Nvidia releases open-weight Nemotron 3.5 Transcribe speech model and Granary v2 dataset
Article timeline
The articles in this story. Times are ET.
- The DecoderNewsNvidia releases open Nemotron 3.5 Transcribe for transcription, translation, and speaker recognition
AINvidia has released Nemotron 3.5 Transcribe, an open speech model that takes text instructions and handles transcription, translation, and speaker recognition in one model. The company says it transcribes 40 languages, translates 31 to or from English, and identifies speakers in 13, with recordings up to an hour long. It also releases the Granary v2 dataset with about 844,000 hours of speech in 48 languages.
vLLMOfficialPickNVIDIA releases Nemotron 3.5 Transcribe with day-0 vLLM servingAINVIDIA released Nemotron 3.5 Transcribe, an open-weight, promptable 30B MoE model with about 4B active parameters. It supports transcription in 40 languages, recordings up to an hour, speaker labels, and audio Q&A, with weights in BF16, FP8 and NVFP4, and serving through vLLM and NeMo Speech on day 0.
NVIDIA AIOfficialNVIDIA releases Nemotron 3.5 Transcribe, an open-weight speech modelAINVIDIA announces Nemotron 3.5 Transcribe, an open-weight, promptable 30B MoE model with about 4B active parameters. The model covers transcription in 40 languages, speaker attribution, translation, contextual prompting, and audio understanding in one checkpoint.
ModelScopeOfficialPickNVIDIA releases Nemotron 3.5 Transcribe, a 30B-A4B open-weight speech modelAIModelScope announces NVIDIA's Nemotron 3.5 Transcribe, a 30B-A4B open-weight model that transcribes up to an hour of speech with speaker labels, translations, and audio answers. The post says it transcribes 40 languages, translates 31 languages to and from English, and supports speaker attribution in 13 languages for recordings with up to eight speakers. It runs on a single GPU and is released under the openmdw-1.1 license.
Artificial AnalysisOfficialNVIDIA's Nemotron 3.5 Transcribe ties for top open-weights speech-to-text accuracyAINVIDIA releases Nemotron 3.5 Transcribe, an open-weights non-streaming speech-to-text model with 30B total and 4B active parameters. It scores 2.8% on Artificial Analysis's AA-WER Index, tying Mistral's Voxtral Small for first among open-weights models, and 1.8% versus 2.3% on AA-AgentTalk, a voice agent conversation dataset. Voxtral Small has 24B active parameters, six times as many as Nemotron 3.5 Transcribe.
- SiliconANGLE · AINewsNvidia releases open Nemotron 3.5 Transcribe model for audio transcription and translation
AINvidia released Nemotron 3.5 Transcribe, an open-weight 30-billion-parameter mixture-of-experts model with 4 billion active parameters, for multilingual audio transcription and translation. It handles up to 100 hours of audio in 40 languages, labels up to eight speakers in a single pass, and reports 30.8% word error rate out of the box. Nvidia also released the Granary v2 dataset and a Hugging Face demo for fine-tuning.
- TechCrunch · AINewsNvidia releases open-weight Nemotron 3.5 Transcribe speech model
AINvidia releases Nemotron 3.5 Transcribe, a free open-weight model that transcribes audio in up to 40 languages, and version 2 of its open-source Granary speech dataset. The 30-billion-parameter model handles recordings up to an hour long, translates 31 languages to and from English, and tells up to eight speakers apart. Nvidia says it runs on a single chip, from Jetson AGX Thor to Blackwell datacenter chips, but it works only with recorded audio, not live streams.
- Hugging Face BlogOfficialPickNVIDIA releases Nemotron 3.5 Transcribe, an open multilingual speech model with Granary v2 data
AINVIDIA releases Nemotron 3.5 Transcribe, an open-weight model that transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13. It also releases the Granary v2 data pipeline and fine-tuning recipes, and in a medical case study fine-tuning cut held-out word error rate by 45.8% relative to the best prompt.