Accent related errors in clinical speech transcription and a LLM-based remedy

npj Digital Medicine 2026 Open access

Read full text
Authors: Yasaman Fatapour, Jamil S. Samaan, Nicholas P. Tatonetti, Aditi Kuchi, Apoorva Srinivasan, Sarvenaz Fatapour, Hongyu Liu, Jacob Berkowitz, Kevin K. Tsang, Michael Ziets, Nadine A. Friedrich, Nitin Srinivasan
+9 more Shehan Thangaratnam, Ryan C King, Ryan Czarny, Trini Nguyen, Yee Hui Yeo, Kim Hyunseok, Yi‐Te Lee, Nicha Wongjarupong, Arash Abiri
DOI: 10.1038/s41746-026-02490-z
Cited by: 2
References: 12

Abstract

Accurate clinical documentation is essential for safe, effective patient care. AI tools powered by automatic speech recognition can streamline this process. Variable performance across speakers with diverse accents leads to transcription errors and clinical risk. In testing Whisper and WhisperX on native and non-native English clinical speech, error rates were significantly higher for non-native speakers. Post-processing with GPT-4o restored lost accuracy. This chained approach (WhisperX-GPT) reduced accent-related errors.

Fields of Study & Subjects

Artificial Intelligence in Healthcare and Education Electronic Health Records Systems Radiology practices and education

Publication details

Journal / source
npj Digital Medicine
Publisher
Springer Science and Business Media LLC
Volume
9
Issue
1
Publication date
2 March 2026
Language
EN