Nvidia Releases Nemotron 3.5 Transcribe Open-Weight Voice AI Model
Nvidia has launched Nemotron 3.5 Transcribe, a free, open-weight model capable of transcribing audio in up to 40 languages and translating between English and 31 other languages. The 30-billion parameter model features speaker diarization for up to eight individuals and can process recordings up to one hour long. To support developers, Nvidia also released version 2 of its Granary open-source speech dataset. The release targets enterprise sectors such as medical transcription and business meeting automation. While the model currently only supports recorded audio rather than live streaming, it is designed for high hardware versatility. It can run on various Nvidia architectures, ranging from Jetson AGX Thor robotics controllers to Blackwell datacenter chips, positioning Nvidia more aggressively in the competitive voice AI landscape alongside rivals like Mistral, ElevenLabs, and Salesforce.
Topics: AI Voice Models Nvidia Nemotron Open Source AI Speech-to-Text
Read the full article at TechCrunch
Part of the story Mistral AI Releases Improved Large 4 (Le Chonk) Open-Weight Model (3 articles)