Nvidia has significantly expanded its offerings in the artificial intelligence voice domain with the introduction of Nemotron 3.5 Transcribe, a new open-weight model designed for comprehensive audio transcription across up to 40 languages. Alongside this, the company has also released version 2 of Granary, its open-source speech dataset, further empowering developers in the rapidly evolving AI landscape.

Nemotron 3.5 Transcribe boasts impressive capabilities, including the ability to translate between 31 languages and English, process audio recordings up to an hour in length, and accurately distinguish between as many as eight different speakers within a single recording. Nvidia asserts that its new model surpasses competitors in key benchmarks, a claim supported by its performance on platforms such as Hugging Face’s Open ASR Leaderboard.

These advanced tools are primarily aimed at developers creating applications for critical business functions, such as transcribing meetings and medical dictation, areas where AI has already established a strong presence. Developers can integrate Nemotron 3.5 Transcribe with the updated Granary dataset and other Nvidia tools, like the NeMo Speech model, to tailor applications that recognize industry-specific terminology in various languages. However, it is important to note that Transcribe is currently optimized for pre-recorded audio and does not support live-stream transcription.

Despite its substantial 30 billion parameters, Nvidia highlights that Transcribe is engineered to operate efficiently on a single chip. This includes a range of Nvidia hardware, from the Jetson AGX Thor robotics controller and DGX Spark chips found in new laptops, to the high-performance Blackwell datacenter chips. This new release joins Nvidia’s growing portfolio of open-weight voice models, which also features the compact yet powerful Parakeet model, with its 1.1 billion parameters. The launch comes amidst a flurry of activity in the AI voice sector, with other major players like Mistral, ElevenLabs, and Salesforce recently unveiling their own innovative voice technologies.

Key Takeaways

• Nvidia launched Nemotron 3.5 Transcribe, an open-weight AI model for audio transcription in up to 40 languages, along with the Granary v2 speech dataset.
• Nemotron 3.5 offers advanced features like 31-language translation, 1-hour recording processing, and 8-speaker differentiation, claiming superior benchmark performance.
• The model targets developers for business and medical transcription apps, is efficient enough to run on a single chip, and contributes to a competitive AI voice technology market.

Editor’s Analysis & Impact

This release from Nvidia significantly strengthens its position in the rapidly expanding AI voice and natural language processing market. By offering an open-weight, high-performance transcription model like Nemotron 3.5, Nvidia is not only providing powerful tools for developers but also fostering innovation within the broader AI ecosystem. The focus on business and medical transcription highlights a strategic move into established, high-value enterprise applications, potentially driving greater adoption of Nvidia’s hardware solutions. The ability to run a 30-billion-parameter model on a single chip underscores advancements in AI efficiency, making sophisticated AI more accessible. This move intensifies competition with other tech giants in the AI voice space, pushing the boundaries of what’s possible in multilingual communication and automated data processing, and setting a new standard for accuracy and accessibility in speech-to-text technologies.

Frequently Asked Questions

Q: What is Nemotron 3.5 Transcribe?
A: Nemotron 3.5 Transcribe is a new open-weight artificial intelligence model developed by Nvidia that specializes in transcribing audio into text across up to 40 different languages. It also offers translation capabilities.

Q: What are the primary applications for Nemotron 3.5 Transcribe?
A: The model is primarily designed for developers to create applications for business meetings, medical transcription, and other professional settings where accurate and multilingual audio-to-text conversion is crucial.

Q: Can Nemotron 3.5 Transcribe process live audio?
A: Currently, Nemotron 3.5 Transcribe is optimized for processing pre-recorded audio and does not support live-stream transcription.