InfoShield: Privacy-Preserving Speech Representations for Mental Health Screening via Information-Theoretic Optimization

In three linesInfoShield minimizes mutual information between speech representations and sensitive attributes (gender, age) for depression screening. TimeAwareMINE with cross-modal attention reduces gender inference from 92.6% to 55.5% and age inference from 55.7% to 30.3%, achieving F1=0.784 on Androids Corpus.

Summary generated by Claude — human-verified