Nvidia Launches Nemotron 3.5 Transcribe for Advanced Audio Processing
New open-weight model enables multi-language transcription and speaker identification for enterprise applications.
1Nvidia Releases Nemotron 3.5
Nvidia officially launched Nemotron 3.5 Transcribe, an open-weight model designed for audio transcription across 40 languages. Alongside the model, the company released version 2 of its Granary speech dataset. These tools aim to improve how developers handle complex audio data, specifically targeting professional environments like business meetings and medical documentation workflows.
2Technical Capabilities and Benchmarks
The model supports translation for 31 languages to and from English while processing recordings up to one hour in length. It features advanced speaker diarization, capable of distinguishing up to eight individual voices simultaneously. Nvidia claims the model outperforms existing competitors on the Hugging Face Open ASR Leaderboard through these specific architectural improvements.
3Targeting Specialized Industry Jargon
Developers can integrate Nemotron 3.5 with the updated Granary dataset and NeMo Speech tools to refine application performance. This customization allows software to recognize industry-specific terminology across multiple languages effectively. By leveraging these resources, companies can build highly accurate transcription solutions tailored for niche sectors like healthcare or corporate legal services.
4Hardware Compatibility and Constraints
With 30 billion parameters, the model is optimized to run on a single chip, including the Jetson AGX Thor and DGX Spark. It also supports Nvidia’s Blackwell datacenter chips, ensuring scalability across different hardware tiers. However, a notable limitation is that Transcribe currently functions only with recorded audio files, excluding real-time streaming capabilities.
5Expanding the Voice Portfolio
Nemotron 3.5 joins Nvidia’s existing library of open-weight voice models, such as the 1.1 billion parameter Parakeet. This addition demonstrates a strategic focus on providing developers with varied model sizes for different computational needs. By offering both lightweight and high-capacity options, Nvidia addresses diverse deployment requirements from edge robotics to large-scale cloud infrastructure.
6Competitive Landscape in Voice AI
The release occurs during a period of intense activity in the voice AI market. Recent developments include Mistral’s Voxtral 3, ElevenLabs’ Realtime launch for voice agents, and Salesforce’s introduction of the Launa-Omni model. These concurrent releases highlight the industry's rapid shift toward more sophisticated, integrated, and accessible voice-based artificial intelligence technologies.
7Strategic Focus on Enterprise Utility
By focusing on business and medical transcription, Nvidia is targeting two of the most established AI use cases in the workplace. These sectors require high precision and the ability to handle specialized vocabulary, which the model’s customization features directly address. This approach prioritizes practical enterprise utility over general-purpose consumer voice applications.
8Future Implications for Developers
The availability of open-weight models like Nemotron 3.5 lowers the barrier for developers building custom speech-to-text applications. As the ecosystem grows, the ability to train models on proprietary datasets will become a key differentiator. Developers should monitor how these tools integrate with existing workflows to maximize accuracy in complex, multi-speaker audio environments.
Original article: TechCrunch — Nvidia Launches Nemotron 3.5 Transcribe for Advanced Audio Processing