NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model on Hugging Face and ModelScope, along with the Granary v2 dataset.
What it does
- Transcribes 40 languages, translates 31 to or from English, and labels speakers in 13.
- Handles recordings up to an hour with up to eight speakers.
- Takes natural-language prompts, so you can steer formatting and teach it domain vocabulary and accents.
- NVIDIA says it’s first among open-weight models on the Hugging Face Open ASR Leaderboard for English.
My take
Transcription plus “who said what” in one open checkpoint is exactly what I want for meeting notes and podcast cleanup without shipping audio to someone else’s cloud. Going on my local-model list.
God Bless, and Happy hacking!
