Microsoft Adds MAI-Transcribe-2 to Dragon Copilot Transcribe
Microsoft announced at Ignite 2026 that Dragon Copilot Transcribe now uses its in-house MAI-Transcribe-2 model for all clinician-patient conversation transcription. Microsoft says the model handles multilingual and code-switched conversations, supports speaker separation and word-level timestamps, and averages a 5.2% word error rate across 60 languages on FLEURS. The company also reports that the service processes about 10 million patient encounters per month and has cut errors by roughly 50% in multilingual workflows.
The model has moved into a medical transcription workflow used in about one-third of U.S. physicians according to Microsoft, giving the release production-scale relevance beyond benchmark claims.