Voice AI in vogue: Nvidia releases new open transcription model

Nvidia released Nemotron 3.5 Transcribe, a free open-weight speech-to-text model for recorded audio in up to 40 languages, alongside version 2 of its open-source Granary speech dataset. Nvidia says the 30-billion-parameter model can translate 31 languages to and from English, process recordings up to one hour long, and distinguish as many as eight speakers. The company claims it outperforms rival systems on benchmarks including Hugging Face’s Open ASR Leaderboard. The model targets developers building workplace tools such as business-meeting assistants and medical-transcription applications. Nvidia says developers can customize systems with Nemotron 3.5 Transcribe, Granary, and NeMo Speech tools to recognize industry-specific terminology across languages. Transcribe does not support livestreamed audio. Nvidia says the model can run on a single chip, including the Jetson AGX Thor robotics controller, DGX Spark hardware for new laptops, and Blackwell datacenter chips. It expands Nvidia’s open-weight voice-model lineup, which also includes the 1.1-billion-parameter Parakeet. The release follows recent voice-AI launches including Mistral’s Voxtral 3, ElevenLabs’ Realtime speech capability for voice agents, and Salesforce’s Launa-Omni voice model.