Speech based emotion classification

In this paper, a speech based emotion classification method is presented. Six basic human emotions including anger, dislike, fear, happiness, sadness and surprise are investigated. The recognizer presented in this paper is based on the Discrete Hidden Markov Model and a novel feature vector based on Mel frequency short time speech power coefficients is proposed. A universal codebook is constructed based on emotions under observation for each experiment. The databases consist of 90 emotional utterances each from two speakers. Several experiments including ungrouped emotion classification and grouped emotion classification are conducted. For the ungrouped emotion classification, an average accuracy of 72.22% and 60% are obtained respectively for utterances of the two speakers. For grouped emotion classification, higher accuracy of 94.44% and 70% are achieved.