Nvidia open-sources Nemotron 3 for real-time speaker tracking

Nvidia: NVIDIA Corporation develops graphics processing units, AI accelerators, and software platforms central to modern computing and machine learning. It released the open-weight Nemotron 3 Diarization model to improve real-time speaker tracking in voice applications. The model addresses limitations in speech-to-text systems by enabling multi-speaker context awareness for AI assistants. VoiceArena: VoiceArena is an independent evaluation lab that runs rigorous, human-verified benchmarks for voice AI models across speech-to-text, text-to-speech, and related tasks. It tested Nemotron 3 Diarization on its diarization benchmark, where the model achieved the leading position among evaluated systems. Nemotron 3 Diarization: Nemotron 3 Diarization is an open-weight streaming model from NVIDIA built on Sortformer architecture for identifying who is speaking in audio. It processes live or recorded conversations, labeling speakers consistently while supporting overlaps and maintaining context across turns. The release integrates it into tools like NeMo to support advanced voice agent development. AI Integration: NVIDIA continues expanding its Nemotron model family within the NeMo Labs Voice Agent framework to support end-to-end voice pipelines. Benchmark Ecosystem: VoiceArena maintains specialized leaderboards that evaluate real-world conversational audio, providing standardized comparisons for diarization and transcription models. Open Source Momentum: The model release follows NVIDIA's pattern of open-weight distributions under permissive licenses to accelerate developer adoption in voice AI.