Speech emotion recognition based on deep belief network

摘要

The paper introduces the continuous model and the discrete model of speech emotion recognition, and then it introduces the popular speech emotion databases. This paper analyzes different characteristics to make a better description of speech emotion. The main works of this paper are the selection of the database, the extraction of emotion features, and the selection of classification algorithm. Then, two methods are used to evaluate the result, including overall and average recognition rate. This paper uses contrastive divergence algorithm on emotion feature extraction. Compared with the traditional algorithms, such as support vector machine (SVM) and artificial neural network (ANN), the accuracy of test emotion sample has a better performance after feature extraction by DBN, to about 5% higher than traditional classification algorithm.