Vocal acoustic parameters as objective biomarkers of major depressive disorder and its severity: a case-control study using WhatsApp audio messages

Depression lacks validated objective biomarkers in routine clinical practice. Voice is a non-invasive signal shaped by the neuromotor and psychomotor systems disrupted in depressive states, making it a promising candidate for objective assessment. We investigated whether acoustic parameters extracted from WhatsApp audio messages distinguish adults with major depressive disorder (MDD) from healthy controls and correlate with clinician-rated severity on the Montgomery-Åsberg Depression Rating Scale (MADRS). Seventy-two participants (31 MDD, 41 controls) sent WhatsApp audio in two tasks: structured counting (1-to-10) and spontaneous narrative (last week). Features were extracted with Praat (formants, shimmer, jitter, harmonic-to-noise ratio) and PyAudioAnalysis (68 features including MFCCs, spectral entropy, zero-crossing rate, and energy), and group comparisons were conducted using Generalised Estimating Equations with false discovery rate correction. Formant 1 was substantially reduced in MDD across both tasks (Cohen’s d = − 0.95 to − 1.03, q < 0.003). Shimmer was markedly elevated in the narrative task (d = + 1.08 to + 1.14, q < 0.001), and spectral entropy and zero-crossing rate were reduced (d up to − 1.33, q < 0.001). MADRS severity correlated most strongly with chroma-11 maximum (Pearson r = + 0.73, q < 0.001) and energy entropy interquartile range (r = + 0.67, q < 0.001). The spontaneous narrative task yielded 62 significant features versus 25 in the counting task. WhatsApp audio messages yield statistically robust vocal biomarkers of MDD detectable in naturalistic recordings on personal smartphones, with formant 1, shimmer, and spectral entropy emerging as the most promising candidates for scalable, non-invasive depression assessment.