Nvidia releases open Nemotron 3.5 Transcribe speech model
Original title: Nvidia's open Nemotron 3.5 Transcribe combines transcription, translation, and speaker recognition
Nvidia says the 30-billion-parameter model keeps about 4 billion active and tops open-weight English ASR models.
Nvidia released Nemotron 3.5 Transcribe, an open speech model that follows text instructions for transcription, translation, speaker identification, summaries, and questions about a recording. It handles up to one hour and eight speakers, and Nvidia says it runs on a single GPU. The company also published the Granary v2 dataset, about 844,000 hours in 48 languages, plus tools for adapting the model. It does not support real-time audio.
Unconfirmed: Accuracy rankings and the medical fine-tuning gains are Nvidia's or cited benchmark results, not independently verified here.
- It transcribes 40 languages, translates 31 to or from English, and identifies speakers in 13.
- Hugging Face English Open ASR lists a 5.33 percent average word error rate.
- Medical fine-tuning cut one benchmark's error rate from 30.8 to 11.8 percent.
- Audio question-answering was trained in English, and features vary by language.
- Vendors
- Nvidia
- Products
- Granary v2
- Organizations
- Hugging FaceArtificial Analysis
From the original article
4 min read at the-decoder.comNvidia has released Nemotron 3. 5 Transcribe, an open speech recognition model that handles transcription, speaker recognition, and translation and takes its instructions as text prompts. It leads several benchmarks and uses only a sixth of the active parameters of its strongest open rival.
By Jonathan Kemper · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.