Speech emotion recognition for tamil language speakers

Abstract

Emotional information in speech signal is an important information resource that expresses the state of mind of a person. For robots to plan their actions autonomously and interact with people, recognizing human emotions is a critical parameter required for emotion recognition. In this paper, the emotion is recognized and identified from Tamil speech signal of the speaker so as to make conversation between human and computer more native and natural. The most human non-verbal cues such as pitch, loudness, spectrum and speech rate are efficient carriers of emotion. In this proposed work, there are three important aspects of designing a speech emotion recognition system. For developing a speech emotion recognition system, three aspects are taken into consideration. They are database, feature extraction and emotion classifiers. A database is created for Tamil language, and the Mel-Frequency Cepstral Coefficient (MFCC) and energy are calculated from both recorded audio signal and real-time audio signal. From these extracted prosodic and spectral features, emotions are classified using support vector machine (SVM) with overall 85.4% accuracy.