Yesterday (November 17) at the Ignite 2026 event, Microsoft announced that its healthcare AI product Dragon Copilot Transcribe has officially switched to its self-developed speech recognition model, MAI-Transcribe-2, which now handles 100% of doctor-patient conversation transcription in Dragon Copilot. In the exam room, this means consultation recordings can be turned into structured clinical records faster and more accurately, so doctors spend less time organizing charts and checking terminology.
10 Million Patient Visits a Month in Production
The scale numbers are the most telling: it handles about 10 million patient visits per month, error rates in multilingual workflows have dropped by about 50%, and about one-third of physicians in the United States are already using it. It can recognize multiple languages and common code-switching scenarios such as Chinese-English and Hindi-English, and it maintains recognition quality even in noisy environments, reducing the risk that multilingual patients’ medical information is distorted because the wrong language is misheard.
5.2% Word Error Rate, 10 Cents an Hour
Technically, it supports speaker diarization, word-level timestamps, keyword biasing, and configurable transcription styles, automatically detects language, and offers two output modes, “verbatim” and “concise,” suited respectively to compliance record-keeping and clinical note writing. On benchmarks, it leads across 60 languages on the FLEURS benchmark with an average word error rate of 5.2%, and it is also in the leading range in Artificial Analysis’s combined accuracy and latency evaluation. It processes long audio up to 10 times faster than some competitors, and Microsoft has also introduced a limited-time price of $0.10 per hour of audio, making clear its intent to grab a share of the medical transcription market.


