On-device speech demo · v1.5.0

Say something. Your browser decodes it.

Your microphone goes to phoneme ASR, then the phonemes go to P2G (phoneme → grapheme). Both models run on the hand-written WebAssembly engine — ~39 KB, no onnxruntime. Audio never leaves your device.

or upload a clip

Loading models…

Phonemes Phoneme ASR

—

P2G decodes phonemes back to text

Text P2G

—

P2G is a small (~7.3M-param) research model, so the text is deliberately approximate (e.g. quilter → kilter). Clearest on short English speech. See the v1.5.0 notes or the text → phonemes demo.