WebGPU · On-device
Nemotron 3.5 ASR
Multilingual streaming speech recognition
WebGPU isn't available in this browser. Use Chrome or Edge (or enable the WebGPU flag).
⬇ Load model (~750 MB)
One-time download, cached on-device.
● Record
Live
File
Options
lang_id
strip language tag
● Record
0.0s
Record from your mic, then transcribe on stop (highest accuracy).
● Start listening
Real-time streaming — text appears as you speak.
Transcribe
wav / mp3 / m4a / etc. — decoded and resampled to 16 kHz locally.
Transcript
—
▸ diagnostics
⌫ clear cached model (~750 MB)
idle
WebGPU