AI voice assistants encode gendered power asymmetries, study finds
Female AI voices labeled submissive and sexualized, male voices dominant in new study.
A new study accepted at Interspeech 2026 examines how sexualized synthetic voices from a popular commercial AI platform reinforce gendered power asymmetries. Researchers at KTH Royal Institute of Technology conducted a listening experiment with a diverse group of participants, asking them to evaluate male- and female-coded voices reading either sexualized scripts or neutral text. They collected quantitative adjective selections, qualitative free-text responses, and conducted acoustic analysis.
Results show female-coded voices are far more likely to be described with sexualized and submissive adjectives, while male-coded voices are associated with dominance and positive traits like confidence and authority. The authors argue this pattern encodes toxic masculinity and heteronormativity, limiting gender expression to a narrow binary. The work highlights a critical need for more diverse and ethical design in commercial voice AI systems.
- Female-coded AI voices are disproportionately labeled as 'seductive', 'submissive', and 'sensual' compared to male voices.
- Male-coded voices are more often associated with 'dominant', 'authoritative', and 'trustworthy' — even in neutral scripts.
- The study involved a diverse listener group and combined quantitative ratings, qualitative feedback, and acoustic analysis.
Why It Matters
Voice AI design risks entrenching gender stereotypes at scale, affecting millions of daily interactions.