How to Dictate on Windows 11: The 2026 Voice Typing Guide

2026-10-06 · 52 min read · Yap team

Every way to turn speech into text on a Windows PC, what each one does with your voice, and how to make it work

Three out of four desktop computers run Windows. In September 2026, Windows held 76.36% of the worldwide desktop operating system market - StatCounter, and every one of those computers ships with dictation built in. The quick answer to "how do I dictate on Windows" fits in one sentence: click into any text box, press Windows key + H, wait for "Listening...", and talk; say "stop listening" when you are done - Microsoft Support.

The longer answer is where people get stuck. Windows does not have one dictation feature; it has several that share almost nothing. Voice typing (Windows+H) sends your speech to Microsoft's online service. Voice access recognizes speech on the PC itself, but it is built around controlling the computer and speaks fewer languages. Microsoft 365 Dictate lives inside Word and Outlook and needs a subscription. Windows 10 still carries an older dictation with a shorter language list, and Copilot+ PCs add a new layer called fluid dictation on top. Each has its own languages, commands, privacy rules and failure modes, and most people only ever discover one of them.

This guide untangles all of it. You will learn how to start dictating in two minutes, how each Microsoft route actually processes your voice, which voice commands and punctuation words work where, how to install and switch languages, how to dictate without an internet connection, what to change in Windows' privacy settings, how dictation works in Word, Outlook and Google Docs, what is different on Windows 10, and how to fix voice typing when it stops working. It ends with a decision framework and an honest look at when a dedicated dictation app is worth paying for.

One disclosure before we start: we make Yap, an on-device dictation app that added a Windows version this month. So every factual claim below links to the vendor's own page as we read it on October 6, 2026, and Yap is scored by the same rules as everything else in the scorecard, where it places third of eight, behind a free Microsoft feature. If the built-in tools are enough for you, this guide will say so, and it will show you how to get the most out of them.

Contents

  1. The 2026 Windows dictation scorecard
  2. Voice typing in two minutes: the quick start
  3. How Windows dictation works under the hood
  4. Voice commands and punctuation
  5. Languages: what is supported and how to switch
  6. Fluid dictation and Copilot+ PCs
  7. Dictating offline on Windows
  8. Privacy: what Windows sends, and how to change it
  9. Dictating in Word, Outlook and Google Docs
  10. Windows 10 versus Windows 11
  11. Troubleshooting: when voice typing is not working
  12. When a dedicated dictation app is worth it
  13. A decision framework

1. The 2026 Windows dictation scorecard

Here is the field on one page: eight ways to dictate on a Windows PC, three of them built into Windows and Microsoft 365 and five of them apps you install. Each one types text from your speech, and each publishes enough about its pricing and data handling to be scored on evidence. Two names are covered in prose instead of in the table: Dragon Professional, whose product page lists no price to score, and Google Docs voice typing, which only works inside Google Docs and leaves the speech processing to whichever browser you use. Our general guide to choosing a dictation app compares apps across Mac and Windows; this table answers the narrower question of what works best on a Windows PC.

Every cell holds a score from 0 to 10 and the fact behind it, so you can disagree with the judgment without redoing the research. The final score is the weighted average, rounded to one decimal, and the table is sorted by it, with ties listed alphabetically. Read the reason before the number, because two options can earn the same score for opposite reasons, and the reason is what tells you whether an option fits the way you work.

#OptionWhat it doesPrivacy (30%)Reach (25%)Language (20%)Cost (25%)Final
1SuperwhisperDictation app with offline models or cloud AI8 - on-device models on Windows; cloud models optional8 - types in any app; Windows 10 and 119 - 100+ languages and dialects8 - free tier with unlimited Whisper models; Pro $8.49/mo8.2
2Voice accessBuilt-in voice control that also dictates10 - on-device; data not shared to the cloud6 - any app, but needs setup, a bar on screen and Windows 11 22H2 or later5 - 15 entries in seven languages10 - free8.0
3YapOn-device dictation for Windows and Mac10 - recognition and history stay on the PC8 - hold Ctrl+Win in any app; Windows 10 and 11, x64 and Arm6 - 25 languages, detected per recording6 - $69/yr or $6.90/mo after 3 free days and a 10-day trial7.7
4Voice typingBuilt-in dictation, Win+H4 - online Azure speech; needs internet9 - Win+H in any text field on Windows 10 and 118 - 46 entries on Windows 1110 - free7.6
5Willow VoiceCloud dictation with an offline mode6 - offline mode runs on device; cloud otherwise8 - types in any app9 - over 100 languages6 - free on a weaker model; Pro $12/mo billed yearly7.1
6Wispr FlowCloud dictation with AI formatting3 - always cloud; data can train models if sharing is on9 - any app, plus Mac, iOS and Android9 - 100+ languages, auto-detected4 - 2,000 free words a week; Pro $12/mo billed yearly6.0
7Aqua VoiceCloud dictation on its own Avalon model3 - cloud only, no offline mode8 - types in any app; Windows 10 and 117 - 49 languages5 - 1,000 free words; Pro $8/mo billed yearly5.6
8Microsoft 365 DictateDictation inside Word and Outlook5 - online; audio and text not stored4 - only inside Word, Outlook and other 365 apps9 - 15 languages plus 36 in preview5 - only with a paid 365 plan5.6

How the criteria work. Privacy (30%) asks where your speech is recognized, whether dictation works without an internet connection, and what the vendor says it keeps. Reach (25%) asks whether the option types into any application, how quickly you can start, and which Windows versions it runs on. Language (20%) counts the language and region entries each vendor lists, and a bigger list only helps if your languages are on it. Cost (25%) reflects what a year of full use costs, with free tiers counted when they are usable every day. Accuracy is deliberately not scored, because no public test runs these eight options on the same recordings; section 7 shows our own measurements for the one engine we can test, and our buyer's guide explains how to measure any app on your own voice in ten minutes.

