On Wednesday, Nvidia introduced Nemotron 3.5 Transcribe, a free and open-weight model designed to transcribe audio in up to 40 languages. In conjunction with this release, the company launched version 2 of Granary, an open-source speech dataset.
Nvidia highlights that Nemotron 3.5 can translate 31 languages to and from English, process recordings up to an hour long, and distinguish between up to eight speakers. The company claims superiority over competitors on benchmarks, such as the Hugging Face Open ASR Leaderboard.
These resources are intended for developers creating applications for business meetings and medical transcription, which are two of the most established uses of AI in the workplace. Nvidia states that developers can utilize the model, the updated Granary dataset, and tools like its NeMo Speech framework to customize applications for industry-specific jargon in each language. However, a limitation exists: Transcribe operates on recorded audio rather than live streams.
Despite this limitation, the company assures that Transcribe, with its 30 billion parameters, can run on a single chip ranging from the Jetson AGX Thor robotics controller and DGX Spark, which powers new laptops, to Blackwell datacenter chips. This addition joins Nvidia’s lineup of open-weight voice models, including the compact yet powerful Parakeet (1.1 billion parameters).
Nvidia’s latest voice offering arrives during a period of intense activity in 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 own voice model.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.