NVIDIA releases Nemotron 3.5 Transcribe and Granary v2 on Hugging Face
Developers can use one available model for several recorded-audio tasks and use Granary v2 data and tools to adapt recognition to specialized speech.
• NVIDIA released Nemotron 3.5 Transcribe for transcription, speaker labels, translation and audio questions.
• It transcribes 40 languages and handles recordings up to an hour long, according to NVIDIA.
• Granary v2 provides a public manifest covering 463,062 hours across 48 languages.
Read more

October 28, 2026 30