These pages are how we exercise browser-whisper in the browser: same library, different flows. After the first Download model or Prepare, weights live in OPFS and later runs can work offline. WebGPU and COOP/COEP headers are required.

Live demo

The polished path: pick a file and transcribe. The model downloads automatically when you start.

OPFS model cache

Mirrors how we test caching: call downloadModel() yourself, watch progress in the log, then transcribe from the cached weights.

Live streaming

Real-time microphone input with voice-activity detection; each utterance is sent to Whisper after you prepare the model.