A new paper on arXiv examines the persistent performance gaps in automatic speech recognition (ASR) systems, focusing on differences between standard and minority English varieties. The authors note that evaluations have repeatedly revealed disparities along lines of race, gender, socioeconomic status, and language variety.

While these gaps are well-documented, the paper argues that they remain an enduring problem rather than a solved issue. The work appears to synthesize existing evidence to highlight how ASR systems continue to underperform for speakers of minority English varieties, even as overall accuracy improves.

Because the abstract is truncated, the full methodology and proposed solutions are not yet visible. Still, the central message is clear: ASR disparities are not isolated incidents but a systemic pattern that requires continued attention.