Speech to Text

idle

language show interim auto-punctuate

mic level

Heads up: this uses the browser's built-in SpeechRecognition API. On Chrome (desktop and Android) audio is sent to Google's servers for transcription. On Safari (iOS/macOS) audio goes to Apple's servers. The browser does this — the page does not. No audio leaves your device through this page itself, but the browser does call out for the recognition step. If that matters to you, this isn't the right tool. See offline alternatives at the bottom.

Transcript

0 words

Tips

Press Space to start/stop quickly (when buttons aren't focused).
If the page goes to "no-speech" and stops, just hit Start again — that's a built-in timeout.
Background tabs may stop receiving audio. Keep this tab visible.
For best accuracy: speak normally, not slowly. The model is trained on natural speech.

Truly offline alternatives

If you need recognition that never leaves your device, the realistic options are whisper.cpp via WebAssembly (75–500 MB models, 5–20× slower than realtime on phones) or Vosk-Browser (50 MB models, faster but less accurate). Both require a heavier setup than a single HTML file and a long initial download. The browser's built-in API is what almost every "speech to text" web demo actually uses — it's just rare for them to admit it.