Google's medical speech tech, MedASR: When doctors dictate

Google has released MedASR, an open- weights medical speech-to-text model aimed at making clinical dictation and documentation faster—without tripping over medical jargon.

- TDG NETWORK

In hospitals and clinics, the “real work” of medicine often continues after the patient has left the room— inside the electronic health record.

Clinicians think and speak in fast, shorthand language (“rule out,” “no acute distress,” “CT chest without contrast”), but typing that language into structured systems is slow. General-purpose speech recognition can help, yet it often stumbles precisely where accuracy matters most: drug names, anatomy, procedures, and specialty-specific phrasing.

That is the niche Google is targeting with MedASR, a Conformer-based automatic speech recognition model built specifically for medical dictation and transcription. The company describes it as a foundational model for healthcare voice applications: it takes monochannel 16 kHz audio and outputs text-only transcriptions. Importantly, it is not positioned as a finished, plug-and-play clinical product; it is meant to be adapted by developers for their own settings and workflows.