DWT and MFCC based human emotional speech classification using LDA

Recent years, identification of gender based on emotional speech is one of the active research areas in developing intelligent human machine interactive (HMI) systems and biometric system. This work aims to identify the gender of the speaker through emotional speech. Two different features extraction methods such as Discrete Wavelet Transform (DWT) and Mel Frequency Cepstrum Coefficients (MFCC) are used for extracting the statistical features from the emotional speech signals. Three different value of MFCC coefficients (13, 15, and 20) and Daubechies wavelet function with three different orders (dB4, dB6 and dB8) in Discrete Wavelet Transform (DWT) were studied and compared to analyze their effect on emotional speech classification. Gender classification was done using Linear Discriminant Analysis (LDA) classifier. As a result of this study, 20 MFCC coefficient gives the highest classification accuracy (angry: 99.54%; happy: 99.76%; sad: 99.91%) on classifying three emotions compared to DWT. Complete comparison of two different feature extraction methods on classifying three emotional speech using LDA is given for justifying our system performance.