model card says it does not feature explicit, automatic language detection, so the language has to be chosen. - Q: How big is Cohere Transcribe? A: Cohere describes it as a 2 billion parameter model. - Q: Does Cohere Transcribe give timestamps or speaker labels? A: No. The model card says it features neither timestamps nor speaker diarization. - Q: Can Cohere Transcribe run offline? A: Yes, when you run the published weights on your own machine. Used through Cohere’s service, it needs a connection. In Flit: Cohere Transcribe is one of the voice models you can choose in Flit, for its 14 languages. It runs on your Mac, and you pick the language, since the model does not detect it. (https://flit.fyi/#languages) Related terms: automatic-speech-recognition, word-error-rate, language-identification, voice-activity-detection, mlx ### Qwen3-ASR (Qwen3-ASR-1.7B, Qwen3-ASR-0.6B) URL: https://flit.fyi/glossary/qwen3-asr Qwen3-ASR is a family of open-source speech recognition models from Alibaba’s Qwen team, released on 29 January 2026. It recognizes 30 languages and 22 Chinese dialects, and identifies which one is being spoken. It comes in two sizes, Qwen3-ASR-1.7B and Qwen3-ASR-0.6B, with a third model, Qwen3-ForcedAligner-0.6B, for timestamps. All three are released under the Apache 2.0 license. Qwen3-ASR is built differently from older speech models. It is a language model that has been given ears: an audio encoder turns speech into tokens, and a Qwen3 language model writes the text. That is also why it is larger than a model that only transcribes. #### The two sizes The names refer to the language model inside, not to the total. The paper explains that Qwen3-ASR-1.7B is built with Qwen3-1.7B, a projector and an audio encoder of 300 million parameters, and that Qwen3-ASR-0.6B uses Qwen3-0.6B with an encoder of 180 million. So each model is somewhat larger than its name suggests. The larger one is the more accurate. The smaller one is the faster, and is the one to choose when memory is tight. #### Languages and dialects The project’s page states support for language identification and speech recognition in 52 languages and dialects: 30 languages and 22 Chinese dialects, along with English accents from several countries and regions. Language identification is part of the model. You do not have to say which language is coming; it works that out and reports it with the transcript. #### What it can do beyond plain transcription Three features are stated by its makers. - Streaming and offline recognition from the same model, so one set of weights serves both live and recorded speech. - Long audio. The paper gives a limit of 20 minutes for a single piece of speech. - Timestamps, through the separate forced aligner, for up to 5 minutes of speech in 11 languages. #### The limits its makers state The project has no section headed limitations, but its documentation sets some out. Streaming is available with one serving backend only, and streaming does not return timestamps. Timestamps need the second model. And the published speed figures were measured on servers handling many requests at once, so they say little about one person dictating on a laptop. #### Running it on a Mac The original weights are published for Python on a graphics card. For Apple silicon, the same weights are converted to Apple’s MLX format and usually quantized to 4 or 5 bits, which brings the download and the memory use down to laptop size. Questions: - Q: How many languages does Qwen3-ASR support? A: Its makers state 30 languages and 22 Chinese dialects, 52 in all, with language identification built in. - Q: Is Qwen3-ASR open source? A: Yes. The models are released under the Apache 2.0 license. - Q: What is the difference between Qwen3-ASR-1.7B and 0.6B? A: The size of the language model inside. The 1.7B model is the more accurate, and the 0.6B model is smaller and faster. - Q: Does Qwen3-ASR detect the language by itself? A: Yes. Language identification is part of the model, so the language does not have to be set in advance. - Q: Does Qwen3-ASR give timestamps? A: Not by itself. Timestamps come from a separate model, Qwen3-ForcedAligner-0.6B, for up to 5 minutes of speech in 11 languages. - Q: Can Qwen3-ASR run offline on a Mac? A: Yes. Converted builds of the published weights run on Apple silicon with no connection once downloaded. In Flit: Qwen3-ASR is one of the voice models you can choose in Flit. It covers 30 languages, detects yours, and runs on your Mac. (https://flit.fyi/#languages) Related terms: language-identification, automatic-speech-recognition, mlx, quantization, streaming-transcription ### MLX (Apple MLX, MLX framework) URL: https://flit.fyi/glossary/mlx MLX is an open-source array framework for machine learning on Apple silicon, made by Apple’s machine learning research team. It is what many language models and speech models use to run on a Mac’s own chip. If you have downloaded a model for a Mac and its name ends in MLX, or it came from a collection of MLX models, this is what that means: the weights have been converted to the format this framework reads. MLX is a tool for developers and researchers, not something you install to use an app. An app that uses it carries it inside. It is released under the MIT license, and the first public version appeared at the end of 2023. #### Why it suits