The result is less obvious than "use the built-in tool" or "buy an app". Superwhisper wins on breadth: a free tier with unlimited local models and more than 100 languages. Voice access, a free Windows feature most people have never opened, comes second because it is the only Microsoft dictation that works fully offline. Yap places third: the strongest privacy score and a simple hold-to-talk shortcut, held back by price and a 25-language list. Voice typing lands fourth, mainly because it needs an internet connection and sends your speech to Microsoft, and the cloud apps and Microsoft 365 Dictate follow, each limited by privacy, cost, or the fact that it only works inside one family of apps.

Why this matters: most Windows users choose a dictation tool by accident, using whichever one they found first, and then blame dictation itself when that tool's particular weakness bites. How to apply it: use the table to pick two candidates that fit your constraints, then run the quick start in section 2 with each. If you must work offline, start with voice access and the on-device apps; if you mostly write in Word, try Microsoft 365 Dictate before anything else, because it is already in the program you use.

2. Voice typing in two minutes: the quick start

Voice typing is the dictation most Windows users mean when they say "Windows dictation", and it takes about two minutes to try. Microsoft lists three requirements: you need to be connected to the internet, have a working microphone, and have your cursor in a text box - Microsoft Support. That last requirement trips people up more than the other two: voice typing only starts when a text field has focus, so pressing the shortcut on the desktop does nothing visible.

Once those three things are true, the rest is a keyboard shortcut. Pressing Windows key + H opens the voice typing menu, which starts listening on its own; you wait for the "Listening..." alert and speak. On a tablet or touchscreen, the microphone button on the touch keyboard does the same thing. To stop, you say a command such as "stop listening" or "pause voice typing", or press the microphone button in the menu. Here is the whole sequence:

  1. Click into a text box, then press Windows + H and wait for "Listening...".
  2. Speak in full sentences, at your normal pace and volume.
  3. Say "stop listening", or press the microphone button, when you are done.

Before your second dictation, open the gear icon in the voice typing menu, because two of its settings change your results immediately. Automatic punctuation lets voice typing insert commas and full stops "based on what you're saying"; without it, you dictate every punctuation mark by name. The default microphone setting decides which microphone voice typing listens to, which matters as soon as you plug in a headset or dock a laptop to a monitor with its own microphone. Microsoft's own screenshot shows the settings panel, opened over an email in Outlook:

The panel holds five controls worth knowing. The voice typing launcher lets you start voice typing from a text box with a click instead of a shortcut. Fluid dictation, which appears only on Copilot+ PCs, cleans up grammar and filler words and is covered in section 6. Filter profanity decides whether swear words are typed out or blocked. The microphone selector sits below them, and the footer reads "Powered by Microsoft Online Speech Tech", which is the plainest statement anywhere in Windows of where your voice is going. Microsoft also lists a wait time before acting setting that controls how long voice typing pauses before treating something you said as a command.

Two more shortcuts round out the basics. Windows key + Alt + H moves keyboard focus into the voice typing menu, so you can reach its buttons without a mouse. And voice typing has echo cancellation, which Microsoft says "automatically suppresses audio coming from your speakers", so a video playing in the background or the Narrator screen reader does not end up in your text. If you use a laptop's built-in microphone in a busy room, that feature helps, but it does not replace a headset.

A word on technique, because it matters as much as settings. Speech recognition models work best on complete phrases, since they use the words around each word to decide what you said, so think the sentence through and then say it in one go rather than word by word. Speak at your normal conversational pace, keep the microphone at a steady distance, and resist the urge to correct each word as it appears. If a phrase comes out wrong, say "delete that" and say it again, which is usually faster than fixing it by hand; section 4 covers the commands that make this work.

Why this matters: the first two minutes decide whether people keep dictating, and the most common first-time failures (no text box selected, punctuation off, the wrong microphone) are settings problems rather than recognition problems. How to apply it: before your first real dictation, open the gear icon, turn on automatic punctuation, pick your microphone, and switch on the launcher if you prefer clicking to shortcuts. Then dictate a short email and read it back before deciding what you think of the accuracy.

3. How Windows dictation works under the hood

Every dictation tool runs the same basic chain: it captures audio from the microphone, turns that audio into words with a speech recognition model, formats the words into text, and inserts the text where your cursor is. What separates Microsoft's three routes is the second step, where recognition runs, because that decides whether dictation needs the internet, who processes your voice, and how many languages are practical. Microsoft has made three different choices for three different products, and the diagram shows them side by side.

Voice typing uses "online speech recognition, which is powered by Azure Speech services", the same cloud speech platform Microsoft sells to developers. Voice access downloads language files "to be used by on-device speech recognition" and then works without a connection - Microsoft Support. Microsoft 365 Dictate in Word and Outlook needs "a microphone and reliable internet connection", and Microsoft says your "speech utterances will be sent to Microsoft and used only to provide you with text results" - Microsoft Support. On Copilot+ PCs, fluid dictation then adds a cleanup step that runs on the PC itself.

