A Sociophonetic Analysis of Racial Bias in Commercial ASR Systems Using the Pacific Northwest English Corpus

TL;DR

This study systematically evaluates racial bias in four major commercial ASR systems using the Pacific Northwest English corpus, linking transcription errors to sociophonetic variation through a novel Phonetic Error Rate metric. By annotating eleven sociophonetic features and applying both traditional WER and PER, the authors demonstrate that vowel-quality variation, especially low-back and pre-nasal mergers, systematically drives errors, with African American speakers most affected. The methodology combines regionally adapted phonetic representations (PNWEdict), controlled WL data, and linear mixed-effects analysis to reveal language-phonetics-driven biases that persist across platforms, rather than architecture-specific issues. The PNWE corpus is established as a valuable resource for bias evaluation, with actionable guidance for data collection and modeling to mitigate dialectal biases in speech technologies.

Abstract

This paper presents a systematic evaluation of racial bias in four major commercial automatic speech recognition (ASR) systems using the Pacific Northwest English (PNWE) corpus. We analyze transcription accuracy across speakers from four ethnic backgrounds (African American, Caucasian American, ChicanX, and Yakama) and examine how sociophonetic variation contributes to differential system performance. We introduce a heuristically-determined Phonetic Error Rate (PER) metric that links recognition errors to specific linguistically motivated variables derived from sociophonetic annotation. Our analysis of eleven sociophonetic features reveals that vowel quality variation, particularly resistance to the low-back merger and pre-nasal merger patterns, is systematically associated with differential error rates across ethnic groups, with the most pronounced effects for African American speakers across all evaluated systems. These findings demonstrate that acoustic modeling of dialectal phonetic variation, rather than lexical or syntactic factors, remains a primary source of bias in commercial ASR systems. The study establishes the PNWE corpus as a valuable resource for bias evaluation in speech technologies and provides actionable guidance for improving ASR performance through targeted representation of sociophonetic diversity in training data.