NVIDIA Open-Sources Nemotron 3.5 Transcribe: 30B MoE ASR in 40 Languages

NVIDIA Developer · youtube · 2026-11-16

NVIDIA released Nemotron 3.5 Transcribe, an open-weight, promptable 30B mixture-of-experts model (4B active parameters) that combines transcription in 40 languages, speaker attribution, translation, and audio understanding in one checkpoint.

The livestream with Hugging Face covers running it via NeMo or transformers, contextual prompting to improve accuracy on names and domain jargon, domain adaptation with the Granary v2 dataset, and choosing between offline models for recorded audio vs streaming recognizers for live use.

Related event: NVIDIA Open-Sources Nemotron 3.5 Transcribe, a 30B MoE Model(2 posts)→

Original post →

More from Models

Models channel →