VOSK Local Speech-to-Text

Download a VOSK small model into the browser and continuously transcribe speech on-device (offline). Like the LLM viewer, the model stays local and audio is never sent to a server. It is PWA-ready, so you can add it to your iPhone / Android home screen.

1) Choose the recognition language 2) Fetch the model (download locally)

* The first run downloads about 40–50 MB (later runs reuse the in-browser cache and work offline). Works on iPhone/iPad/Android too.

Initializing model...

3) Choose the input source

* "Device audio" uses the screen-sharing mechanism. When you start, turn on "Share audio" in the share dialog. It works in desktop Chrome / Edge and is unavailable on most mobile environments such as iPhone/iPad.

4) Transcription

Idle

Speech recognition uses vosk-browser (Kaldi/VOSK compiled to WebAssembly) and runs entirely on your device. Add it to the home screen to launch it as a standalone app without browser UI.