Nvidia releases open-weight Nemotron-3-Diarization speaker model
Nvidia has introduced Nemotron-3-Diarization, an open-weight speaker diarization model whose single checkpoint handles both offline processing and streaming for up to eight speakers. Output resolution is configurable in multiples of 10 ms, and chunked inference removes any maximum audio duration limit. The model ships under the openmdw-1.1 license for commercial and non-commercial use.
- One checkpoint covers offline and streaming up to 8 speakers
- Output resolution configurable in multiples of 10 ms
- Streaming uses Arrival-Order Speaker Cache and FIFO queue
- openmdw-1.1 license allows commercial use
Read next
AI