Self-supervised Speech Representations Still Struggle with African American Vernacular English
This work addresses the underperformance of ASR systems for marginalized language varieties like AAVE, which reinforces stigmatization, but it is incremental as it shows SSL alone may not bridge the gap.
The study investigated whether self-supervised learning (SSL) speech models can reduce the performance gap in automatic speech recognition (ASR) between African American Vernacular English (AAVE) and Mainstream American English (MAE), finding that models like wav2vec 2.0, HuBERT, WavLM, and XLS-R perpetuate bias with higher word error rates on AAVE, especially for utterances with more AAVE features.
Underperformance of ASR systems for speakers of African American Vernacular English (AAVE) and other marginalized language varieties is a well-documented phenomenon, and one that reinforces the stigmatization of these varieties. We investigate whether or not the recent wave of Self-Supervised Learning (SSL) speech models can close the gap in ASR performance between AAVE and Mainstream American English (MAE). We evaluate four SSL models (wav2vec 2.0, HuBERT, WavLM, and XLS-R) on zero-shot Automatic Speech Recognition (ASR) for these two varieties and find that these models perpetuate the bias in performance against AAVE. Additionally, the models have higher word error rates on utterances with more phonological and morphosyntactic features of AAVE. Despite the success of SSL speech models in improving ASR for low resource varieties, SSL pre-training alone may not bridge the gap between AAVE and MAE. Our code is publicly available at https://github.com/cmu-llab/s3m-aave.