{
  "id": 16918294,
  "title": "NVIDIA Nemotron 3.5 Transcribe and Granary v2: Intelligent Multilingual Transcription You Can Prompt and Adapt",
  "url": "https://urgent.news/2026/10/28/nvidia-nemotron-3-5-transcribe-and-granary-v2-intelligent",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-28T14:58:17.000Z",
  "source": {
    "name": "Hugging Face",
    "slug": "hugging-face",
    "url": "https://huggingface.co/blog/nvidia/nemotron-transcribe-granary"
  },
  "original_language": "en",
  "account": "NVIDIA Nemotron 3.5 Transcribe and Granary v2 deliver intelligent, multilingual transcription and speaker identification. This open-weight model processes recorded speech, generating text, speaker labels, translations, and answering audio questions. It handles up to 40 languages, 31 of which can be translated between English, and supports speaker diarization for up to 13 languages. Transcribe runs on a single NVIDIA GPU, from DGX Spark to Blackwell data-center systems.\n\nDevelopers can guide output through natural language, select processing locations, and adapt the model to domain vocabulary and accents. Transcribe leads on open-weight models for transcription accuracy and speaker attribution, holding the top positions on multiple leaderboards. Its multilingual data and reusable curation pipeline extend prompt-level control to training, enabling developers to define the written output they need and adapt the model to their specific speech applications.\n\nIn a medical case study, fine-tuning reduced the word error rate on held-out speakers by 45.8% relative to the best prompt. The same checkpoint can produce different outputs from a single meeting recording, depending on the given instruction. Nemotron 3.5 Transcribe successfully transcribed overlapping dialogue from the 1940 film His Girl Friday, demonstrating its capability to handle fast-paced, densely spoken conversations.\n\nFor healthcare applications, Transcribe can transcribe medical discussions while preserving standard medical spelling and writing numbers in conventional written form. Word-level timestamps are also supported. As with any speech model, it's essential to spot-check names, speaker changes, and answers against the recording.",
  "summary": null,
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}