Exploring the Complexity of Parkinson's Patient Speech for Depression Detection task: A Qualitative Analysis

This study examines the acoustic features of Parkinson's patients with depression. More specifically, the research investigates whether interpretable, handcrafted acoustic feature-based methods, previously used for automatic speech-based depression detection, can be applied to detect depression in PD patients. We approach this by conducting a comparative study of speech-based depression detection tasks, using the DAIC-Woz corpus for typical speech and the PD-Depression corpus for atypical speech. We investigate how the acoustic descriptors of typical depressed speech differ from those extracted from the speech of PD patients suffering from depression. Our finding indicates that while typical depressed speech exhibits pronounced fluctuations in pitch and vocal stability, the speech of Parkinson's patients presents a more varied range of spectral features, reflecting the complexities of their condition caused by hypokinetic dysarthria.