Private, on-device speech recognitionWhisper Base · Multilingual · WebGPU

Live transcription

Your microphone audio stays in this browser. A Web Worker runs the speech model locally so the interface remains responsive.

Status

Checking WebGPU support...

Transcript

The speech model is being prepared.

Audio is captured with an AudioWorklet, resampled to 16 kHz, and sent in 3-second segments with a 0.4-second overlap.