How To Enable On-Device Dictation in Chrome

This page uses Chrome’s Web Speech API. It can run in two modes: Local audio never leaves your computer — transcription happens on-device using a model Chrome downloads once. Cloud audio is sent to Google’s servers for transcription (the historical default).

Requirements for local mode

• Google Chrome 138 or newer on Windows, macOS, or Linux. (ChromeOS is not yet wired up to the on-device Web Speech API.) Check your version at chrome://version.
• A one-time language pack download (a few hundred MB). The page will offer this on the first click.
• Microphone permission for this site.

Step-by-step

1. Click Start Dictation below. If you see the button labeled “Download local model & start”, click it — Chrome will download the English language pack. This can take 30 seconds to several minutes depending on your connection. The status line shows progress.
2. When prompted, allow microphone access for this site.
3. Once the model is installed, the button will say “Start Dictation (local)” and you’re set. Future visits skip the download entirely.
4. Speak naturally. Click Stop Dictation when finished.

Verifying the local model installed

1. Open a new tab and go to chrome://components.
2. Look for an entry called “Speech Recognition” or “SODA en-US Models”.
3. The version should be something like 1.x.x.x. If it shows 0.0.0.0, click Check for update, then reload this page.

Troubleshooting

• Status says “Cloud mode” — your Chrome may be older than 138, or you’re on ChromeOS. Update Chrome and reload.
• “Microphone permission denied” — click the lock icon in the address bar → Site settings → Microphone → Allow.
• Install seemed to succeed but recognition fails with “language-not-supported” — Chrome occasionally needs a full restart after the first SODA download. Quit Chrome completely and reopen.
• Nothing happens after clicking — open DevTools (F12), check the Console for messages prefixed [stt] and any red errors.

Privacy note

In local mode, no audio is transmitted off your device — it’s processed entirely in your browser using the downloaded model. In cloud mode (the fallback), audio is sent to Google’s speech recognition service. The status indicator next to the button always shows which mode is active.