SPEECH EMOTION RECOGNITION USING CLASSIFICATION ALGORITHMS

Abstract

Emotion Recognition from one’s speech is natural activity in human beings. Emotion recognition aims at identifying the emotional state of a speaker from his/her speech signal. The emotion recognition is useful in applications that are lie detection, in car board system, authentication systems and automatic emotional detection in call centers. There are different categories of emotions such as joy, fear, disgust, surprise, anger, sadness, boredom and neutral. In this proposed work, emotional speech files are collected from Berlin Emotional Speech Database (EMO-DB) covering exclusively 3 emotions Neutral, Anger and Sad. Information on emotion is encoded mainly phonetic and acoustic properties of spoken language. Prosodic features and voice quality also infers emotion characteristics. The emotion speech files are processed to extract features like energy, pitch, intensity and Mel-Frequency Cepstral Coefficient (MFCC). Emotion recognizer is designed with classifiers like Multilayer Perceptron (MLP) and Support Vector Machine (SVM). The experiment carried out for male and female speech files with acoustic features separately and acoustic features along with short term spectral features. The performances of the classifiers are evaluated with predictive accuracy.