Speech emotion detection using supervised, unsupervised and feature selection algorithms

Abstract

A tremendous research is being done on Speech Emotion Recognition (SER) in the recent years with its main motto to improve human machine interaction. In this thesis work,we have introduced a scheme for emotion recognition from speech. We have classi ed three emotions (happy, angry and sad) for both male and female. Recognition task has been done using Mel-frequency Cepstrum Coe cient (MFCC) based features.Four classi ers are used for the purpose of classi cation. They are Random Forest, Gradient Boosting, SVMand CNN. Among them, CNN has shown the best accuracy of 71.17%. Random Forest has shown an accuracy of 61.26%, Gradient Boosting 60.36% and SVM60 36%. After using RFE method, PCA and P-Valuefor less signi cant feature reduction the accuracy improved to 62.16% for Random Forest, 62.16% for Gradient Boostingand 61.26% for SVM.