---
title: MedASR Medical Speech Recognition
emoji: 🏥
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 6.2.0
app_file: app.py
pinned: false
license: apache-2.0
---

# 🏥 MedASR - Medical Speech Recognition Demo

This is a HuggingFace Space demo for the [MedASR](https://huggingface.co/google/medasr) model.

## About MedASR

MedASR is a speech-to-text model based on the Conformer architecture, pre-trained specifically for medical dictation. It is designed to handle:

- ✅ Medical terminology
- ✅ Radiology dictation
- ✅ Physician-patient conversations
- ✅ Various medical specialities

## Model Details

| Property | Value |
|----------|-------|
| **Model Type** | Automated Speech Recognition |
| **Architecture** | Conformer |
| **Parameters** | 105M |
| **Input** | Mono-channel audio @ 16kHz |
| **Output** | Text |
| **License** | Health AI Developer Foundations |

## Usage

1. Click the microphone icon to record audio or upload an audio file
2. Click the "Transcribe" button
3. View the transcribed medical text

## Performance

| Dataset | MedASR WER |
|---------|-----------|
| RAD-DICT | 6.6% |
| GENERAL-DICT | 9.3% |
| FM-DICT | 8.1% |
| MIMIC | 6.6% |

## Citation

bibtex
@inproceedings{wu2023last,
title={Last: Scalable Lattice-Based Speech Modelling in Jax},
author={Wu, Ke and Variani, Ehsan and Bagby, Tom and Riley, Michael},
booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
pages={1–5},
year={2023},
organization={IEEE}
}       

## Resources

- [Model Card](https://huggingface.co/google/medasr)
- [GitHub Repository](https://github.com/google-research/medasr)
- [Quick Start Notebook](https://github.com/google-research/medasr)
- [Fine-tuning Notebook](https://github.com/google-research/medasr)

## License

The use of MedASR is governed by the Health AI Developer Foundations terms of use.