Microsoft Upgrades Dragon Copilot with MAI-Transcribe-2 for Healthcare
Microsoft announced that its Dragon Copilot Transcribe medical AI now uses the proprietary MAI-Transcribe-2 speech model, reducing multilingual error rates by 50%. The model currently handles 100% of doctor-patient dialogue transcription tasks within the platform, serving about one-third of U.S. physicians. This upgrade significantly improves clinical documentation workflows by converting multi-language and mixed-language consultations into structured notes faster and more accurately. It directly reduces administrative burdens for physicians and minimizes the risk of medical information distortion for diverse patient populations. MAI-Transcribe-2 supports 60 languages with automatic language identification, speaker diarization, word-level timestamps, and keyword biasing. It achieves an average word error rate of 5.2% on the FLEURS benchmark and processes long-form audio up to 10 times faster than some competitors.
## BACKGROUND
Dragon Copilot is an AI-powered assistant utilized by medical professionals to streamline clinical documentation and integrate voice dictation into electronic health records. Speech-to-text models like MAI-Transcribe-2 convert spoken words into written text, which is crucial for handling medical dictations and patient consultations accurately.