Why would one company make three different choices? From first principles, it is a trade between model size, language coverage and independence from the network. A cloud service can run large models for dozens of languages and improve them without shipping anything to your PC, which is how voice typing covers 46 language and region entries on Windows 11. An on-device model has to fit on every PC that uses it and download once per language, which is one reason voice access covers far fewer. Microsoft has not published the models behind voice typing, so we cannot score its accuracy directly, but the company's commercial Azure Speech model is one of the most accurate English systems on Hugging Face's public Open ASR Leaderboard, averaging a 3.81% word error rate across its English test sets - Open ASR Leaderboard.

The trade-offs show up in daily use. A cloud route needs a working connection for every sentence, so voice typing fails on a plane, on a weak hotel network, or behind a corporate proxy that blocks the service, and every word you dictate is transmitted. An on-device route works anywhere and keeps audio local, but the first setup downloads the language files, and the model is fixed until the next update. Microsoft itself frames voice access as on-device: its setup screen says your data "isn't shared to the cloud", while voice typing's own panel says it is "Powered by Microsoft Online Speech Tech".

There is also a fourth route that still matters to anyone who has used Windows for a long time. Windows Speech Recognition, the classic command and dictation feature, is being replaced: Microsoft says that on Windows 11 22H2 and later it "will be replaced by voice access starting in September 2024", while older versions of Windows keep it - Microsoft Support. If you trained Windows Speech Recognition to your voice years ago and wonder where it went, voice access is its successor.

Why this matters: the processing location is not a technical footnote. It decides whether dictation works offline, how fast it responds on a bad connection, which languages are available, and who handles your voice, and you cannot change it with a setting, only by choosing a different route. How to apply it: decide first whether you need offline dictation or must keep speech on the PC. If you do, your Microsoft option is voice access, and your app options are the on-device ones in the scorecard. If not, voice typing is the fastest place to start.

4. Voice commands and punctuation

Dictation has a problem that typing does not: the same sound can be a word you want typed or an instruction you want obeyed. "Delete that" is a command when you mean to remove the last phrase and three ordinary words when you are writing about deleting files. Every dictation system has to guess, and the three Microsoft routes guess differently, which is why commands that work in one place do nothing in another. Knowing the handful that matter, and how each route separates commands from text, removes most of the frustration people report in their first week.

Voice typing on Windows 11 keeps its command list short and predictable. You can stop with "stop listening", "stop dictation" or "pause voice typing", remove the last phrase with "delete that" or "scratch that", select it with "select that", press keys with "press enter", "backspace" or "tab", and take back the last change with "undo that" or "revert" - Microsoft Support. Microsoft also added the wait time before acting setting so that slower speakers are not cut off mid-thought, which is the same collision problem attacked from the timing side. Voice access has a much larger vocabulary, because it was designed to control the whole PC:

What you wantVoice typing (Windows 11)Voice access
Stop listening"Stop listening", "pause voice typing""Voice access sleep", "mute"
Remove the last phrase"Delete that", "scratch that""Delete that", "strike that"
Fix a misheard wordNot available; select and retype"Correct that", "correct (word)"
Spell a nameNot available"Spell out", "spell that"
Teach a new wordNot available"Add to vocabulary"

That table explains why voice access, despite its fiddlier setup, is the better Microsoft tool for people who dictate long documents. "Correct that" opens alternatives for the last phrase instead of forcing you back to the keyboard, "spell out" handles names the recognizer has never heard, and "add to vocabulary" teaches it a word for next time - Microsoft Support. Voice access also lets you force a phrase to be typed rather than obeyed by saying "type" before it, and it has a dictation mode ("switch to dictation mode") that stops interpreting commands entirely while you write a long passage. Those two features are the cleanest answers on Windows to the word-or-command problem.

Punctuation works in two ways. With automatic punctuation turned on, voice typing inserts commas and full stops from the rhythm of your speech, which is faster and usually right for ordinary sentences. You can also say the marks: in English, Microsoft lists "comma", "period" or "full stop", "question mark", "exclamation mark", "colon", "semicolon", "open quotes" and "close quotes", "ellipsis", and even "smiley face" and "heart emoji" for the matching emoticons. Voice access supports spoken punctuation too, and it can punctuate automatically, but Microsoft notes that automatic punctuation is off by default there; you turn it on from the voice access bar under Settings, Manage options - Microsoft Support.

A practical habit follows from all this. Let automatic punctuation handle commas and full stops, and say only the marks a model cannot infer from tone, such as quotation marks, colons and the occasional question mark it misses. Pause briefly before a command so it is heard as a separate utterance, and keep commands to the two or three you use constantly rather than memorizing the list. On Windows 10, the older dictation has a richer navigation vocabulary ("go after that", "select the next three words", "start spelling"), but Microsoft notes that those dictation commands "are available in US English only", so anyone dictating in another language there is limited to text and basic punctuation.

Why this matters: most dictation errors that people describe as "it typed the wrong thing" are command collisions or punctuation mistakes rather than recognition errors, and they are the ones you can fix with habit instead of a new tool. How to apply it: turn on automatic punctuation, learn "delete that", "select that" and "stop listening" first, and if you dictate long documents on Windows 11, try voice access's dictation mode and "correct that" before concluding that Windows dictation cannot handle real writing.

5. Languages: what is supported and how to switch

Language support is where Microsoft's three routes differ most, and where the differences cost multilingual users the most time. Voice typing lists 46 language and region entries on Windows 11, from Bulgarian to Vietnamese, including six varieties of English and three of Chinese. Microsoft 365 Dictate in Word and Outlook lists 15 fully supported entries and 36 more in preview, with the warning that "Preview Languages may have lower accuracy or limited punctuation support". Voice access names 15 language and dialect entries across seven languages: English, Spanish, German, French, Chinese, Japanese and Italian.

