Quantification of stylistic differences in human- and ASR-produced transcripts of African American English

TL;DR

This paper examines how stylistic differences between human and ASR transcriptions of African American English affect accuracy measures like WER, highlighting the influence of transcription choices and linguistic features on ASR performance evaluation.

Contribution

It categorizes stylistic differences in multiple transcription types and analyzes their impact on WER, providing insights into ASR output variability for underrepresented language varieties.

Findings

Stylistic differences significantly affect WER comparisons.

Transcription choices influence ASR performance metrics.

Analysis clarifies how training data decisions impact ASR outputs.

Abstract

Common measures of accuracy used to assess the performance of automatic speech recognition (ASR) systems, as well as human transcribers, conflate multiple sources of error. Stylistic differences, such as verbatim vs non-verbatim, can play a significant role in ASR performance evaluation when differences exist between training and test datasets. The problem is compounded for speech from underrepresented varieties, where the speech to orthography mapping is not as standardized. We categorize the kinds of stylistic differences between 6 transcription versions, 4 human- and 2 ASR-produced, of 10 hours of African American English (AAE) speech. Focusing on verbatim features and AAE morphosyntactic features, we investigate the interactions of these categories with how well transcripts can be compared via word error rate (WER). The results, and overall analysis, help clarify how ASR outputs are a function of the decisions made by the training data's human transcribers.