Quantification of stylistic differences in human- and ASR-produced transcripts of African American English
This addresses evaluation challenges for underrepresented speech varieties, though it is incremental in focusing on categorization and analysis rather than new methods.
The study analyzed stylistic differences between human and ASR transcripts of African American English, finding that these differences impact word error rate comparisons and clarify how ASR outputs depend on human transcriber decisions in training data.
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.