Comparison between different feature extraction techniques to identify the emotion 'anger' in speech

Abstract

In this paper, three different techniques of feature extraction for identification of emotion in speech have been compared. Traditional feature like LPCC (Linear Predictive Cepstral Coefficient) and MFCC (Mel Frequency Cepstral Coefficient) have been described. Linear features like LFPC which is FFT based have been explained. Finally TEO (Teager Energy Operator) based nonlinear LFPC features in both time and freqnency domain have been proposed and the performance of the proposed system is compared with the traditional features. The comparison of each approach is performed using SUSAS (Speech Under Simulated and Acid Stress) and ESMBS (Emotional Speech of Mandarin and Burmese Speakers) databases. It is observed that proposed system outperforms the traditional systems. Analysis will be carried for identification mainly of the emotion 'Anger' in this paper. © Institute for Computer Sciences, Social Informatics and Telecommunications Engineering 2012.