Application for Detecting Depression, Parkinson’s Disease and Dysphonic Speech
Abstract
In this Show&Tell presentation we demonstrate an application that is able to assess a voice sample according to three different voice disorders: depression, Parkinson’s disease and dysphonic speech. Affection probability of each disorder is analyzed along with their severity estimation. Although the acoustic models (support vector machine and regression models) are trained on Hungarian voice samples, English samples can also be utilized for assessment. The results are displayed by as pie chart for probabilities and separate severity scores. The input of the application is a read text with a fixed linguistic content. It is possible to load a pre-recorded voice sample or create a live recording. The developed system could evaluate a speaker’s voice sample, assisting medical staff.