From the source
Microsoft AI released MAI-Transcribe-1.5, a multilingual speech-to-text model covering 43 languages with a best-in-class Word Error Rate on the FLEURS benchmark.
The model includes keyword biasing that improves recognition of domain-specific terms by up to 30% on FLEURS, and can transcribe an hour of audio in under 15 seconds.
It is being integrated into Copilot, Teams, GitHub, and Dynamics 365 Contact Centre, and is available via Foundry.



