NVIDIA Releases Nemotron 3.5 Transcribe: Unified Open-Weight Model for Speech Transcription, Diarization, and Translation

NVIDIA released Nemotron 3.5 Transcribe on October 28, 2026, an open-weight, promptable model designed to consolidate speech transcription, speaker attribution (diarization), translation, and audio question answering into a single checkpoint. The model is now available on Hugging Face and ModelScope.

Nemotron 3.5 Transcribe supports transcription in 40 languages, translation to or from English for 31 of them, and speaker attribution in 13 languages. It handles recordings up to one hour long and labels up to eight speakers. Developers can guide output using natural language instructions, such as "Transcribe this recording and label each speaker" or "Translate this audio into English."

In terms of performance, the model excels across several independent benchmarks. On the Hugging Face Open ASR Leaderboard's English track, it ranks first among open-weight models with an average Word Error Rate (WER) of 5.33%. On the Artificial Analysis Speech-to-Text Leaderboard, it achieves 2.8% AA-WER with approximately 4 billion active parameters, tying for first place with models that have 24 billion active parameters, demonstrating high efficiency. For speaker attribution, it posts the lowest error rate on David AI’s 21-language diarization leaderboard and ranks first in VoiceArena’s 14-system comparison.

Beyond the model itself, NVIDIA also released the Granary v2 multilingual dataset and its reusable data curation pipeline. Granary v2 covers 463,062 hours of audio across 48 languages, a significant expansion from previous versions. Together with NeMo Speech fine-tuning tools, developers can use these datasets and recipes to adapt the model to specific domains, such as medical terminology. In a medical case study, fine-tuning reduced the word error rate on held-out speakers by 45.8% relative to the best prompt.