Nvidia has introduced Nemotron 3.5 Transcribe, a free, open-weight AI model designed to turn recorded speech into text. The company says it supports up to 40 languages, can translate 31 languages to and from English, and can distinguish as many as eight speakers in recordings lasting up to an hour.
The release also includes version 2 of Granary, Nvidia's open-source speech dataset. Developers can combine these resources with Nvidia's NeMo Speech tools to adapt applications to specialized vocabulary across different industries.
Potential uses include transcribing business meetings and medical dictation. The model processes recorded audio rather than live streams. Nvidia says its 30-billion-parameter system can run on a single chip, including its Jetson AGX Thor and DGX Spark platforms, as well as Blackwell data-center chips.
Nemotron 3.5 joins Nvidia's other open-weight voice models, including the smaller Parakeet. As speech AI tools become more adaptable and accessible, they could help developers build transcription services tailored to more languages and professional settings.
