Private voice dictation
An AI speech model runs inside your browser. Your audio is never uploaded, and punctuation and capital letters come built in.
The first start downloads the model once; your browser keeps it for next time. Small models are weaker outside English, so use clear speech and a quiet room.
Ready.
How it works: your microphone audio is split at pauses and transcribed on your device by an open-source Whisper model. Text appears after each pause, with a rough live preview while you speak. Needs a recent Chrome or Edge for the best speed (WebGPU); other browsers fall back to the slower CPU mode. Compare with instant browser dictation or transcribe a file with audio to text.