What it is
You want what's said on a round or in a consult turned straight into text. But medical language breaks a consumer speech recognizer — drug names, surgical procedures, anatomy. GENSHI Voice recognizes first, then has an AI re-read the result in a medical context and repair what came out garbled. Turning a mishearing like「閉鎖コーヘルニア」back into「閉鎖孔ヘルニア」is the post-processing step's job.
It ships as a Python SDK, so in-hospital systems and MedLocal can call it from their own code.
From making sound to hearing it
At my previous job, Livetoon, I was on the making side — building speech synthesis from scratch. GENSHI Voice is the reverse: the listening side. Time spent down in the fine detail of waveforms on the making side turns into an instinct, on the listening side, for where things tend to break. Same audio, but do both the entrance and the exit and it looks different.
Recognition output is meant to be checked by a clinician before use. Proper nouns and figures especially aren't meant to be taken at face value.