Voice AI in vogue: Nvidia releases new open transcription model

Nvidia released Nemotron 3.5 Transcribe, a free open-weight speech transcription model that handles audio in up to 40 languages, along with version 2 of Granary, its open source speech dataset. Nvidia claims the model can translate 31 languages to and from English, process recordings up to an hour long, and distinguish up to eight speakers, and says it beats rivals on benchmarks including Hugging Face's Open ASR Leaderboard. The tools target developers building business meeting and medical transcription apps, allowing customization to recognize industry jargon in each language; one limitation is that Transcribe works with recorded audio rather than livestreams. At 30 billion parameters, it can run on a single chip, from the Jetson AGX Thor robotics controller and DGX Spark laptop chip to Blackwell datacenter chips, joining Nvidia's open-weight voice models such as the 1.1-billion-parameter Parakeet. The release comes amid a busy period of AI voice launches from Mistral, ElevenLabs, and Salesforce.

(Source：TechCrunch)