Two more data points complete the picture. The older Windows 10 dictation names US English plus 12 more entries: Simplified Chinese, four other varieties of English, French for France and Canada, German, Italian, Brazilian Portuguese and Spanish for Mexico and Spain. Google Docs voice typing, which you can use on Windows in Chrome or Edge, lists 118 languages and accents, the longest list here, though its voice commands are "available only in English" - Google Docs Editors Help. The chart puts the counts side by side, with Yap's for comparison:

The counts measure different things, so read them with care. An "entry" can be a language or a regional variety, so Google's 118 includes several varieties of Arabic, English and Spanish, and Microsoft's 46 counts English six times. A long list also says nothing about quality per language: our guide to choosing a dictation app shows that even strong multilingual models make two or more times as many errors outside English, which is a reminder to test your second language before relying on it. And Yap's 25 are the European languages its on-device model covers; if you dictate in Hindi, Japanese or Thai, it is the wrong tool, and voice typing or Google Docs is the better start.

Switching languages is the other half of the story. Voice typing follows your Windows input language, so you first install the language in Settings, Time & language, Language & region, by selecting "Add a language", and then switch with Windows key + Spacebar or the language switcher in the taskbar - Microsoft Support. Microsoft notes that optional language features, including speech recognition, "aren't required for voice typing to work", so you do not need the large speech pack for voice typing. If you see the message "Voice typing isn't available in the current language", your input language is not on the supported list, and switching to one that is will fix it.

That design has a cost for anyone who writes in two languages in the same hour: each switch is a keyboard shortcut and a mental context change, and forgetting to switch produces nonsense in the wrong language. Apps built on multilingual models take the opposite approach. NVIDIA's Parakeet TDT 0.6B v3, the model Yap runs on Windows, "automatically detects the language of the audio and transcribes it without requiring additional prompting" - NVIDIA model card, so each recording is heard in the language you actually spoke. Wispr Flow advertises the same behavior for its 100+ languages in the cloud.

Why this matters: for people who write in more than one language, switching friction, not accuracy, is usually what ends their use of dictation. How to apply it: list every language you write in and check each one against the route you plan to use; if they are all on voice typing's list and you rarely switch, Windows+H plus Windows+Spacebar is enough. If you switch constantly, test an app with automatic language detection on two sentences in two languages, back to back, without touching any setting.

6. Fluid dictation and Copilot+ PCs

Fluid dictation is Microsoft's newest addition to voice typing and the clearest sign of where Windows dictation is heading. Microsoft describes it as a feature that "automatically corrects grammar, punctuation, and filler words as you speak, reducing the need for manual editing", and says it is "powered by on-device small language models (SLMs) to ensure fast and private processing" - Microsoft Support. In practice, that means the "um", the restarted sentence and the missing comma are smoothed out before the text lands, which is the same job the cloud rewriting in apps like Wispr Flow performs, done on your own hardware.

The catch is the hardware. Fluid dictation is "a Copilot+ PC feature" available "in all English locales", and Copilot+ PCs are a specific class of Windows 11 computer. Microsoft defines them by a neural processing unit, or NPU, "capable of at least 40 TOPS", meaning 40 trillion operations per second - Microsoft, and it adds that devices "with fewer than 16 GB of RAM or lower storage configurations are not Copilot+ PCs" even when their chip has such an NPU - Microsoft. If your PC is not a Copilot+ PC, the fluid dictation switch simply does not appear.

If you do have a Copilot+ PC, the setup is automatic but not instant. The small language models download in the background the first time you launch voice typing, and until they finish, the fluid dictation option in the settings panel appears dimmed; Windows shows a progress tip and then notifies you when it is ready. Once active, fluid dictation is on by default, works "across all text editing surfaces", and turns itself off in secure fields such as password and PIN boxes. Voice access can use it too, with the commands "turn on fluid dictation" and "turn off fluid dictation".

Like any rewriting system, fluid dictation sometimes changes what you meant. Microsoft built an escape hatch for exactly that: say "revert" or "undo that", or press Ctrl+Z, and voice typing restores the original sentence. Use it whenever precision matters more than polish, for example in a quotation, a technical term the cleanup "corrects" into a common word, or a deliberately informal sentence. If you find yourself reverting often, turn the feature off in voice typing settings; the raw transcription underneath is the same either way.

It is worth being precise about what fluid dictation does not change. The cleanup runs on the device, but it is part of voice typing, and voice typing's recognition is still Microsoft's online service, so a Copilot+ PC does not make voice typing work offline. For fully on-device dictation on Windows, you still need voice access or an app that runs its recognition model locally, which section 7 covers. The direction of travel is clear, though: Microsoft is moving more of the dictation pipeline onto the PC, starting with the step that benefits most from an NPU.

Why this matters: cleanup is the difference between dictated text you can send and dictated text you must edit, and on a Copilot+ PC you get it for free without sending your text to a third-party rewriting service. How to apply it: if you own a Copilot+ PC, open voice typing once, let the models download, and dictate a messy paragraph to see the cleanup at work; keep "revert" in mind for the moments it overreaches. If you do not, do not buy new hardware for this feature alone, because an app with a good model and a little care with punctuation gets you most of the way.

7. Dictating offline on Windows

Voice typing stops working the moment your connection does, because every sentence is recognized online; Microsoft's requirements list "connected to the internet" first. The older Windows 10 dictation has the same dependency: "To use dictation, your PC needs to be connected to the internet." For many people that never matters. For anyone who writes on trains and planes, works on a locked-down corporate network, handles material that should not leave the machine, or simply dislikes the idea of every word crossing the internet, it is the deciding limitation.

