Entropy-Based Vocal Biomarkers Improve Depression Detection Over Static Features

A recent study featured on arXiv introduces entropy-driven temporal vocal dynamics as innovative digital biomarkers for the automated detection of depression, showing superior performance compared to conventional static aggregation techniques. Analyzing the DAIC-WOZ corpus, which includes 142 labeled participants, researchers reconstructed acoustic trajectories at the utterance level and evaluated various feature sets using leakage-aware validation. The static pooling method yielded an AUC of 0.593, while trajectory dynamics improved this to 0.637. Notably, entropy biomarkers achieved the most significant enhancement (AUC 0.646; nested cross-validated AUC 0.615; permutation p = 0.017), outpacing recurrence quantification, sample entropy, fractal complexity, and coupling biomarkers. This research indicates that entropy-based temporal analysis reveals clinically relevant behavioral dynamics that static features may overlook, providing a more sensitive approach for depression detection.