"It's not a representation of me": Examining Accent Bias and Digital Exclusion in Synthetic AI Voice Services

Technical performance varies significantly by accent: The study found disparities in how well Speechify and ElevenLabs performed across five regional English-language accents, meaning some users experience lower-quality synthetic speech depending on how they speak.

AI voice services may reinforce linguistic discrimination: Current speech generation technologies risk amplifying existing accent-based prejudices by performing better for some accents than others, potentially disadvantaging speakers of underrepresented regional or linguistic varieties.

User perception matters as much as technical quality: Through surveys and interviews, researchers discovered that users' lived experiences with accent discrimination shape how they perceive and react to accent variations in AI voices—suggesting that technical fixes alone won't solve the problem.

Digital exclusion is a real consequence: The disparities in performance across accents create new barriers to accessing AI voice services equitably, meaning some people may be effectively excluded from or have worse experiences with these increasingly common technologies.

Developers and policymakers need actionable guidance: The study calls for inclusive design practices and regulation of speech technologies, indicating that responsible AI requires both technical improvements and policy oversight to ensure these tools don't perpetuate discrimination.