NVIDIA has released Nemotron 3.5 Transcribe and Granary v2, an open speech-to-text model. The model is intended for tools that transcribe recorded audio, including meeting and medical transcription applications. TechCrunch reports that it can transcribe up to 40 languages but does not support live audio. The model is available now through Hugging Face and is listed as a 34B automatic speech recognition model.
Why it matters
Recorded-audio transcription can help developers build meeting and medical workflows without relying on live speech processing. Its lack of live-audio support leaves real-time transcription to other providers.
Key points
- NVIDIA Nemotron 3.5 Transcribe and Granary v2 is listed for automatic speech recognition.
- The model can transcribe up to 40 languages.
- It supports recordings, not live audio.
- Hugging Face lists the model size as 34B.
- The model page was updated 28 minutes ago.
Based on reporting by Hugging Face Blog and TechCrunch.
Join the conversation! Your thoughts help the community grow.