a Mac Apple’s description names one difference from other frameworks: the unified memory model. On Apple silicon the processor and the graphics chip share one pool of memory. MLX arrays live in that shared memory, so work can move between the two without copying data back and forth. On most other computers a graphics card has its own, separate memory, and a model has to fit in it. On a Mac a model can use the memory the whole machine has, which is why a laptop can run models that would otherwise need a dedicated card. #### What it runs on Apple’s documentation says operations can run on the CPU and the GPU. It does not list the Neural Engine. To install it on a Mac you need Apple silicon and macOS 14 or later. The same documentation now also describes builds for Linux, including one for Nvidia graphics cards, so MLX is no longer for Apple hardware only. Apple silicon remains what it was designed around. #### The MLX family MLX itself only handles arrays and the arithmetic on them. The packages built on top are what most people meet. - MLX LM, from the same team at Apple, for generating text with large language models and fine-tuning them. - MLX Whisper, from Apple’s examples, for speech recognition with Whisper. - MLX Swift, which lets a Mac or iPhone app written in Swift use MLX directly. - Community libraries for audio and speech, which are built on MLX but are not Apple’s. #### MLX models and quantization Model names in MLX collections often end in 4bit or 8bit. That is quantization: the weights are stored with fewer bits each, which makes the download smaller and the model lighter on memory, at some cost in accuracy. The MLX tools can produce these builds when a model is converted. #### MLX and Core ML Apple has two frameworks that run models on its devices, and they are for different jobs. Apple describes MLX as an open-source array framework for experimenting with, training and fine-tuning generative models on Apple silicon. It describes Core ML as the way to integrate machine learning models into an app. In practice many apps use both: Core ML for a model that suits the Neural Engine, and MLX for a large language or speech model that runs on the graphics chip. Questions: - Q: What does MLX stand for? A: Apple does not give an expansion. It describes MLX as an array framework for machine learning on Apple silicon. - Q: Does MLX work on Intel Macs? A: No. On a Mac, Apple’s documentation requires Apple silicon and macOS 14 or later. - Q: Does MLX use the Neural Engine? A: Apple’s documentation says MLX operations run on the CPU and the GPU. It does not list the Neural Engine. - Q: Is MLX free? A: Yes. It is open source under the MIT license. - Q: Do I need to install MLX to use an app built with it? A: No. An app that uses MLX includes it. Installing MLX yourself is for writing code or running models from the command line. - Q: What is an MLX model? A: A model whose weights have been converted to the format MLX reads, often quantized to 4 or 8 bits so that it fits comfortably in a Mac’s memory. In Flit: Flit runs its formatting model and three of its voice models, Cohere Transcribe, Qwen3-ASR and Omnilingual, with MLX, on your Mac. (https://flit.fyi/features#privacy) Related terms: core-ml, quantization, neural-engine, local-llm, on-device-processing ### Core ML (CoreML, Apple Core ML) URL: https://flit.fyi/glossary/core-ml Core ML is Apple’s framework for running machine learning models inside an app, on the device. It uses the CPU, the GPU and the Neural Engine, and needs no network connection to make a prediction. Core ML has been part of Apple’s systems since 2017, in macOS 10.13 and iOS 11. An app hands it a model and some input, such as audio, an image or text, and it returns the model’s answer. Where the arithmetic happens is Core ML’s decision. You do not install or open Core ML. It is part of macOS. You meet the name in an app’s description, in a model’s file name, or when a download says it contains Core ML weights. #### What it does Apple’s description: Core ML provides a unified representation for all models, and an app uses it to make predictions, and to train or fine-tune models, all on a person’s device. On performance, Apple says Core ML optimizes on-device performance by leveraging the CPU, GPU and Neural Engine while minimizing its memory footprint and power consumption. The Neural Engine is the part of Apple’s chips built for this kind of arithmetic, and Core ML is the usual way for an app to reach it. #### Why on-device matters Apple makes the argument itself: running a model strictly on a person’s device removes any need for a network connection, which helps keep a person’s data private and the app responsive. For speech, that means the audio does not have to leave the Mac to become text. Whether a particular app keeps it there is up to the app, but the framework makes it possible. #### The file types Core ML models come in three forms, and the extension tells you which. - `.mlmodel`: the original model definition file. - `.mlpackage`: the newer package format, which keeps the weights separate from the program. It is supported from Xcode 13. - `.mlmodelc`: a compiled model, which is what an app loads. A folder with this ending inside an app or its downloads is a Core ML model ready to run. #### Where the models come from Most models are trained