Emotion Classification using Dual-Tree Complex Wavelet Transform

Abstract

In this paper, a new method of feature extraction is proposed for speech emotion classification using dual-tree complex wavelet transform (DTCWT). DTCWT is a multiresolution technique, which decomposes the speech signal into number of sub-band signals. Different emotions have different impacts on frequency-bands. The feature extracted from DTCWT sub-band coefficients may therefore be better capture the emotion information of different sub-bands. The mean, skewness, kurtosis and energy values are evaluated from each sub-band coefficient, and used as a feature. The performance of the proposed feature is evaluated using two databases, EMODB and IEMOCAP. For classification purpose, support vector machine (SVM) classifier is used. Recognition results show that the proposed feature out-performs the linear prediction coefficients (LPC) and a Teager-energy-operator (TEO) based feature (TEO-CB-Auto-Env). Combination of the proposed feature with the mel frequency cepstral coefficients (MFCC) further increases the classification performance.