{"version":1,"type":"story","url":"https://digestai.news/story/nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual","json":"https://digestai.news/story/nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual.json","markdown":"https://digestai.news/story/nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual.md","slug":"nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual","headline":"NVIDIA releases open-weight Nemotron 3.5 Transcribe with multilingual transcription and speaker attribution","summary":"NVIDIA has launched **Nemotron 3.5 Transcribe**, an open-weight, promptable model for multilingual transcription, speaker attribution, translation, and audio question-answering. It supports 40 languages for transcription, translates 31 into or from English, and attributes up to eight speakers in 13 languages. The model handles recordings up to an hour long and can be fine-tuned for domain-specific vocabulary or accents using **Granary v2**, a multilingual data pipeline and fine-tuning toolkit also released by NVIDIA. Nemotron 3.5 Transcribe leads on the Hugging Face Open ASR Leaderboard for English (5.33% WER) and multilingual (4.62% mean WER) among open-weight models, while achieving top scores on speaker attribution benchmarks like David AI’s 21-language diarization leaderboard (18.26% DER) and VoiceArena’s English full-session benchmark (32.4% cpWER). It runs on a single NVIDIA GPU, from edge devices like Jetson AGX Thor to data-center systems like Blackwell, processing 100 hours of audio in under three minutes on an RTX PRO 6000 Blackwell GPU.\n\nGranary v2, part of the Nemotron speech collection, provides 463,062 hours of multilingual speech data across 48 languages, up from 25 in its previous version. The pipeline includes reusable curation tools for aligning transcripts, task labels, and metadata, enabling developers to fine-tune the model for specialized domains. A medical case study demonstrated a **45.8% relative reduction in word error rate (WER)** when fine-tuning on held-out speakers, cutting errors from 22.44% to 12.17%. The model is available on Hugging Face and ModelScope, with hosted endpoints offered by Baseten, DeepInfra, and Simplismart. NVIDIA also provides tutorials for fine-tuning, deployment, and agent-driven workflows.","keyPoints":["Nemotron 3.5 Transcribe supports 40 languages for transcription, 31 for translation, and speaker attribution in 13 languages, handling up to 8 speakers per recording","Leads on Hugging Face Open ASR Leaderboard (5.33% WER for English) and David AI’s 21-language diarization benchmark (18.26% DER), outperforming proprietary APIs in accuracy-to-parameter efficiency","Granary v2 data pipeline and fine-tuning tools cut WER by 45.8% in a medical domain case study, with 463,062 hours of multilingual speech data across 48 languages"],"whyItMatters":"Nemotron 3.5 Transcribe fills a gap for open, high-accuracy multilingual speech models, enabling developers to build applications for real-time transcription, translation, and speaker attribution without proprietary APIs. Its fine-tuning capabilities and leaderboard-topping performance could accelerate adoption in domains like healthcare, education, and customer support, where accuracy and domain specificity are critical.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["NVIDIA","Hugging Face","ModelScope","Baseten","DeepInfra","Simplismart"],"models":["Nemotron 3.5 Transcribe","Nemotron 3 Diarization","Nemotron 3.5","Granary v2","vLLM","NeMo Speech"],"people":[]},"firstPublishedAt":"2026-10-28T14:58:17Z","updatedAt":"2026-10-28T15:49:39Z","sourceCount":7,"hasPrimarySource":true,"sources":[{"outlet":"Hugging Face","title":"NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt","url":"https://huggingface.co/blog/nvidia/nemotron-transcribe-granary","publishedAt":"2026-10-28T14:58:17Z","type":"primary","primary":true,"lead":true},{"outlet":"The Decoder","title":"Nvidia's open Nemotron 3.5 Transcribe combines transcription, translation, and speaker recognition","url":"https://the-decoder.com/nvidias-open-nemotron-3-5-transcribe-combines-transcription-translation-and-speaker-recognition","publishedAt":"2026-10-28T15:49:39Z","type":"press","primary":false,"lead":false},{"outlet":"Unite.AI","title":"NVIDIA Launches Open-Weight Speech Model and 463,062-Hour Audio Dataset","url":"https://unite.ai/nvidia-launches-open-weight-speech-model-and-463-062-hour-audio-dataset","publishedAt":"2026-10-28T15:45:12Z","type":"press","primary":false,"lead":false},{"outlet":"siliconangle.com","title":"Nvidia debuts Nemotron 3.5 Transcribe, an open-weight model for multilingual audio transcription and translation, and Granary v2, its open-source speech dataset","url":"https://siliconangle.com/2026/10/28/nvidia-releases-all-in-one-adaptable-nemotron-3-5-audio-transcription-and-translation-model","publishedAt":"2026-10-28T15:20:02Z","type":"press","primary":false,"lead":false},{"outlet":"NVIDIA Blog","title":"Thin, Light and Built for Everything: NVIDIA RTX Spark Windows Laptops Gain New Models, Agent Tools and Games","url":"https://blogs.nvidia.com/blog/local-ai-rtx-spark-laptops-updates-models-agents-games","publishedAt":"2026-10-28T15:00:46Z","type":"primary","primary":true,"lead":false},{"outlet":"tech.yahoo.com","title":"NVIDIA RTX Spark Laptops Expand With New AI Models, Creator Tools and Games","url":"https://tech.yahoo.com/computing/articles/nvidia-rtx-spark-laptops-expand-150000090.html","publishedAt":"2026-10-28T08:01:00Z","type":"press","primary":false,"lead":false},{"outlet":"techcrunch.com","title":"Voice AI in vogue: Nvidia releases new open transcription model","url":"https://techcrunch.com/2026/10/28/voice-ai-in-vogue-nvidia-releases-new-open-transcription-model","publishedAt":"2026-10-28T08:01:00Z","type":"press","primary":false,"lead":false}],"sourceNotes":null,"discussions":[],"thread":{"title":"Open Source AI Multilingual Model Race","url":"https://digestai.news/thread/ibm-releases-granite-speech-5-0-2b-for-multilingual-speech-recognition","storyCount":3},"cite":{"text":"Digest AI, \"NVIDIA releases open-weight Nemotron 3.5 Transcribe with multilingual transcription and speaker attribution\", 28 October 2026, https://digestai.news/story/nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual","publisher":"Digest AI","title":"NVIDIA releases open-weight Nemotron 3.5 Transcribe with multilingual transcription and speaker attribution","datePublished":"2026-10-28T14:58:17Z","url":"https://digestai.news/story/nvidia-releases-open-weight-nemotron-3-5-transcribe-with-multilingual"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}