Voice to text not working on Windows?

First check: is your mic unmuted and Online Speech Recognition turned on?

Before any of the eight fixes, run three thirty-second checks.

One: confirm the mic is not muted. Many laptop keyboards have a physical mic-mute key; headset mics often have an inline switch. Open Settings > System > Sound > Input, pick your device, speak. The input level bar should bounce.

Two: confirm Online Speech Recognition is on. Windows 11: Settings > Privacy & security > Speech. Windows 10: Settings > Privacy > Speech. When this is off, voice typing does not work; only device-based features like Narrator remain. Single most-missed fix on the SERP.

Three: confirm your cursor is inside a text box. Microsoft lists three prerequisites for Win+H: internet, working microphone, cursor in a text box. Press Win+H on the desktop and the flyout opens, listens, types nothing, because there is nothing to type into.

If those three are clean and Win+H still does nothing, move to Fix 1.

Fix 1: Toggle Online Speech Recognition off, wait 10 seconds, toggle it back on

The boring fix works on most Windows PCs. Same panel as the first check: Settings > Privacy & security > Speech on Windows 11, Settings > Privacy > Speech on Windows 10. Toggle Online Speech Recognition off. Count to ten. Toggle it back on.

Why ten seconds? Windows needs a moment to deregister the cloud-speech endpoint and tear down the local recognition pipeline before it rebuilds them. Flip too fast and you land in the same stuck state. Test in Notepad first; if Win+H works there but breaks in Word or your browser, the problem is per-app microphone permission.

Fix 1 leads the list because Google's AI Overview surfaces it first, Microsoft's own troubleshooting page omits it, and the two best-ranking competitor articles both miss it. If you only read one fix here, read this one.

Fix 2: Microphone permission is the problem (not your mic)

When Win+H works in Notepad but breaks in Word, Chrome, Slack, or Teams, the issue is per-app microphone permission. Windows treats voice typing as a feature inside the focused app, so that app needs mic access on its own.

Start > Settings > Privacy & security > Microphone. Three toggles to verify: Microphone access for the whole device (must be on), Let apps access your microphone (must be on), and the catch-all Let desktop apps access your microphone further down. Win+H falls under the catch-all for many third-party desktop apps.

Microsoft documents this as the named error "Voice typing needs access to your microphone". Same fix with or without the error string. Resolves roughly half of "Win+H stopped working" support threads I have read.

Fix 3: Windows picked the wrong microphone again

Microsoft leads its troubleshooting page with "make sure the microphone you want to use is selected in Settings" for a reason. Windows switches the default input device when a Bluetooth headset disconnects, when you plug in a USB mic, or when a cumulative update reorders the device list.

Start > Settings > System > Sound > Input. Pick the device you want from the dropdown. Speak and watch the input-level meter; it should bounce. Flat meter, dead device, pick another. If it bounces but Win+H still hears nothing, drag the input volume to roughly 75% and retry.

A side note: a twenty-dollar USB mic does more for transcription accuracy than any software fix. I have spent five years writing this off as snobbery. I was wrong.

Fix 4: The voice typing bar's gear icon — pick your mic explicitly

This one is the underdocumented win. Press Win+H, the voice typing pill appears, and on its right edge is a gear icon. The gear exposes Voice typing launcher, Automatic punctuation, Filter profanity, and Default microphone. The Default microphone item is the catch.

Even when Settings > System > Sound > Input is set correctly, the flyout sometimes points at a different mic — usually whichever device was active when you first invoked Win+H after a reboot. Click the gear, click Select default microphone, pick the device you want.

Only this fix lives inside the voice typing flyout itself. Google's AI Overview surfaces it as step 3; Microsoft's troubleshooting page does not mention it. If Fixes 1 through 3 did not stick, this is usually the one.

Fix 5: Restart Windows Audio and Windows Audio Endpoint Builder (the services.msc fix)

When the flyout opens, the mic is selected correctly, but no audio reaches the recogniser, the audio stack itself is wedged. Two services run the Windows audio subsystem: Windows Audio (internal name AudioSrv) and Windows Audio Endpoint Builder (AudioEndpointBuilder). Restarting them clears the wedge.

Press Win + R, type services.msc, Enter. Scroll to Windows Audio Endpoint Builder first; it manages the audio device list, so other services depend on it. Right-click, Restart. Windows will warn that Windows Audio depends on it and will restart with it; let it. When both show Running, close services.msc, try Win+H.

Google's AI Overview surfaces this as step 4; Microsoft's troubleshooting page does not mention it.

Fix 6: The "speech service are managed by your organization" bug (Group Policy)

If you see the exact error "Voice typing is not available — Speech service are managed by your organization" — Microsoft's verbatim wording, grammatical "are" included — this is the Group Policy bug. Shows up on personal Windows 11 PCs after 22H2 cumulative updates or after running a privacy-tweak script that flips Group Policy keys for speech access.

The safe path uses the Group Policy Editor. Win + R, type gpedit.msc, Enter. Navigate to Computer Configuration > Administrative Templates > Control Panel > Regional and Language Options. Find Allow users to enable online speech recognition services, set to Enabled or Not Configured, OK, close, reboot, not just sign out. The policy is only re-read at boot.

