Whispering

Record a quick note, run Whisper in your browser with transformers.js + WebGPU, and copy the transcript.

Keywords

Artificial Intelligence, Machine Learning, Data Science, Personal Blog, Research Notes, tools, audio

Speak, transcribe, copy. Everything happens in-browser; no audio leaves the page.

Backend

WebGPU (fastest, if supported) WASM fallback

Model

whisper-tiny.en (fastest) whisper-base.en (better) whisper-small.en (best, slowest) whisper-large-v3 (very high quality, heavy)

Pick a backend and model, then load.

We’ll ask for microphone access. Audio stays local.

Tip: tiny ≈ fastest; small is higher quality; large-v3 is best but may be slow and need WebGPU.

Backend: – Model: –

00:00

Auto-fix punctuation (Llama 3.2 1B) LLM: loading…