Nvidia releases open transcription model for up to 40 languages
🎧 Listen to this story
Nvidia released Nemotron 3.5 Transcribe, a free, open-weight model for transcribing recordings in up to 40 languages, along with version 2 of its open-source Granary speech dataset. The company says the model can translate 31 languages to and from English, process recordings of up to an hour, and distinguish up to eight speakers, while beating rivals on benchmarks such as Hugging Face’s Open ASR Leaderboard.
The model targets applications for business meetings and medical transcription and, Nvidia says, can be customized to recognize industry jargon. Its main limitation is that it does not work with live audio. The company says the 30-billion-parameter model can run on a single chip, including Jetson AGX Thor, DGX Spark and Blackwell chips.
Why it matters · editorial interpretation
The combination of open weights, multilingual support and customization for specialized jargon could make it easier to build transcription tools for business and healthcare. Because the model does not process live audio, its use is limited to recordings, despite Nvidia’s claim that it runs on a single chip.