Phonon-2 in your browser

Speech recognition with the 164 MB Phonon-2 model, running on this device: WebAssembly on the CPU, or WebGPU / WebGL on the GPU. Audio never leaves the page.

Model

not loaded
Downloads once from Hugging Face (163 MB), then loads from the browser cache.

Compute

Live dictation

A pause ends each sentence.
Your words appear here.

Transcribe a file

LibriSpeech, 29 s