The code for Clinical BERTScore: An Improved Measure of Automatic Speech Recognition Performance in Clinical Settings, an Interspeech 2023 submission.

The main contributions of this work are:

1. The Clinician Transcript Preference (CTP) dataset, consisting clinician preferences describing which transcript errors are less acceptable in a clinical context. The data can be found here.
2. A metric of textual similarity that correlates much more strongly with clinician preferences, and incorporates medical knowledge. It can be used to train or evaluate ASR models in clinical settings.

This repo contains:

1. Code for reading the CTP.
2. Code for analyzing the CTP.
3. Code to run inference with CBERTScore.