with other tools and then converted. Apple publishes Core ML Tools for that, to convert models from training libraries such as PyTorch and TensorFlow into the Core ML format. The same tools can compress a model on the way in, by quantization or by palettization, which makes it smaller and often faster on the Neural Engine. #### Core ML and MLX Apple describes the two differently. Core ML is for integrating models into apps. MLX is an open-source array framework for experimenting with, training and fine-tuning generative models on Apple silicon. Apple’s documentation says MLX runs on the CPU and the GPU, while Core ML can also use the Neural Engine. Apple’s pages now also point developers who work with the newest model architectures to a separate framework, Core AI. Core ML remains the long-standing way to run a model in an app. Questions: - Q: What is Core ML used for? A: Running machine learning models inside apps on Apple devices: recognizing speech, classifying images, reading text and similar tasks, without sending the data to a server. - Q: Does Core ML need the internet? A: No. Apple says that running a model strictly on the device removes any need for a network connection. - Q: Does Core ML use the Neural Engine? A: Yes. Apple says Core ML leverages the CPU, GPU and Neural Engine. - Q: What is an .mlmodelc file? A: A compiled Core ML model, the form an app loads and runs. - Q: What is the difference between Core ML and MLX? A: Apple describes Core ML as the way to integrate models into apps, and MLX as an array framework for experimenting with and fine-tuning generative models on Apple silicon. Core ML can use the Neural Engine; MLX runs on the CPU and GPU. - Q: Is Core ML on Intel Macs? A: Yes. Core ML is part of macOS from version 10.13. The Neural Engine is only in Apple silicon, so on an Intel Mac it uses the CPU and GPU. In Flit: Flit’s Parakeet voice models (V3, Ultra and Unified) and Nemotron stream through Core ML on your Mac. Voxtral Realtime streams too, but runs on the GPU through ggml. (https://flit.fyi/features#privacy) Related terms: mlx, neural-engine, quantization, on-device-processing, parakeet ### Quantization (model quantization, 4-bit, 8-bit) URL: https://flit.fyi/glossary/quantization Quantization is storing a model’s numbers with fewer bits each, for example 4 or 8 bits where the original used 16 or 32. It makes the model smaller and lighter on memory, usually at a small cost in accuracy. A model is mostly a very long list of numbers, called weights. As trained, each weight is typically a 32-bit or 16-bit floating point value. Quantization rewrites them in a coarser form. Apple’s documentation defines it as reducing the number of bits that represent a number. This is the reason the same model can appear as several downloads of different sizes, with labels such as 4bit, 8bit or Q4. They are one model at different levels of compression. #### How much smaller Size follows the bits almost directly. Halve the bits per weight and the file is about half the size. One published example makes it concrete. For an 8 billion parameter language model, the 16-bit file is about 15 GiB, the 8-bit file about 8 GiB, and a common 4-bit build about 4.6 GiB. The 4-bit build works out at about 4.9 bits per weight, not exactly 4, because some parts are kept at higher precision. #### What you give up Accuracy, a little. Coarser numbers mean the model’s arithmetic is slightly off everywhere, and the errors add up. How much it shows depends on the model and on how far you go. Apple’s guidance is that compressing to 8 bits, and for some models to 6 or 4, can work with only a slight decrease in accuracy. #### What you gain besides size Apple lists the possible gains of compressing a model as a smaller memory footprint, lower latency and lower power use, alone or together. On a laptop, memory is often the one that matters: a model that fits comfortably leaves room for everything else you have open. #### Reading the labels The label tells you the precision of the weights. - fp32 or float32: full precision, as most models are trained. - fp16, bf16 or float16: half precision. Often the form a model is published in. - 8bit or int8: a quarter of full precision. - 4bit or int4: an eighth of full precision. The usual choice for running large models on a laptop. - Q4, Q5, Q8 with letters after them: the same idea in one common file format, where the letters name the method used. #### Quantization and palettization Apple’s tools offer two ways to store weights in fewer bits. Linear quantization maps the range of the original values onto a small range of integers. Palettization, also called weight clustering, groups the weights into clusters and stores each weight as an index into a short lookup table. Apple’s guidance is that palettization typically works best on the Neural Engine, and that 4-bit quantization can work well for models running on a Mac’s graphics chip. Questions: - Q: What does 4-bit mean on a model? A: That its weights are stored in about 4 bits each, where the original used 16 or 32. The file is much smaller, and the model is slightly less accurate. - Q: Does quantization make a model worse? A: Slightly. It trades some accuracy for a smaller size and lower memory use. Apple’s guidance is that 8 bits, and for some models 6 or 4, costs only a slight decrease in