Windows' own answer is voice access, and it is better than its accessibility branding suggests. Microsoft describes it as a way to "control their PC and author text using only their voice and without an internet connection" - Microsoft Support, available on Windows 11 version 22H2 and later. You turn it on under Settings, Accessibility, Speech, or by searching for "voice access" from Start, and you can set it to start before or after you sign in. The first launch is the only step that needs a connection: you pick a language, and voice access downloads the files for on-device recognition. Microsoft's own screenshot shows that welcome screen mid-download:

Read the bullet points on that screen, because they are the clearest privacy statement Microsoft makes about any of its dictation features: voice access processes "your resulting voice data on your device", "Your data isn't shared to the cloud", and the network is used for "necessary services, like downloading voice access language files". After the download and a short microphone setup, voice access shows a bar across the top of your screen and listens. For dictation, three habits help: say "switch to dictation mode" before long passages so nothing is taken as a command, turn on automatic punctuation from the bar's settings, and use "voice access sleep" rather than closing it when you pause, so waking it is instant.

Voice access has real limits as a dictation tool, and they explain its middling "reach" score. It speaks seven languages. It runs only on Windows 11 22H2 or later. It is designed around commanding the PC, so in its default mode it is always deciding whether you meant a command, and its bar is fixed at the top of your main screen. And it is the only Microsoft route where you have to start a separate app; if voice access is running and you press Windows+H, Windows tells you that you "can use voice access for typing too". For people who need full voice control of the PC, those are features. For people who just want offline dictation, they are friction.

That is the gap the on-device apps in the scorecard fill. Superwhisper says its Windows app uses "on-device AI models" for offline dictation in any app on Windows 10 and 11 - Superwhisper. Willow Voice offers an offline mode "powered entirely by your device" - Willow. And Yap runs NVIDIA's Parakeet v3 model on the PC's processor through the open-source sherpa-onnx runtime - sherpa-onnx on GitHub, using at most four threads so the app you are typing into stays responsive; the model downloads once, about 0.7 GB, and after that recognition never touches the network. We measured that exact Windows model and runtime on the same 60 public recordings we use for every engine decision, against the Whisper Base model our app used before:

The recordings are 20 English audiobook clips from LibriSpeech's harder test set, 20 Dutch clips from Google's FLEURS, and 20 meeting clips from the AMI corpus, chosen before any model ran and scored identically, with punctuation and letter case ignored. We ran the Windows engine's exact model files and runtime on a Mac's processor rather than on a PC, so the numbers describe the engine, not any particular laptop's speed. They are small samples, about eight minutes of audio in total, and they say nothing about your microphone or accent, which is why the ten-minute test in our buyer's guide is still the step that decides.

Whatever you choose, verify offline behavior yourself rather than trusting a feature list. Turn off Wi-Fi, open a text editor, and dictate a sentence. If text appears, recognition is happening on the PC; if the app shows an error or waits, it depends on the network. Then check the two other places your words can travel even in an on-device app: AI rewriting features, which often call a cloud model, and history sync, which may upload transcripts for cross-device access. Our guide to offline dictation on a Mac covers the same test from the Mac side.

Why this matters: offline dictation is the one capability Windows' default voice typing cannot provide at any price, and it is the only way to be certain your speech never leaves the computer. How to apply it: if you are on Windows 11 and write in one of voice access's seven languages, set it up once and try dictation mode for a day; if you need Windows 10 support, a hold-to-talk shortcut, more languages or a recoverable history, try an on-device app, and run the airplane-mode test before you trust any of them with sensitive work.

8. Privacy: what Windows sends, and how to change it

Dictation is the most revealing way to give a computer text. A recording carries your voice, the room you are in and anyone speaking near you, and it captures everything you say while the microphone is open, including the sentence you abandon. So the privacy question for dictation is not abstract: it is which company receives that audio, whether it is kept, and whether it can be used to train models. On Windows, the answer depends entirely on the route, and Microsoft documents each one more clearly than most app vendors do.

For voice typing, the relevant setting is called online speech recognition. Microsoft explains that it "uses Microsoft cloud-based services for more accurate recognition", that "voice data is sent to Microsoft only to provide the service and create text transcriptions", and that "Microsoft does not store, sample, or listen to voice recordings without your permission" - Microsoft Support. You control it under Start, Settings, Privacy & security, Speech. Turning it off is a real choice with a real cost: Microsoft notes that "only device-based features (such as Narrator) are available" when it is off, which means voice typing stops working. Here is where each Microsoft route stands:

RouteWhere speech is processedWhat Microsoft says about keeping it
Voice typingOnline, Azure Speech servicesNot stored, sampled or listened to without your permission
Fluid dictationCleanup on the device, after online recognitionDescribed as "fast and private processing" on the PC
Voice accessOn the device"Your data isn't shared to the cloud"
Microsoft 365 DictateOnline"Does not store your audio data or transcribed text"

The permission Microsoft refers to is voice clip contribution, an opt-in program in which samples of your speech help improve the service. Microsoft states that "you can use voice typing without contributing voice clips" and that "contributing voice clips is optional"; the links to start or stop contributing sit in the voice typing settings panel, the same panel shown in section 2. If you use a work or school account, Microsoft adds that "sharing voice data isn't available, but you can still use voice typing", so on many company laptops the choice is already made for you.

Third-party apps answer the same questions in their own terms, and the answers vary more than the marketing suggests. Wispr Flow says that "transcription always occurs on the cloud" and that, if you allow data sharing, your "audio, transcript, edits" may be used to "evaluate, train, or improve AI models" - Wispr Flow. Aqua Voice processes audio in the cloud and may store transcript data for users "with Privacy Mode disabled" - Aqua Voice. Google Docs voice typing hands the question to your browser, which "controls the speech-to-text service". And on-device apps such as Yap keep both recognition and history on the PC, so there is nothing to send.

