What we know

5 Transcribe and Granary v2, two AI models designed for intelligent, multilingual transcription that can be prompted and adapted for various use cases. These models aim to provide high-quality speech-to-text capabilities across multiple languages, with customization options for specific domains. 5 Transcribe is an open, multilingual speech-to-text model supporting 53 languages. 5 Transcribe can be prompted and adapted for domain-specific transcription tasks. Granary v2 is a large-scale, multilingual dataset for training and evaluating transcription models.

Why it matters

5 Transcribe is a large language model focused on speech-to-text transcription, supporting multiple languages and designed to be adaptable through prompting. Granary v2 is a dataset and framework to support the training and evaluation of such models, emphasizing multilingual and domain-specific transcription. The release is part of NVIDIA's broader push into generative AI and language technologies, aiming to provide open, adaptable models for enterprise and research use.

What is still unknown

Independent tests of accuracy, supported languages, and a broader release date are not in the reporting.