Depression Detection

The model was better at classifying controls than identifying depressed participants. Depression can affect multiple speech properties, including prosody, energy, timing, articulation, speech rate, and vocal expressivity. The study hypothesizes that depression changes the recurrence structure of vocal trajectories. Depression may be reflected in altered state-space recurrence organization of conversational vocal behavior. The supervised analysis used PHQ-8 binary labels to distinguish depression-positive and depression-negative participants. The entropy classifier was framed as a high-specificity risk-stratification aid rather than a diagnostic system. The study addresses the need for scalable and objective computational screening tools for major depressive disorder. Speech-derived digital biomarkers are presented as a scalable, passive, and more objective way to assess depression.