What changed
Nvidia released Nemotron 3.5 Transcribe, a free open-weight transcription model, alongside version 2 of its Granary speech dataset. The company says the model transcribes audio in up to 40 languages and can also translate 31 languages to and from English.
Why it matters
The release expands the open-weight speech stack at a time when voice interfaces are becoming a major product surface. Nvidia says the model is designed for recorded audio, supports up to an hour of audio, and can run on a single chip, which makes it relevant for edge devices, laptops, and datacenter deployment.
Practical takeaway
Developers building meeting tools, medical transcription systems, or multilingual assistants now have a new model to test against closed alternatives. The combination of model and dataset also makes customization more practical for specialized vocabularies.