Voice AI in vogue: Nvidia releases new open transcription model
Summary
- Nvidia released Nemotron 3.5 Transcribe, a free open-weight model supporting transcription in up to 40 languages and translation for 31 languages.
- The 30-billion parameter model handles hour-long recordings, distinguishes up to eight speakers, and runs on a single chip from Jetson AGX Thor to Blackwell data center hardware.
- Alongside the model, Nvidia released version 2 of Granary, its open-source speech dataset, targeting developers building business meeting and medical transcription apps.
- The model currently supports recorded audio rather than live streaming, and joins other recent voice AI releases from competitors like Mistral and ElevenLabs.
Why it matters
Nvidia's new open-weight transcription model provides developers with a powerful tool for custom speech applications that can run efficiently on single hardware chips across edge and data center environments.