Speech Based Emotion Recognition Using Combination of Features 2-D HMM Model

In this article, we study how the numbers and statistical values of speech characteristics affect the precision of recognition of feelings from human speech. We recognize two efficient characteristics with Gaussian Mixture Model (GMM), namely Mel Frequency Cepstrum Coefficients (MFCCs) and Auto Correlation Function Coefficients (ACFC) found directly from the audio signal. Using GMM super vector molded by importances of MFCCs with berlin emotional folder considering six earlier proposed emotions: fear, anger, happy, disgust, sad and neutral. Our technique accomplished an emotion recognition rate of 74.45%, suggestively better than 59.00% achieved formerly. We also perform experiments with a distinct set of feelings to demonstrate the wide applicability of our technique: anger, boredom, fear, happiness, neutrality and sadness. Our 75.00 percent emotional identification rate again exceeds 71.00 percent of the concealed markov model technique with MFCC, delta MFCC, cepstral coefficient and speech energy.