NVIDIA open-sources 100M-param diarization model that separates 8 speakers locally in real time
Prompt Engineering · youtube · 2026-09-25
NVIDIA released Nemotron 3 Diarization, a 100M-parameter open model that identifies up to 8 speakers in real-time streaming audio, even when two people talk over each other. Speakers are numbered by first appearance, and a small speaker memory keeps labels stable when someone leaves and returns. Paired with a speech-to-text model like Parakeet, every line of meeting notes gets attributed to the right person — and it runs entirely on-device. Model, live demo, papers (Sortformer, Streaming Sortformer) and the NeMo toolkit are all public.
Related event: NVIDIA Open-Sources Nemotron 3 Diarization, Tops Benchmark(16 posts)→
More from Multimodal
- One week after Gemini 4 Argon launch, 10 wild 3D worlds and simulations — FinanceYF5 · 2026-10-23
- MiniMax H3 video generation launches on Japan's ai& platform from $0.0125/second — DavidBennett__ · 2026-10-23
- Moss-Image-SFT showcase: photorealistic neon-lit portrait with full inference settings — SkirtSwimming8950 · 2026-10-23
- 156 AI-painted pumpkins: ChatGPT draws "based on everything it knows about you" — Kyrannio · 2026-10-23
- LoRA Studio preview brings Krea 2 LoRA training to 12GB GPUs, seeks feedback — homebrewedstuff · 2026-10-23
- FAPL From HKUST and Tencent Makes Video DiTs Actually Use MLLM Latents — tencent · 2026-10-23