Nvidia releases Nemotron 3.5 Transcribe for multilingual audio

SiliconANGLE reports Nvidia released Nemotron 3.5 Transcribe, an open-weight model for audio transcription and translation. Nvidia says the model transcribes 40 languages and handles recordings up to an hour long; audio Q&A was trained in English. Prompts can guide it to label speakers or translate recordings. Nvidia also released an open data pipeline for fine-tuning; developers can download the model checkpoint or try a Hugging Face demo.