Nvidia Announces Nemotron 3.5, an Open Speech Conversion Model Supporting 40 Languages

Nvidia has released Nemotron 3.5 Transcribe, a free and open-weights model capable of voice transcription in up to 40 languages, along with the second version of the open-source audio dataset Granary.

The Nemotron 3.5 Transcribe model announced by Nvidia has the ability to translate 31 languages into English and transfer from English into those languages. Processing recordings of up to an hour, the system can distinguish between eight different speakers and successfully outperforms competitors in benchmarks such as the Hugging Face Open ASR Leaderboard.

Aimed at helping developers design applications for fields such as business meetings and medical transcription, the tools can be customized for industry jargon. The 30-billion-parameter Transcribe model, which does not support live streaming, can run on a single chip ranging from the Jetson AGX Thor robotics controller to Blackwell data center chips.