Speech to text.

Records audio in your browser and runs an ONNX build of Whisper (whisper-base.en) through WebAssembly via transformers.js. No API calls, no uploads. The first visit downloads the model weights (~150MB) and caches them locally.