NVIDIA's open model turns meetings into text in 40 languages
Nemotron 3.5 Transcribe transcribes, translates and tells speakers apart, and takes plain-language instructions. Weights are on Hugging Face.
NVIDIA shared a free tool that writes down what is said in a recording, translates it and knows who is speaking. It supports 40 languages and any developer can use it.
Hype check
The noise matches the real change.
Our editorial opinion, based on the sources below. How we score
Your call: hype or real?
Real = it will matter in daily use. Hype = louder than it is. One vote, and it can’t be changed.
Our view and readers’ views are opinions, not statements of fact about a product.
Why it matters
One open model covers transcription, translation and speaker labels, with training data and tools published for developers.
Watch: it shows how far speech-to-text has come.
The details
- The model transcribes speech in 40 languages, translates 31 of them to or from English, and labels speakers in 13.
- It handles recordings up to an hour long and up to eight speakers.
- You can steer it with plain prompts such as "translate this recording" or "who spoke?", and add special terms like drug names.
- NVIDIA says it is first among on the Hugging Face Open ASR Leaderboard for English.
- In a medical case study, fine-tuning cut word error rate by about 45.8% relative to the best prompt.
- The model and Granary v2 data are on Hugging Face and ModelScope; quantized versions are coming soon.
Receipts
The company or lab said it. Primary source linked. Open the sources and check us.
Written with the help of AI tools and checked against the sources above. How we work
See a mistake? Tell us and we’ll fix it.