There is a second stop that people forget, and it matters more each month as dictation becomes the way people talk to AI tools. When you dictate a prompt into an AI chat window or a coding assistant, the speech service is only the first company to see your words; the text then goes to the AI tool, under that tool's own data terms. For coding assistants in particular, Founden's 20-minute opt-out audit for AI coding tools shows where each tool's training switch lives, including GitHub Copilot's, which it reports has been on by default for individual plans since April 24. Founden and Yap have the same founder, and we build this site on Founden.

Why this matters: dictation privacy is decided by architecture first and by settings second, and on Windows the two built-in routes sit at opposite ends: voice typing is online by design, voice access is on-device by design. How to apply it: check Settings, Privacy & security, Speech to see whether online speech recognition is on, look at the voice clips links in the voice typing panel, and decide which route your work allows. If you dictate into AI tools, settle that tool's training setting as well, because your spoken prompt ends up there either way.

9. Dictating in Word, Outlook and Google Docs

System-wide dictation types wherever your cursor is, but it knows nothing about the document it is typing into. In-app dictation is the opposite trade: it only works inside one program, and in return it can do things a system-wide tool cannot, such as applying bold, creating a list or adding a comment by voice. If most of what you write lives in Word, Outlook or Google Docs, the dictation inside those programs deserves a serious try before you look anywhere else.

In Word, you open a document and select Home, Dictate "while signed into Microsoft 365 on a mic-enabled device", wait for the Dictate button to turn on, and start speaking. Microsoft is explicit that "the dictation feature is only available to Microsoft 365 subscribers", and that it needs "a microphone and reliable internet connection" - Microsoft Support. The command set is the richest of any Microsoft route: "delete from [x] to [y]", "select that", "bold that", "align center", commands to create lists and add comments, plus spoken punctuation in each supported language. And Microsoft states that the service "does not store your audio data or transcribed text".

Outlook works the same way. In the new Outlook, you go to the Message tab and select Dictate; in classic Outlook, you select the message body, then Dictate on the Message tab while signed in to Microsoft 365 - Microsoft Support. To resume after a pause, Microsoft gives the shortcut Alt + ` (the backtick key) or the microphone icon in the ribbon. The dictation settings, behind a gear icon, let you change the spoken language and microphone, toggle auto punctuation where the language supports it, and set a profanity filter that masks sensitive phrases with asterisks.

Google Docs offers the same idea for free, inside the browser. You open a document and click Tools, then Voice typing, and a microphone box appears - Google Docs Editors Help. It works in the latest versions of Chrome, Edge and Safari, and it lists 118 languages and accents, the most of any tool in this guide. Two caveats: voice commands are "available only in English", with both account and document language set to English, and Google says "your web browser controls the speech-to-text service", so what happens to your audio depends on the browser rather than on Docs. If you make a mistake, you can fix it without stopping, and words underlined in gray offer alternatives with a right-click.

When does in-app dictation beat a system-wide tool? When formatting is part of the writing: a report with headings and lists, a long email with bolded action items, a document you review with comments. When does it lose? As soon as your writing crosses programs. Dictate a reply in Slack, a search in the browser and a note in a to-do app, and in-app tools cover none of it, while Windows+H or a dictation app covers all three. Many people end up using both: Microsoft 365 Dictate inside long Word documents, and a system-wide tool everywhere else.

If Word's Dictate button is missing or silent, Microsoft's troubleshooting page covers the usual causes: a missing Microsoft 365 subscription, a Windows version older than Windows 10, a muted or unplugged microphone, a room that is too noisy, or "We don't have access to your microphone", which often means another application is holding the microphone - Microsoft Support. Close the meeting app that still has the microphone open, and try again.

Why this matters: the best dictation tool for a document-heavy job is often the one already inside the document, and many Microsoft 365 subscribers pay for a feature they have never opened. How to apply it: if you have Microsoft 365, dictate your next long document with Home, Dictate and learn three formatting commands; if you live in Google Docs, try Tools, Voice typing in Chrome; and keep a system-wide tool for everything outside those two programs.

10. Windows 10 versus Windows 11

Windows 10 is officially finished, but it is far from gone. Microsoft ended support on October 14, 2025, and says Windows 10 PCs "will continue to function" but no longer receive security updates unless they are enrolled in Extended Security Updates, which run through October 12, 2027 - Microsoft. Nearly a year later, StatCounter still counts more than a quarter of Windows desktops on Windows 10, and the chart shows how gradual the move has been:

Windows 11 overtook Windows 10 between June and July 2025, and by September 2026 it held about 71% of Windows desktops against about 28% for Windows 10. For dictation, that remaining quarter matters, because Windows 10 users get a noticeably older experience. Pressing Windows key + H on Windows 10 opens "dictation" rather than voice typing, a small toolbar that listens in whatever text box you are in. Microsoft's own screenshot shows it in the old Mail app:

The differences are substantial. Windows 10 dictation also needs an internet connection; its language list is US English plus 12 other entries rather than 46; its editing commands, while rich, work in US English only; and two of the features this guide recommends most are missing entirely. Voice access is a Windows 11 feature, available from version 22H2, and fluid dictation requires a Copilot+ PC, which by definition runs Windows 11. In other words, a Windows 10 user who wants offline dictation has no built-in option at all.

