MedASR: An Open-Source Model for High-Accuracy Medical Dictation

TL;DR

MedASR is an open-source, high-accuracy medical dictation model designed to be small, fast, and effective, addressing data scarcity and improving transcription quality for healthcare applications.

Contribution

The paper introduces MedASR, a novel open-source model optimized for medical dictation, with innovative training and inference techniques for clinical transcription.

Findings

Achieves 58% relative WER reduction on Eye Gaze dataset

Addresses clinical data scarcity and class imbalance effectively

Provides a transparent, high-performance healthcare transcription backbone

Abstract

We present MedASR, an open-source 105M-parameter model engineered for high-accuracy medical dictation. Prioritizing a "small, fast, and accurate" design, MedASR addresses 3 core pillars (1) Data: overcoming clinical corpora scarcity and class imbalance; (2) Modeling: efficient long-form training; and (3) Inference: accurate transcription via a pseudo-streaming sliding-window approach. Our evaluation shows that MedASR achieves a 58% relative WER reduction on Eye Gaze compared to Whisper Large-v3. By open-sourcing MedASR, we provide a transparent, high-performance backbone for specialized health-care applications, breaking down the barriers to clinical documentation often obscured by proprietary systems.