WebGPU Speech-to-Text

Speech transcription that runs Whisper on your GPU inside the browser tab. No server and no API key.

Pick a Whisper model, from tiny to small, and it downloads once and stays cached in the browser. Inference runs in a Web Worker on WebGPU, falling back to WebAssembly where WebGPU is missing. Voice activity detection ends a recording when you stop speaking, and a continuous mode keeps recording and transcribing for hands-free dictation.