Nvidia has unveiled Nemotron 3.5 Transcribe, an open-weight AI model designed to transcribe recorded audio in up to 40 languages, enhancing transcription accuracy and speaker differentiation for sectors like business meetings and healthcare documentation.
- Transcribes 40 languages, translates 31 to/from English
- Targets meeting and medical transcription workflows
- Runs on Nvidia’s edge and datacenter AI chips
What happened
Nvidia announced the release of Nemotron 3.5 Transcribe, a free and open-weight speech-to-text model capable of processing audio recordings in 40 languages. The model supports transcription and translation capabilities for 31 languages with English as a pivot language, addressing diverse global needs. It can differentiate up to eight speakers and handle recordings as long as one hour.
Alongside Nemotron 3.5, Nvidia released version 2 of the Granary open source speech dataset and updated tools like the NeMo Speech model, which developers can use to customize voice applications to recognize specialized vocabulary across industries such as business meetings and medical transcription.
Why it matters
This launch marks a significant advancement in AI transcription technology, providing a highly accessible solution for developers focused on post-event audio transcription rather than real-time voice recognition. By supporting numerous languages and enabling multi-speaker separation, Nvidia’s model caters to the growing demand for precise transcription in enterprise and healthcare settings.
Nemotron 3.5’s ability to run efficiently on varied Nvidia hardware, from the Jetson AGX Thor chip used in robotics to DGX Spark chips for laptops and Blackwell datacenter processors, makes it a versatile option for deploying transcription models at different scales and environments.
What to watch next
While Nemotron 3.5 focuses on recorded audio, the real-time transcription space continues to be dominated by competitors like ElevenLabs, which recently launched Realtime for instant speech applications. Monitoring how Nvidia evolves its technology to bridge this gap could be crucial for future market positioning.
Developers and enterprises will likely experiment with Nvidia’s dataset Granary 2 and the customizable frameworks provided to tailor voice AI solutions to specific industry languages and jargon, potentially improving AI adoption in sectors with complex terminology.