InfoShield: Privacy-Preserving Speech Representations for Mental Health Screening via Information-Theoretic Optimization
Researchers have developed InfoShield, a novel method for privacy-preserving speech analysis aimed at mental health screening. This technique uses information-theoretic optimization to reduce the correlation between speech representations and sensitive user attributes like gender and age. Experiments demonstrate that InfoShield significantly lowers the accuracy of inferring these attributes from speech while maintaining high performance in detecting depression. AI
IMPACT Enhances privacy in AI-driven mental health tools, potentially increasing user trust and adoption.