Study reveals speech recognition systems fail millions of dialect speakers daily — and the emotional cost of constant adjustment

arXiv cs.CL · 2026年4月24日

AI要約

•Researchers at four U.S. locations (Atlanta, Gulf Coast, Miami Beach, Tucson) documented how automatic speech recognition (ASR — the AI that converts spoken words to text) systematically fails speakers of regional English dialects, forcing them to repeat themselves or adjust their speech patterns to use basic voice features.
•Unlike past research that only measured error rates, this study captured what those failures actually feel like: participants described frustration, exclusion ('This wasn't made for me'), and the cumulative fatigue of constant workarounds — revealing that broken speech recognition isn't just a technical problem, it's an emotional burden.
•If you use voice assistants, voice-to-text dictation, or call center voice recognition and have a Southern, Gulf Coast, or regional accent, this explains why those tools work worse for you than they do for others — and shows that the AI industry has been measuring success in the wrong way (error rates) instead of measuring what matters (whether real people can actually use the technology without exhaustion).