accuracy. - Q: Is a 4-bit or an 8-bit model better? A: The 8-bit model is closer to the original, and the 4-bit model is about half its size. Choose 8-bit if you have the memory and 4-bit if you do not. - Q: Does quantization make a model faster? A: It can. Apple lists lower latency among the possible gains, along with a smaller memory footprint and lower power use. - Q: Can any model be quantized? A: Most can, with tools that convert the published weights. How well it holds up varies from model to model. - Q: Is quantization the same as compression? A: It is one kind of model compression. Others include pruning, which removes weights, and palettization, which stores weights as entries in a lookup table. In Flit: Flit’s formatting model and its Qwen3-ASR voice models are quantized builds, and Voxtral Realtime ships as a 4-bit GGUF build. That is how they fit in a Mac’s memory alongside your other apps. (https://flit.fyi/features#privacy) Related terms: mlx, core-ml, local-llm, whisper-large-v3-turbo, neural-engine ### Real-time factor (RTF, RTFx) URL: https://flit.fyi/glossary/real-time-factor Real-time factor (RTF) is the time a speech recognition system takes to process audio, divided by the length of that audio. An RTF of 0.5 means one minute of speech is transcribed in 30 seconds; below 1 is faster than real time. It is the standard way to say how fast a speech model is, independent of how long the recording was. The definition goes back to evaluations run by NIST, the US standards body, which called it the speed factor: total processing time divided by the duration of the source signal. Its inverse, RTFx, has become the more common figure, because bigger is better and the numbers are easier to read. Both describe the same thing. #### The formula RTF is processing time divided by audio duration. If 10 seconds of audio take 2 seconds to transcribe, the RTF is 0.2. RTFx turns it over: audio duration divided by processing time. The same example gives an RTFx of 5. The Open ASR Leaderboard, which publishes RTFx for the models it ranks, explains that an RTFx of 1 means a system processes speech as fast as it is spoken, and an RTFx of 2 means it takes half the time. #### How to read a figure Check which of the two you are looking at, because they point in opposite directions. - RTF: lower is faster. Below 1 keeps up with live speech; above 1 falls behind. - RTFx: higher is faster. Above 1 keeps up with live speech. - An RTF of 0.01 and an RTFx of 100 are the same speed: an hour of audio in 36 seconds. #### Why published figures are so high Model cards often quote an RTFx in the hundreds or thousands. Those are measured on a data-center graphics card with many recordings processed at once. One model’s card, quoting a figure above 3,000, adds that performance may vary depending on audio duration and batch size. A laptop transcribing one voice gets nothing like that, and does not need to. The figure is useful for comparing models with each other, not for predicting what you will see. #### What it does not tell you RTF is an average over a whole recording. It says nothing about when the first word appears, or how long you wait after you stop speaking, and for dictation those are what you feel. A model with an excellent RTF that waits for a whole chunk of audio before writing can feel slower than a streaming model with a worse one. For live use, latency is the measure to ask about, and RTF is the check that the model can keep up at all. #### What changes it The same model gives different figures in different conditions. - The hardware, and whether the model runs on the CPU, the graphics chip or the Neural Engine. - The size of the model, and whether it has been quantized. - The length of the audio. Very short clips carry a fixed start-up cost that makes the ratio look worse. - Batching. Processing many recordings at once raises RTFx sharply on a server. Questions: - Q: What is a good real-time factor? A: For live dictation, anything comfortably below 1, so the model keeps ahead of your speech. For transcribing files, lower simply means less waiting. - Q: What is the difference between RTF and RTFx? A: They are inverses. RTF is processing time divided by audio length, so lower is faster. RTFx is audio length divided by processing time, so higher is faster. - Q: How do you calculate real-time factor? A: Divide the time taken to transcribe by the duration of the audio. Ten seconds of audio transcribed in two seconds is an RTF of 0.2, or an RTFx of 5. - Q: Does a low real-time factor mean low latency? A: Not necessarily. RTF is an average over the recording. A model can have a low RTF and still make you wait, if it reads a whole chunk before writing anything. - Q: What does an RTF of 1 mean? A: The system takes exactly as long to transcribe the audio as the audio lasts. It only just keeps up with live speech. - Q: Why is my speed lower than the figure on the model card? A: Published figures are usually measured on a server graphics card with many recordings in a batch. One recording on a laptop is a different test. In Flit: Flit’s text lands about 0.7 seconds after you let go of the key. That wait, not an average over the recording, is the figure Flit is tuned for. (https://flit.fyi/features) Related terms: latency, streaming-transcription, word-error-rate, quantization, parakeet