Nvidia has unveiled Nemotron 3.5 Transcribe, a free, open-weight model capable of transcribing audio in up to 40 languages. Alongside it, the company released version 2 of Granary, its open-source speech dataset.

According to Nvidia, Nemotron 3.5 can translate 31 languages to and from English, process recordings as long as an hour, and distinguish between up to eight separate speakers. The company also claims the model outperforms competitors on benchmarks, including Hugging Face's Open ASR Leaderboard.

These tools are aimed at developers creating applications for business meetings and medical transcription—two of the most well-established workplace applications of AI. Nvidia says developers can leverage the model, the updated Granary dataset, and tools like its NeMo Speech model to tailor apps that recognize industry-specific jargon in each language.

One limitation: Transcribe handles recorded audio rather than live streams. Nevertheless, the company promises that Transcribe, with 30 billion parameters, can run on a single chip—ranging from the Jetson AGX Thor robotics controller and the DGX Spark chip powering new laptops to Blackwell datacenter chips. It joins Nvidia's lineup of open-weight voice models, which includes the remarkably compact yet capable Parakeet (1.1 billion parameters).

Nvidia's latest voice offering arrives during a flurry of AI voice releases: Mistral shipped Voxtral 3 earlier this month, ElevenLabs launched Realtime for instant speech in voice agents, and Salesforce released Launa-Omni, its voice model.