Recurrence-Based Nonlinear Vocal Dynamics as Digital Biomarkers for Depression Detection from Conversational Speech

A new study from Himadri S Samanta, published on arXiv, proposes recurrence-based nonlinear vocal dynamics as digital biomarkers for depression detection. Using the DAIC-WOZ corpus with 142 labeled participants, the researchers modeled frame-level COVAREP trajectories as nonlinear dynamical systems, extracting recurrence-based biomarkers from 74 vocal channels. Logistic regression with feature selection and stratified cross-validation achieved a mean AUC of 0.689, outperforming static acoustic baselines, entropy-dynamics features, Hurst exponent features, determinism features, and Lyapunov-like instability proxies. Permutation testing confirmed statistical significance at p=0.004, with a pooled cross-validated AUC of 0.665 (95% bootstrap CI: [0.568, 0.758]).

This approach moves beyond conventional static acoustic descriptors by capturing the temporal organization of vocal state trajectories. The key insight is that depression alters how the vocal system revisits acoustic states over time—a subtle but measurable signal. While the AUC of 0.689 is modest, it significantly outperforms existing methods and suggests that nonlinear state-space analysis could complement traditional diagnostic tools for mental health. The work supports the development of scalable, non-invasive digital screening technologies that could be deployed via telehealth or mobile apps, potentially improving early detection and monitoring of depression in clinical and remote settings.