Labs
Everything here runs client-side. Open an experiment and it fetches its model once, then works offline on your hardware.
EXP.01 voice-lab
Voice Lab
An 82M-parameter speech model running on your own hardware. Your text never leaves the page. What does travel: the model weights from Hugging Face, the WASM runtime from jsDelivr, and a half-megabyte voice file the first time you pick each voice.
91/300
// telemetry
modelkokoro-82M v1.0 · Apache-2.0
weights92 MB · q8 quantized
backend— not loaded
threads1 · not isolated
last run—
EXP.02 listen-lab
Listen Lab
Whisper, running on your own hardware. Hold the button, say something, let go. Your voice never leaves this page: the only things on the wire are the one-off model download and the anonymous click counter this site already runs.
transcript
Nothing yet. Hold the button and say something.
// telemetry
modelwhisper-base v1 · Apache-2.0
weights77 MB · q8 quantized
backend— not loaded
threads1 · not isolated
last run—