Speech Emotion Recognition Based On Rough Set And SVM

Along with the rapid development of computer technology, the application of computer is almost used in every field. As using and communicating with computer have become more and more popular, the traditional human-computer interface (HCI) does not satisfy the widely application of computer. A harmonious, intelligent HCI becomes an important research area in computer technology. The design and realization of a computer which can recognize, understand, and express emotion is an import part of this research area.Affective computing is computing that relates to, arises from or deliberately influences emotions. These facts combined with abilities computers are acquiring in recognizing, understanding and expressing affect. The expression or physiological signals of human's emotions are obtained through different sensors, and "affective models" is used in affective computing to recognize these signals, understands human's affect and responds correspondingly.Speech emotion recognition is a key part of affective computing. The emotion features are extracted precisely from the wave signals by computer and used to recognize the emotion state. The word is never recognized in speech emotion recognition while it is recognized precisely in speech recognition which never considers the emotion information in a speech contour.Speech emotion recognition can be viewed as a pattern recognition problem. Such raw features as energy, pitch, speech rate, and formant are extracted from speech signal. Statistical features derived from these raw features are used as the input of a classifier. However, irrelevant and redundant features may exist and in some cases it may provide better classification accuracy by removing these unnecessary features.This paper uses Rough set theory and Support Vector Machine in emotion recognition of mandarin speech. We preprocess the SVM input training set by applying feature selection which is conducted by attribute reduction of rough set theory. By feature selection, the dimension of the input vector is decreased and the unnecessary features are removed. Then, SVMs are used to recognize the six basic human emotions such as happiness, angry, surprise, sadness, fear and normal. The recognition results and the analysis of recognition experiments are also reported. A speech emotion recognition system developed with C++ is introduced. At the end of this paper, we discuss about some problems that have not been solved in this paper and the future works in this field.