Speech Emotion Recognition with Class-dependent Feature Selection Methods

This study proposes a novel class-dependent feature selection model to improve the performance of speech emotion recognition. Particularly, we adopt the markov blanket technique to select discriminative features for each emotion class, and use the support vector machine classifier to build the emotion recognition model. To make better decision in multi-label classification, the binary discriminative output of the support vector machine is transformed to a probability output for solving the voting conflict problem. Extensive experimental results on the a publicly available dataset show that in comparison with information gain, principal component analysis, and class-independent feature selectors, the proposed method significantly reduces the feature dimensionality and obtains better classification accuracy.