NVIDIA Releases Nemotron 3 Diarization, Massively Outperforming Rivals
Sam Witteveen · youtube · 2026-09-23
Sam Witteveen reviews NVIDIA's new Nemotron 3 Diarization model for speaker diarization, supporting both batch and streaming modes, which he says massively outperforms other models at establishing who said what.
The video covers the Nemotron speech family, what diarization is and how models are scored (DER), a breakdown of the model, and hands-on demos on DGX Spark with NeMo — including processing a full podcast and exporting to text or SRT. The model is open on Hugging Face.
Related event: NVIDIA Open-Sources Nemotron 3 Diarization, Tops Benchmark(16 posts)→
More from Multimodal
- Chimera vocal synth models the vocal tract as branching acoustic tubes, ditching phoneme stitching — rickasaurus · 2026-11-11
- Simulating a drop of blood with Gemini 4 Carbon and Antigravity Science skills — AI_Andrew · 2026-11-11
- Dev ports POE2's radiance cascade technique for dramatic voxel city lighting — mflux · 2026-11-11
- AI a cappella fix: swap vowel from 'a' to 'o' to tame the high notes — rickasaurus · 2026-11-11
- MiniMax H3 Image day-one tips: one checkpoint for t2i and edits — ggbro_itsover · 2026-11-11
- MiniMax H3-Image tested: 8.41MP images in ~35s on a 16GB GPU — mininuvixi · 2026-11-11