Caution for Windows Home: gpedit.msc is not included. The Microsoft Q&A fallback is a reg delete on the HKLM and HKCU Software\Policies subtrees as administrator. Fine on a personal PC, never on a work or school-managed device. Full sequences at Microsoft Q&A threads 3884630 and 4266488.

If gpedit.msc shows the policy already set to Not Configured and the banner still appears, double-check that your display language and speech language match at Settings > Time & language > Language & region and > Speech — a mismatch can occasionally surface the same banner. Set both to the same locale, reboot, retry.

Fix 7: Update or roll back your microphone driver

Windows-specific. Audio chip drivers are vendor-supplied (Realtek, Conexant, Intel Smart Sound), and one bad cumulative update is enough to break the mic for one user out of a thousand.

Win + X > Device Manager. Expand Sound, video and game controllers. Right-click your audio device, Update driver > Search automatically for drivers. If Windows says you have the latest, try the laptop manufacturer's support site. Dell, Lenovo, HP, Surface all ship audio bundles outside Windows Update.

If the breakage started right after a system update, roll back instead. Right-click the device, Properties, Driver tab, Roll Back Driver. Grey button means no cached driver; download the previous version manually. Reboot either way.

Fix 8: Run the built-in Speech troubleshooter (if you're on Windows 11 22H2 or older)

Microsoft is removing the legacy Speech troubleshooter, along with Keyboard, Power, and Hardware and Devices, in upcoming Windows 11 releases. The deprecation page lists Speech in the removal-only category, no Get Help replacement planned. On Windows 11 22H2 or older you will find it at Settings > System > Troubleshoot > Other troubleshooters > Speech. On 24H2 or 25H2 builds you may not.

If it is there, run Recording Audio first; it catches device-level issues the Speech-specific one cannot. Then run Speech. The wizard walks mic checks, language packs, and the same Online Speech Recognition toggle Fix 1 covered.

The troubleshooter is a wrapper around the previous seven fixes. Useful as a sanity-check; not where to start.

When the fixes don't stick: why Win+H will keep breaking

After the second or third time you fix this, you notice the pattern: voice typing runs for a week or two, then stops. That is not your imagination. It is the architecture.

Windows voice typing depends on five subsystems that drift on their own schedules: the Windows Audio service pair, the per-app microphone permission table, the Group Policy speech-services switch, the Online Speech Recognition cloud toggle, and the vendor audio driver. Throw in cumulative updates that reset privacy toggles, OEM driver bundles that override Microsoft defaults, and the ongoing deprecation of the troubleshooter wizard, and you have a feature that breaks every few months because at least three different Microsoft teams own pieces of it. The sequence above will get you running. It will not stop the breakage from coming back.

Hitting the same wall inside a browser tab instead of the desktop? Here is Google Docs voice typing not working, fixed step by step.

Hitting the same wall inside Microsoft Word, where Dictate is missing or grayed out? Our guide to Word dictation not working explains the Microsoft 365 catch and how to dictate into Word anyway.

When to skip Win+H entirely (and what I'd reach for instead)

Windows voice typing is the right tool for short bursts. For a fifty-word reply or a Teams message, Win+H is free, built in, and covers roughly 43 language/region pairs per support.microsoft.com. If you mostly dictate two-sentence Slack replies and quick search queries, stay on Win+H — installing a third-party app for that is overkill. If your use case is meeting transcription, skip both Win+H and Whisper and use Otter or Fireflies; they are built for that and we are not. We start being worth the install when you find yourself running the same fix sequence every six weeks, or when you need a language Microsoft does not list.

Whisper by Remskill runs locally on Windows 10 and 11 using one of four English Whisper models — Base (~140 MB), Small (~480 MB), Medium (~1.5 GB), Turbo (~1.5 GB, six times faster than Large with about 99% accuracy) — or any of four multilingual variants up to Large v3 (~3 GB) handling 99 languages on the multilingual variants. The .en variants lock the selector to English. Parakeet TDT v3 (~600 MB) is the second engine, 5–10× faster than Whisper on CPU, English plus 24 European languages. Cloud mode is the third path — your own OpenAI key, gpt-4o-mini-transcribe as default.

The default Windows hotkey is Ctrl+Space: hold to record, release to paste at the cursor in any app. The full local pipeline is free for any authenticated user, no card at signup. Whisper Pro adds the OpenAI Cloud surface on top. The installer is an x86_64 build on Cloudflare R2. Try Whisper free on your Windows PC. Every app, no Group Policy collision, no toggle that keeps turning itself off.

If you read this far and Win+H still does not work, the structural answer is in the section above this FAQ: five drifting subsystems and a toggle-off-on dance that keeps coming back. I built Whisper by Remskill partly because I got tired of it. Use Microsoft's voice typing for short bursts; reach for something else when you find yourself fixing the same eight things every other month. My older daughter still thinks the computer is sick. She is not entirely wrong.