Nvidia has released Nemotron 3.5 Transcribe, an open-weight model designed to turn recorded speech into text across as many as 40 languages. The company also published version 2 of Granary, its open-source speech dataset.
According to Nvidia, the model can translate 31 languages to and from English, process recordings of up to an hour and distinguish between as many as eight speakers. Nvidia says it outperforms competing systems on benchmarks including Hugging Face’s Open ASR Leaderboard, though those performance claims come from the company.
The model is aimed at developers building tools for business meetings, medical transcription and other workplace applications. Nvidia says developers can combine Transcribe with Granary and its NeMo Speech tools to customize systems for specialized terminology and different languages.
A notable limitation is that Nemotron 3.5 Transcribe handles recorded audio rather than live streams. Nvidia says the 30-billion-parameter model can run on a single chip, including Jetson AGX Thor, DGX Spark and Blackwell data-center hardware.
The release adds to a crowded wave of voice-AI launches. Mistral, ElevenLabs and Salesforce have also introduced voice-focused products recently, spanning transcription, real-time agents and broader multimodal applications.