Combining clustering and classification approaches for speech-based emotion recognition problem

Communication is an important ability of a human, which is based on linguistics and the emotional component. In the field of technology, the emotion recognition is still a challenge, especially when the recognition is based solely on the voice, which is the primary means of human communication. Selecting of relevant features for automatic classification and recognition is an important step. Recognition efficiency of speaker’s emotions depends on the database used in the system. Recognition of speaker’s emotions is a difficult task, since it requires a set of consecutive operations, such as voice activity identification, feature extraction, training and classification. Speech-based emotion recognition is one of the most popular and common task in the field of the computer linguistics. In this area, the main criterion is the accuracy of the classification procedures. In current work, a variety of data mining techniques, such as artificial neural networks, logistic regression, support vector machines, are proposed to solve the problem of automatic emotion recognition. To improve the performance of emotion recognition we used pre-clustering and classification approaches. The method of principal component analysis is used for selecting important features. Testing of the proposed approach was carried out with the task of emotion recognition based on acoustic characteristics.