Nvidia releases open-weight Nemotron 3.5 Transcribe model for 40-language transcription and translation

Nvidia on October 28 released Nemotron 3.5 Transcribe, a free, open-weight model that turns recorded speech into text, together with Granary v2, a new version of its open-source speech dataset. The model is available on Hugging Face and ModelScope.

According to Nvidia's post on Hugging Face, a single checkpoint handles transcription, speaker attribution, translation and audio question answering. It transcribes 40 languages, translates 31 of them to or from English, attributes speakers in 13, handles recordings up to an hour long and labels up to eight speakers. Users can steer it with natural-language prompts, such as asking what decisions were made in a meeting, and can supply domain terms to improve accuracy.

SiliconANGLE reported that the model has 30 billion parameters with about 4 billion active in a mixture-of-experts design, allowing it to run on a single Nvidia GPU from a Jetson AGX Thor or DGX Spark up to Blackwell data-center hardware. Nvidia said a forthcoming NVFP4 quantized version processed 100 hours of audio in under three minutes on one RTX PRO 6000 Blackwell Server Edition GPU. The company says the model ranks first among open-weight models on Hugging Face's English Open ASR Leaderboard and tops several independent speaker-attribution benchmarks. It works on recorded audio, not live streams.

The public Granary v2 manifest covers 463,062 hours across 48 languages, up from 25 languages in the first version. TechCrunch noted the tools target business-meeting and medical transcription, two of AI's most established workplace uses, and arrive amid a busy stretch of voice AI releases from Mistral, ElevenLabs and Salesforce.

---

Sources: NVIDIA / Hugging Face, TechCrunch, SiliconANGLE, NVIDIA Research. Image: NVIDIA