Nemotron speech recognition, in your browser
Streaming transcription of your microphone, in dozens of languages, on your GPU through WebGPU with tract. Nothing leaves your machine.
Starting…
The model is NVIDIA's Nemotron 3.5 ASR Streaming 0.6B (OpenMDW-1.1): a streaming Conformer encoder and an RNN-T decoder. Its 2.6 GB of weights come straight from Hugging Face on the first visit and stay on this device for the next ones. tract streams the log-Mel features in 100 ms steps and the encoder in 320 ms attention chunks, on the GPU; the RNN-T decoder runs greedily on the CPU. The model tags the language it hears, shown in the transcript. Source code