Third-party tools fill that gap, because most of them still support Windows 10. Yap, Superwhisper and Aqua Voice all state support for both Windows 10 and Windows 11, and Dragon Professional 16 is "optimized for Windows 11 and backwards-compatible to Windows 10" - Nuance. Microsoft 365 Dictate also works on Windows 10, since Microsoft lists "Windows 10 and above" as the requirement. The bigger question for a Windows 10 user is not dictation but security: a PC that no longer receives updates is a poor place to keep anything sensitive, dictated or typed.

If you are not sure which version you have, Microsoft's instructions take ten seconds: right-click Start, choose Settings, select System and then About, and look under Windows specifications for the edition and version - Microsoft Support. The same page lists Device specifications, including the processor type and the amount of RAM installed, which answers two other questions from this guide at once: whether your PC has the 16 GB of RAM a Copilot+ PC needs, and whether it runs on a standard x64 processor or an Arm chip, which matters whenever an app offers separate installers for each.

Why this matters: a guide written only for Windows 11 would fail more than a quarter of Windows desktops, and the gap is widest exactly where people need it most, offline dictation. How to apply it: if you are still on Windows 10, use Windows+H for quick online dictation, consider an app with Windows 10 support for offline use, and plan the move to Windows 11 or Extended Security Updates; if you are on Windows 11, check that you are on version 22H2 or later so voice access is available.

11. Troubleshooting: when voice typing is not working

Most voice typing failures fall into a handful of causes, and Microsoft documents the common ones on a dedicated page. The good news is that almost none of them are recognition failures: they are permissions, microphone selection, language settings or the network. Working through them in order, from the cheapest check to the most involved, fixes the large majority of "voice typing isn't working" problems in a few minutes, without reinstalling anything.

Start with the error message, if you see one, because voice typing names its two most common problems directly. "Voice typing needs access to your microphone" means Windows is blocking the microphone; Microsoft's fix is Start, Settings, Privacy & security, Microphone, and making sure Microphone access is on. "Voice typing isn't available in the current language" means your input language is not on the supported list; install a supported language or switch to one with Windows + Spacebar - Microsoft Support. The table covers the rest:

SymptomLikely causeWhat to do
Windows + H does nothingNo text box has focusClick into a text field first, then press the shortcut
A message about voice accessVoice access is already runningDictate through voice access, or close it first
"Listening" but no text appearsWrong or quiet microphonePick the microphone under Sound, Input, and raise its input level
Words come out wrongNoise, distance or the wrong languageMove somewhere quieter, use a headset, check Windows + Spacebar
Voice typing stopped working entirelyOnline speech recognition or the network is offTurn on online speech recognition under Privacy & security, Speech, and check the connection

The microphone row deserves the most attention, because it is the most common cause and the easiest to misdiagnose. Microsoft's microphone guide walks through the check: open Settings, System, Sound, select your microphone under Input, open its properties, and select Start test, then speak normally and play back the sample, adjusting the input volume slider if needed - Microsoft Support. A laptop docked to a monitor often switches to the monitor's or webcam's microphone without telling you, and the voice typing settings panel has its own default microphone choice that can differ from the system default. If the built-in microphone stays poor, Microsoft's advice in several places is the same: try a headset or an external microphone.

The last row is the one people rarely suspect. Voice typing depends on the online speech recognition setting, and when it is off, Microsoft says only device-based features remain. Privacy-minded users sometimes turn it off months earlier and forget. The same row covers captive Wi-Fi portals in hotels and on trains, where the connection can look fine while Microsoft's speech service cannot be reached until you sign in to the network. If you need dictation in those places, that is a sign you need an offline route rather than a fix.

Three smaller cases round out the list. If the fluid dictation option is dimmed on a Copilot+ PC, the on-device models are still downloading; wait for the notification. If Word's Dictate button is missing, check that you are signed in with an active Microsoft 365 subscription. And if a third-party dictation app types everywhere except one program, check whether that program is running as administrator: Windows only lets an application send simulated input to applications "at an equal or lesser integrity level" - Microsoft Learn, so an app running with normal rights cannot type into an elevated window. Mac users will find the equivalent checklist in our guide to Mac dictation not working.

Why this matters: people abandon dictation after a failure they could have fixed in a minute, usually a permission or microphone setting, and conclude that "Windows dictation doesn't work". How to apply it: keep this order: text box, error message, microphone permission, microphone choice and level, language, online speech recognition and network. If all of those check out and accuracy is still poor, the problem is the environment or the engine, and a headset or a different tool is the next step.

12. When a dedicated dictation app is worth it

The built-in routes are free, and for occasional dictation they are enough. A dedicated app earns its place when one of Windows' structural limits gets in your way every day, not when it has a longer feature list. Reasoned from the gaps this guide has found, the cases are specific: you need dictation offline on Windows 10 or in a language voice access does not cover; you switch languages constantly and want each recording detected automatically; you want the same shortcut and behavior in every app, with a history to recover text when insertion fails; you want AI formatting that rewrites rambling speech into clean prose; or you work across a Mac, a Windows PC and a phone and want one tool on all of them.

