Model reference · open weights

medasr

Audio google Speech→text 1 build Its own licence terms 70k dl/mo

medasr is an open-weight audio or speech model from Google. medasr (FP32) weighs 211 MB; the smallest configuration that runs it is RTX 3060 12 GB.

MedASR is a 105M parameter speech-to-text model developed by Google, based on the Conformer architecture and pre-trained on approximately 5,000 hours of English medical dictation.
It is designed for developers in the healthcare sector to use as a starting point for applications involving medical terminology, such as radiology dictation.
The model accepts mono-channel 16kHz audio input and outputs text, with its use governed by the Health AI Developer Foundations terms of use.

Summary of the google/medasr model card, 2026-10-01

What it is

What it runs on

Memory and cards for medasr (FP32)

211 MBweights, file size

1.6 GBruntime overhead

How it works

Details

Languages