By Denys MedvedievSetup guide
Speech to text in Word, properly
Word has had built-in Dictate since 2017. Most people still walk past it. Here is where it is, how to turn it on, when it runs out of road, and what to reach for when it does.
Last updated: May 2026
Microsoft Word's built-in speech-to-text is called Dictate — Home tab, far-right corner. It is free in Word for the web at office.com, included with Microsoft 365 on Word desktop, and works in 15 fully supported languages plus around 33 in preview. It needs an internet connection. Three situations make it the wrong tool: you are offline, you want to dictate outside Word, or your language is not on Microsoft's list. In those cases, a system-wide tool like Whisper handles the gap — Cloud (OpenAI BYOK), local Parakeet (English + 24 EU languages), or local Whisper (99 languages). One hotkey, any app.
Speech to text in Word, the short version
Speech to text in Microsoft Word is called Dictate. It lives on the Home tab. On Word desktop you need Microsoft 365 and an internet connection. On Word for the web at office.com the same feature is in the free tier. Click the microphone icon, allow Word to use your mic, and speak. Text appears as you talk.
That paragraph is the answer for about 80% of readers. The rest of this guide is for the other 20%, who hit one of three walls. One: Word desktop without Microsoft 365, where Dictate is not there. Two: dictation that needs to work outside Word — in Outlook, Slack, a browser, your code editor. Three: offline mode, longer sessions, languages Microsoft does not fully list, or work you would rather not send to a server.
If any of those apply, you will want a system-wide tool. The one I make is called Whisper. I will show both options and tell you when each is the right pick.
Turn on Dictate in Word in three steps
Prerequisites. A working microphone, an internet connection, and either Word for the web at office.com (free) or Word desktop signed in with Microsoft 365. Total time: 90 seconds from cold start to first dictated sentence.
Step 1 — Open Word and find the Dictate button. (~15 seconds.)
On Word desktop, open any document and look at the far right edge of the Home tab. The Dictate button is a small microphone icon labelled Dictate. On Word for the web, same icon, same spot. If you do not see it, jump to the troubleshooting section below.
Step 2 — Click Dictate and grant microphone access. (~20 seconds.)
Click the icon. The first time, your browser or operating system will ask if Word can use the microphone. Say yes. The button turns red — it is listening. You can also skip the click with the shortcut: Alt + ` (backtick) on Windows or Option + F1 on Mac.
Step 3 — Speak in short, complete thoughts. (~55 seconds to land your first sentence and fix it.)
Start with one sentence. Wait half a second. Read what appeared. Fix punctuation with your keyboard, or use a voice command (next section). Stop dictating by clicking the button again, hitting Escape, or walking away. Dictate is patient.
Hard parts.
Two things trip up first-timers. Talking too fast in long unbroken sentences: Dictate handles natural pacing but loses the thread on a paragraph delivered without breath. And ambient noise: a washing machine in the next room quietly drops accuracy by ten or fifteen percent. A $20 USB mic does more than any setting will.
Voice commands every Word user should know
Once Dictate is running, you switch from speaking text to speaking commands by saying the phrase out loud. There is no toggle. Dictate listens for commands inline.
Punctuation.
"Period", "comma", "question mark", "exclamation mark", "colon", "semicolon", "open quote", "close quote", "new line", "new paragraph". The last two are the ones I missed for embarrassingly long: I was hitting Enter manually for a week before realising "new paragraph" was a thing.
Editing.
"Delete that" removes the last bit you said. "Delete sentence" or "delete paragraph" do what they say. "Bold that", "italicize that", and "underline that" format the most recent chunk.
Navigation and lists.
"Go to end of paragraph" and "move down a line" do sweeping fixes without a trackpad. "Start list" begins a bulleted list, "next line" adds an item, "exit list" gets you out.
Most people get away with five: "period", "comma", "new paragraph", "question mark", and "delete that". The rest is muscle memory you will build, or look up the one time you need it.
When Word's built-in Dictate is enough (and when it is not)
Dictate is the right tool when three things are true: you have Microsoft 365 (or you are in Word for the web), you have steady internet, and you mostly write inside Word. For most office workers, that is the whole answer.
It stops being the right tool when any of these apply:
Word standalone (Office 2016/2019) without Microsoft 365— Dictate is not in those desktop builds. Word for the web is the workaround.
You want to dictate outside Word too— Outlook, Slack, Notion, VS Code, the browser. Dictate is locked to Microsoft 365 apps.
You need offline work— Dictate needs internet at all times. On a train, plane, or in a hotel with bad latency, it stops.
