Nvidia's open Nemotron 3.5 Transcribe combines transcription, translation, and speaker recognition

What happened

Nvidia released Nemotron 3.5 Transcribe, an open speech recognition model that follows text instructions to transcribe, translate, identify speakers, and answer questions about recorded audio. It transcribes 40 languages and handles recordings up to an hour long with up to eight speakers, but does not support real-time processing. The release includes Granary v2, containing about 844,000 hours of speech in 48 languages, plus dataset-building tools and fine-tuning guides.

Read next

• More changes in today’s briefingGoogle DeepMind announces Genie 4 for real-time world simulation with native audio

Reported by 2 sources

The story so far

NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model, with vLLM announcing launch-day serving support through vLLM and NeMo Speech. NVIDIA described it as a promptable 30B mixture-of-experts model with approximately 4B active parameters that combines transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding in one checkpoint. The Decoder reported that the model follows text instructions to process recordings up to an hour long with up to eight speakers, but does not support real-time processing. vLLM said its serving support includes promptable transcripts, speaker labels and audio question answering, with model weights available in BF16, FP8 and NVFP4 formats. According to The Decoder, NVIDIA also released Granary v2, a dataset containing about 844,000 hours of speech in 48 languages, along with dataset-building tools and fine-tuning guides.