Oct 28, 2026
NVIDIA releases open weights for multilingual speech transcription, translation and audio understanding. The model supports transcription across 40 languages, recordings up to one hour and speaker-attributed transcripts.
Brings prompt-controlled speech processing to self-hosted workflows under OpenMDW 1.1. It processes recorded audio rather than real-time streams; language and task coverage vary. NVIDIA also releases its Granary v2 speech dataset.