A language Microsoft does not fully support— only 15 supported plus ~33 in preview. Korean and Polish are preview; Ukrainian is not on Microsoft's list at all. Code-switching mid-sentence makes it worse.
Sensitive work— Microsoft says Dictate does not store audio or text, a fair trust framework, but the audio still leaves your computer. Legal, HR, healthcare: many people would rather not.
If none of those five apply, the article ends here for you. Use Dictate.
If you want the paths weighed against each other in one place, every voice typing option in Word covers Win+H, Dictate, and a system-wide tool side by side.
Using Whisper for speech to text in Word and everywhere else
Whisper is the tool I built and ship for the cases above. It runs on Windows and Mac, sits in the system tray, and listens for a global hotkey. You press the key, you talk, you release. Whatever you said gets pasted at the cursor: in Word, in Outlook, in Slack, in any text field.
The default hotkey on Windows is Ctrl + Space. On Mac it is the Right Option key, used as push-to-talk — hold while you speak, release to stop. Right Option was picked because Cmd + Space opens Spotlight on macOS, and one of us was tired of it interrupting recordings. Both hotkeys are customisable.
The flow inside Word: put your cursor where you want the text. Press the hotkey. Say "draft a polite follow-up to Marc asking for the spec by Friday." Release. Three seconds later, that sentence is in your document. No menu, no ribbon, no sign-in. If you want the text cleaned up by an AI before it lands, you say "Hey whisper" first and the rest of what you say becomes a prompt.
Local mode runs entirely on your laptop. Audio does not leave the machine. The model lives on disk after a one-time download (~140 MB to ~3 GB depending which one you pick). Cloud mode sends audio to OpenAI through your own API key; Remskill is never in the middle.
Pick the right transcription model for your machine
Whisper does not pick for you. Three options sit in Settings, and you choose based on what matters most: hardware, language coverage, or quality.
Cloud mode (OpenAI BYOK).
Two transcription tiers: gpt-4o-mini-transcribe (fast, low per-minute cost) and gpt-4o-transcribe (higher quality). Runs on any hardware; it is just network. You bring your own OpenAI key, the cost lands directly with OpenAI. Pick this for the highest accuracy and web search in one tool.
Local — Parakeet (NVIDIA TDT, ~600 MB).
Fastest local option. Five to ten times faster than Whisper on CPU. English plus 24 European languages: Spanish, French, German, Italian, Polish, Ukrainian, the rest of the European set. No translate-to-English. Pick this if you speak English (or one of the 24 European languages) and want speed.
Local — Whisper (eight model sizes).
Slower than Parakeet, but the multilingual variants cover 99 languages and support translate-to-English. Pick this for translation, less-common languages, or finer control like custom vocabulary and beam-size tuning.
Rough hardware map: an M1/M2 Mac with 16 GB handles any of them, including Whisper Large v3 at ~3 GB. A 2018 Intel MacBook with 8 GB runs Cloud, Parakeet, or the smaller Whisper builds; skip Medium and Large. Windows with 16 GB and a discrete GPU runs Large v3 and Turbo well. If in doubt, start with Whisper Small: it is where most people land.
Speech to text in Word on Mac, mobile, and the web
On Mac.
Word for Mac with Microsoft 365 has the same Dictate button on the Home tab; the shortcut is Option + F1. macOS also has its own built-in dictation that works in every app, including Word. Start it from the Microphone function key, the menu bar (Edit > Start Dictation), or a custom keyboard shortcut. There is no length cap; it stops on 30 seconds of silence or when you press Escape. The voice to text on Mac walkthrough covers both Apple's option and the third-party route.
On mobile.
Word for iOS and Android both have Dictate in the toolbar, leaning on the OS native voice input: Siri on iOS, Google voice typing on Android. Phone quality is usually below desktop because the noise floor is higher.
On the web.
office.com gives you Word for the web free, with Dictate in the free tier. Cleanest free path if you do not have Microsoft 365 on desktop. Runs in any modern browser, saves to OneDrive.
For one consistent dictation experience across Word, Outlook, Slack, and the browser, you want a system-wide tool — one hotkey everywhere. The Windows walkthrough covers that setup end-to-end.
Languages, accents, and the 99-language question
Microsoft Word Dictate's language list splits into two buckets. Supported: 15 languages, including English in six regional variants, French in two, German, two Spanish variants, Italian, Japanese, Hindi, Brazilian Portuguese, and Mandarin. Preview: another ~33 — Arabic variants, Bulgarian, Czech, Dutch, Greek, Hebrew, Hungarian, Korean, Polish, Russian, Swedish, Thai, Turkish, Vietnamese, and more. Preview languages work, but Microsoft warns accuracy is lower and command coverage thinner.
