The model, which boasts 30 billion parameters, claims to outperform existing benchmarks on the Hugging Face Open ASR Leaderboard. Alongside the release, Nvidia updated its Granary dataset to version 2, providing developers with the resources to refine speech recognition for specific industry jargon. While the technology is optimized for a wide range of hardware—from Jetson AGX Thor robotics controllers to Blackwell datacenter chips—it currently remains limited to recorded audio rather than live streaming applications.
This release intensifies an already crowded field for voice-based AI. Competitors are moving quickly, with Mistral recently shipping Voxtral 3, ElevenLabs introducing its Realtime voice agent platform, and Salesforce debuting its own model, Launa-Omni. Nvidia intends for its new tool to complement existing resources like the 1.1-billion-parameter Parakeet, offering a scalable suite of options for developers building specialized voice-to-text applications.
Comments (0)
No comments yet. Be the first!