Nvidia releases open-weight Nemotron 3.5 Transcribe speech model supporting 40 languages
Nvidia released Nemotron 3.5 Transcribe, a free, open-weight speech-to-text model that transcribes audio in up to 40 languages and translates 31 languages to and from English. The 30-billion-parameter mixture-of-experts model, with roughly 4 billion active parameters, can handle recordings up to an hour long, distinguish up to eight speakers, and run on a single chip. Nvidia also released Granary v2, its open-source speech data pipeline. The model targets business meeting and medical transcription apps but works with recorded audio, not live streams.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Reporting timeline
Nvidia Releases Open-Weight Speech Transcription Model Nemotron 3.5 Transcribe, Open-Sources Granary v2 Pipeline
Nvidia announced on October 28 the release of Nemotron 3.5 Transcribe, an open-weight speech transcription model that supports 40 languages. Alongside the model, Nvidia publicly released Granary v2, the training data pipeline used for the model. The announcement emphasizes open weights and an open data pipeline, positioning the release as a contribution to transparent, reproducible speech AI development. No performance benchmarks, licensing terms, or availability details beyond the open-weight and open-pipeline framing were provided in the source.
Read sourceNVIDIA Releases Nemotron 3.5 Transcribe Open-Weight 30B MoE Speech Model
NVIDIA released Nemotron 3.5 Transcribe, an open-weight 30B mixture-of-experts model with roughly 4B active parameters. According to the source, the model can transcribe speech in 40 languages, perform speaker attribution (diarization), translate, handle contextual prompting, and understand audio. The post also states that Nemotron 3.5 Transcribe ranked first in VoiceArena's 14-system comparison of diarization plus automatic speech recognition. The release positions the model as a multilingual, multi-capability audio transcription system with a relatively small active-parameter footprint for its total size.
Read sourceNVIDIA Releases Open Nemotron 3.5 Transcribe Speech Model
NVIDIA is releasing an open speech model, Nemotron 3.5 Transcribe, that can transcribe meetings, label speakers, translate, and answer questions about recordings. According to the post, it supports transcription in 40 languages, translation between English and 31 languages, speaker labels for up to eight speakers in 13 languages, and recordings up to an hour long. Users can supply names and technical terms to guide transcripts or ask questions about the audio, with audio Q&A trained in English. The model has roughly 30B parameters with about 4B active per token and can run on a user's own NVIDIA GPU. NVIDIA's launch materials reportedly rank it first among open-weight models on Hugging Face's English Open ASR leaderboard and first in VoiceArena's 14-system test of transcription with speaker attribution. It is built for recorded audio, with live streaming using a separate model.
Read sourceShow 2 older updatesHide older updates
Nvidia Open-Sources Nemotron 3.5 Transcribe Speech Model With Translation and Speaker ID
Nvidia released an open-source speech recognition model called Nemotron 3.5 Transcribe, according to The Decoder. The model can transcribe 40 languages, translate between 31 languages and English, and identify up to eight speakers. It supports recordings up to one hour long and allows users to control output or ask questions about audio content through prompts. The announcement positions the model as a combined transcription, translation and speaker-diarization tool rather than a single-purpose ASR system.
Read sourceNvidia releases open-weight Nemotron 3.5 Transcribe speech-to-text model
Nvidia released Nemotron 3.5 Transcribe, a free, open-weight speech-to-text model that transcribes audio in up to 40 languages and translates 31 languages to and from English. The 30-billion-parameter model can handle recordings up to an hour long, distinguish up to eight speakers, and run on a single chip, from the Jetson AGX Thor robotics controller and DGX Spark chip to Blackwell datacenter chips. Nvidia also released version 2 of its open-source Granary speech dataset. The tools target developers building business meeting and medical transcription apps, and Nvidia says developers can customize them to recognize industry jargon in each language. A drawback is that Transcribe works with recorded audio, not live streams, leaving real-time use to rivals such as ElevenLabs, which launched Realtime for instant speech in voice agents. The release comes amid a busy period of AI voice launches, including Mistral's Voxtral 3 and Salesforce's Launa-Omni.