Whisper works differently. The multilingual variants (Small, Medium, Large v3, and Large v3 Turbo) cover 99 languages. The English-optimised variants (Base, Small, Medium, and the distilled Turbo English) are English only; selecting one locks the language picker to English. Parakeet sits between: English plus 24 European languages. OpenAI's cloud transcription models cover the same broad language inventory as Whisper, with improved accuracy.
Practical answer: if Microsoft fully supports your language, Word Dictate is fine. If your language is in Microsoft's preview tier — Russian, Hebrew, Korean, Polish, Czech, Hungarian, most of Eastern Europe — Whisper's multilingual large model is usually more accurate. If your language is not on Microsoft's list at all (Ukrainian is one notable absence), Whisper or Parakeet is the only path. For code-switching mid-sentence, Whisper Large v3 handles it without a manual toggle.
Heavier regional accents cost a few percent of accuracy in every engine. The fix is not a different tool; it is a better microphone.
Fixing dictation when the Dictate button is missing or grayed out
Three problems, three specific fixes from Microsoft's own page.
Dictate button missing.
Two checks. First, confirm you are signed into Word with an active Microsoft 365 subscription — Dictate on desktop is gated behind that. File > Account shows your sign-in state. Second, confirm you are on Windows 10 or above. Older Windows, and the standalone Office 2016/2019 builds, do not ship Dictate on desktop.
Dictate button grayed out.
The document is in Read-Only mode. Look at the top of the window for "Read-Only" or "Protected View", click "Enable Editing" in the yellow bar, and the button comes back.
Microphone problems.
Dictate is on, but nothing happens when you speak. Three things to check, in order. First, OS permission: Windows Settings > Privacy & Security > Microphone, with "Let apps access your microphone" on and Word in the list. Second, the input meter — does it move when you speak? Third, the right input device, if you have several (laptop mic, USB, Bluetooth headset). The wrong mic is the most common culprit.
Accuracy issues are almost always audio in, not the model. A built-in laptop mic 60 cm away with a fan in the background is rough conditions. Move closer, kill the fan, or spend $20 on a USB mic. That mic does more than any setting will.
If Word's Dictate button works but Win + H is dead in your other apps, that's a Windows-wide problem, not a Word one — eight verified fixes for Windows voice typing that stopped working walks through Online Speech Recognition, mic permission, and the Group Policy bug.
For a broader look at accuracy claims across the whole market, see our guide to picking transcription software — it covers where the accuracy numbers actually come from.
If your trouble is specifically the Dictate button itself — missing, grayed out, or throwing a microphone error — fix Word dictation not working covers the Microsoft 365 requirement and every button-level cause.
Privacy: where your voice goes
Three trust models, three different answers.
Word Dictate.
Audio is sent to Microsoft's servers for processing. Microsoft states the service does not store audio or transcribed text: utterances are sent only to provide text results. Fair statement from a vendor with a public trust framework, but the audio still leaves your computer. Worth reading if your workplace has policies against cloud transcription.
Whisper local mode.
Audio is processed on your computer using a model you have downloaded locally. Nothing reaches any server: not us, not OpenAI. The model lives on disk after the one-time download; the laptop is the whole stack. Right choice for legal drafts, healthcare notes, internal HR.
Whisper cloud mode (BYOK).
Audio goes directly from your computer to OpenAI through your own API key. Remskill is never in the middle. We never see the audio, never see the transcript. Trust relationship is between you and OpenAI directly.
The boring truth is that for most everyday writing, Word Dictate's trust model is fine. For work where "audio leaves the laptop" is a hard no, local mode in a system-wide tool is the answer.
Pricing, if you decide to go beyond Word's built-in option
Microsoft Word Dictate is included with Microsoft 365 on desktop. If you already pay for Microsoft 365 for the rest of Office, there is no extra cost. Word for the web at office.com is free, Dictate included.
Whisper is free for the entire local pipeline once you create an account — Parakeet, Whisper local models, AI text enhancement via Ollama, history, presets. No payment method required at signup. The Cloud surface (OpenAI cloud transcription, Cloud AI enhancement, web search) is the paid layer, and the upgrade flow includes a 7-day Cloud trial. Numbers and plan details live at Whisper pricing page. I would rather link there than quote them here — those details belong in one canonical place, not scattered across a dozen articles.
Want to see it on your computer?
Download Whisper, hold the hotkey, watch the transcript appear — in Word, Outlook, Slack, or any text field you are already in.
Free for all local features. No card required.
