Clinical depression detection for adolescent by speech features

Depression is a common mental illness with a prevalence of 10-15% in general population. It is one of the most prevalent of all psychopathology disorders and if left untreated may lead to many negative consequences (i.e., suicidality, poor physical health, economic cost). Depression affects a person's psychosocial function and can eventually lead to family problems, problems in social life and loss of job. The increase in the prevalence of clinical depression in adolescents (i.e., those aged 13-20 years) has been connected to a scope of genuine results, especially an expansion in the quantity of suicide endeavors and passing's, making it a general wellbeing concern. Indications of depression often first appear during adolescence at a time when the voice is changing, in both males and females, recommending that particular investigations of these marvels in pre-adult populaces are justified. This study reports on an investigation of 14 acoustic features used as potential indicators of clinical depression that were grouped into five main feature categories(i.e Teager energy operator (TEO)-based, Cepstral (C), Prosodic (P), Spectral (S), Glottal (G) ) extracted from adolescents' speech using Distress Analysis Interview Corpus (DAIC) database. Most interestingly, the TEO-based features, when used on their own, clearly outperformed all other features and feature combinations. TEO together with glottal feature using GMM classifier outperformed all other feature combination with accuracy 90.58%. Also it is seen that TEO together with MFCC provide accuracy (85.88%) GMM, (76.4%) SVM and (76.5%) SVM+GMM classifier.