Classification of Depressed Speech Samples with Spectral Energy Ratios as Depression Indicator

摘要

This research study aimed to investigate the characteristics of the Spectral Energy Ratios (SER) determined from the Power Spectral Density (PSD) of the spoken speech samples used to represent the severity level of emotional illness such as Depression in quantitative measure. Situation could be getting worst for a person who suffers from such illness with the elevated severity of symptom. When the symptom of severe depression strikes, a depressive person might be at high risk of committing suicide. The prevention of suicide is necessary for depressed persons to save life by admitting them in time and providing the proper treatment under supervision of clinical specialist. Prediction is primarily one of the most important tasks in the prevention of life-threatening risk from suicide. Researcher has attempted to adapt the speech processing techniques into a clinical diagnosis of emotional illness. In this study a full-band energy and further several sub-band energies estimated from the four frequency bands with each 625-Hz bandwidth were computationally extracted from the categorized speech samples and consequently formed the parameter models for classifications. As result shown, the averaged value of correct classification was obtained to be effectively approximate 80%, when training and validating classifiers with 35% and 65% of the extracted SER features, respectively.