Different apps answer different gaps, which is why the scorecard has no single winner for everyone. Superwhisper and Yap answer the offline gap with on-device models; Wispr Flow, Aqua Voice and Willow Voice answer the formatting and cross-device gaps with cloud models; and Dragon Professional, long the professional standard on Windows, remains available for Windows 10 and 11, though its page shows no price and asks you to contact the company. If you are weighing the cloud apps specifically, our comparisons with Wispr Flow and Aqua Voice go through them point by point, and our page on Dragon for Mac covers what happened to Dragon on the other platform. For the paid plans with yearly billing, here is what a year costs:

Prices alone mislead in both directions. Superwhisper is missing from the chart because its free tier already includes unlimited dictation on local Whisper models, with Pro at $8.49 a month - Superwhisper, and both Wispr Flow and Aqua Voice offer limited free tiers: 2,000 words a week and 1,000 words in total, respectively. On the other side, a cheap app that mishears you costs more in corrections than an expensive one that does not, which is why the accuracy test matters more than the price chart. Whatever you consider, use the trial period to dictate real work, not a demo paragraph.

Since we make one of these apps, here is exactly what Yap does on Windows and what it does not. You hold Ctrl + Win, speak, and release, and your words are typed where your cursor was; pressing Space while holding the keys keeps recording hands-free. Speech is recognized on the PC by NVIDIA's Parakeet model in 25 languages, each recording in the language you spoke, and audio is neither uploaded nor saved. Typing stops if you switch apps, open a menu or land in a password field, and your words stay in a local history. Its own settings screen shows the defaults:

What it does not do matters just as much. Yap has no AI rewriting, no custom vocabulary yet, and no free tier: it is free for its first 3 days with a free account and no card, then costs $6.90 a month or $69 a year after a 10-day trial that requires a card - Yap pricing. It runs on Windows 10 and 11, on both standard x64 PCs and Arm laptops, and the same subscription covers a Mac. If those trade-offs fit, the Yap for Windows page has the details and the download page the installers; if they do not, Superwhisper or voice access may serve you better, and the scorecard says so.

Why this matters: paying for a dictation app only makes sense when it removes a limit you hit every day, and the limits differ by person: offline, languages, formatting or devices. How to apply it: name the one Windows limit that bothers you most, pick the two apps that answer it, and run both through a week of real work before you subscribe to either.

13. A decision framework

Most of the choice comes down to four questions asked in order, because each answer removes options that later answers cannot bring back. Privacy comes first, because it is a constraint of your work rather than a taste: if speech must stay on the PC, voice typing and every cloud app are out before features enter the picture. Where you write comes second, because Microsoft 365 Dictate is excellent inside Word and Outlook and useless anywhere else. Devices and rewriting come third, and only then the questions of accuracy and comfort, which you settle by testing rather than reading.

The diagram turns those questions into a short list. Treat each end point as the one to three options to try for a few days, not as a verdict, because the final decision depends on your voice, your microphone and your rooms, which no table can see.

A few situations show how it plays out. Someone who dictates a few emails a day on Windows 11 should simply turn on automatic punctuation and use Windows + H; nothing else in this guide will save them enough time to matter. A lawyer, clinician or consultant whose work cannot leave the machine should start at voice access if their language is covered, and move to an on-device app if they need Windows 10, more languages or a recoverable history. A writer who lives in Word with a Microsoft 365 subscription should learn Dictate's formatting commands before buying anything.

Three more cases are common enough to name. A multilingual writer should test automatic language detection, because the switching cost of Windows + Spacebar adds up quickly, and should check each language against the tool's list, where Google Docs' 118 entries cover the most ground. A Copilot+ PC owner should try fluid dictation first, since it brings on-device cleanup at no cost. And a developer who dictates prompts to AI coding tools should pick for reach and speed, then settle the coding tool's own training setting, because the prompt ends up there whichever dictation route carried it.

Anyone on a tight budget has several free routes on Windows. Voice typing, voice access and Google Docs voice typing cost nothing; Superwhisper's free tier includes unlimited dictation on local Whisper models; and Willow Voice offers unlimited free dictation on what it calls a weaker model - Willow. Exhaust those first. If a free route recognizes your voice well, the only reasons left to pay are offline use, a history, automatic language switching and convenience, and a free trial is enough to judge whether those are worth money to you.

Why this matters: a structured choice prevents the most common mistake in this category, choosing by the most visible feature and discovering the deal-breaker a month later, usually offline use, languages or where the text is allowed to go. How to apply it: answer the four questions honestly, take the options at your end point, and spend twenty minutes dictating the same real paragraph into each of them before you decide.

Conclusion

Dictating on Windows is easy to start and surprisingly easy to get wrong, because "Windows dictation" is really four different systems. Voice typing (Windows + H) is the fastest way in: free, in any text box, 46 language entries on Windows 11, and dependent on Microsoft's online service for every sentence. Voice access is the hidden gem: free, fully on-device after a one-time download, with the best editing commands Microsoft offers, but limited to Windows 11 and seven languages. Microsoft 365 Dictate is the best tool inside Word and Outlook and nowhere else. And Windows 10's older dictation still serves more than a quarter of Windows desktops, online only.

The order of decisions matters more than any single feature. Decide first whether your speech may leave the PC; then where you write; then whether you need phones or AI rewriting; and only then compare accuracy, by dictating your own real text into two or three candidates. Our scorecard puts Superwhisper first and voice access second for Windows in 2026, with Yap third and voice typing close behind, and the honest summary is that many people need nothing beyond the free tools, configured properly.

If you do need more, the reason is usually offline use, languages or consistency across apps, and that is where an app earns its price. Yap is one option for on-device dictation on Windows 10 and 11, free for its first 3 days with a free account and no card, from the download page; Superwhisper and the cloud apps are others, and the scorecard and decision framework above are there to help you choose between them on evidence rather than marketing.

This guide reflects Windows dictation features, apps, prices and data policies as of October 6, 2026. Microsoft updates voice typing, voice access and fluid dictation through Windows updates, and vendors change plans and policies often, so check the linked pages before you rely on a detail.