Automatic Human Utility Evaluation of ASR Systems: Does WER Really Predict Performance?

Favre, B.; Cheung, K.; Kazemian, S.; Lee, A.; Liu, Y.; Munteanu, C.; Nenkova, A.; Ochei, D.; Penn, G.; Tratz, S.; Voss, C.; Zeller, Frauke

Abstract

We propose an alternative evaluation metric to Word Error Rate (WER) for the decision audit task of meeting recordings, which exemplifies how to evaluate speech recognition within a legitimate application context. Using machine learning on an initial seed of human-subject experimental data, our alternative metric handily outperforms WER, which correlates very poorly with human subjectsf success in finding decisions given ASR transcripts with a range of WERs.

Citation

Favre, B., Cheung, K., Kazemian, S., Lee, A., Liu, Y., Munteanu, C., Nenkova, A., Ochei, D., Penn, G., Tratz, S., Voss, C., & Zeller, F. (2013, August). Automatic Human Utility Evaluation of ASR Systems: Does WER Really Predict Performance?. Presented at Interspeech 2013, Lyon, France

• Presentation Conference Type: Conference Paper (published)
• Conference Name: Interspeech 2013
• Start Date: Aug 25, 2013
• Publication Date: 2013
• Deposit Date: Apr 29, 2023
• Pages: 3463-3467
• Book Title: Proc. Interspeech 2013
• DOI: https://doi.org/10.21437/Interspeech.2013-610
• Public URL: http://researchrepository.napier.ac.uk/Output/3085862