A Novel Approach for Depression Detection from Speech using MFCC and Machine Learning Algorithm

Abstract

Gloom is a typical psychological sickness. It is one of the most common of all psychopathology issue and whenever left untreated may prompt many negative outcomes (i.e., suicidality, poor physical wellbeing, monetary expense). Living in depressive state is one of the main reasons of unhealthy and disturbed life. Depression detection process can be split into two primary groups: concentrating on an individual’s performance and concentrating on its status.  So, a novel method is proposed for predicting depression using machine learning and the objective of this research work is to acknowledge individual’s state with high performance and to predict depression before the person actually gets into depression. In this research work, depression is evaluated by real-time audio sections, and therefore the proposed approach detects the state of an individual by analyzing the voice showing that they are depressed or not. Audio segments captured are analyzed by the use of audio processing technique Mel-frequency cepstral coefficients. In past decades, lot of work has been done on depression detection but in this research work various moods of an individual are observed by the voice irrespective of the words spoken and language used, predicts state of depression with the depression score and with good performance. Moreover, mental alertness is connected with depression so, it is linked with voice with the help of the moods of a person. In addition to it, depression levels using speech provides levels as severely depressed, moderately depressed and not depressed. The procedure of information gathering includes multiple interviews. It can be split up into three categories on the basis of its causing emotion: positive, negative and unbiased emotion which is used for training the model and real time speech from individuals is used for testing the model. The age group for real time data collection is above 18 years. With the help of AI calculations, the paper presents MFCC feature extraction and decision level hybrid approach for the equivalent. Classification is done on fused as well as independent features using Multi Support Vector Machine (MSVM). The results crossed the given standard on validation data by 15%. on sound highlights. Therefore, the proposed model opens up the possibilities for better managing and controlling the health-related activities.