Comparing statistical classifiers for emotion classification

Speech emotion recognition has been widely used in human computer interaction and applications. This paper has classified emotion into two classes: happy and angry. All the speech signal is preprocessed from Malay spoken speech database. Emotional information is obtained by applying two well-established acoustical features that are Mel Frequency Cepstral Coefficients (MFCC) and Short Time Energy (STE). The performance of the classification is done by comparing four types of classifiers which are Naïve Bayes, Multi-Layer Perceptron (MLP), C4.5 and Random Forest. Result shows that Random Forest has achieved the highest accuracy of ~90% exceeding C4.5, Multilayer Perceptron (MLP) and Naïve Bayes. Naïve Bayes shows the lowest score of ~76% accuracy.