NVIDIA released Nemotron 3 Diarization, an open-weight multi-speaker speaker diarization model for real-time and offline use
Model release Provisional 86% confidence first seen
NVIDIA launched Nemotron 3 Diarization, an open-weight 100M-parameter model for real-time and offline multi-speaker diarization, including overlapping speech. The release expands streaming diarization capacity from four to up to eight speakers and reports improved diarization accuracy on VoiceArena’s Diarization-Bench v1, achieving a 14.72% DER.
NVIDIA
Nemotron 3 Diarization
Diarization-Bench v1
Hugging Face
MarkTechPost
Streaming Sortformer
VoiceArena