Research Of Emotion Recognition In Speech Based On Multiple Classifiers Fusion Of SVM And CRF

In recent years, with people constantly improve on the requirements of human-computer interaction technology, affective computing has been the significant research directions in computer application fields. Speech is one of the most convenient means of communication between people and it is one of the fundamental methods of conveying emotion as well as semantic information. Therefore, as a branch of affective computing speech emotion recognition receive much more attention. With the technological development of various disciplines, speech emotion recognition has been made great progress, but because of the complexity of human emotions, there is still no acknowledged definition of human emotion. The building of emotional speech database, the selection and extraction of speech emotional characteristic parameters, and the speech emotional recognition have not formed systematic theory. Because these reasons, speech emotional recognition is still in the preliminary stage, it is need more deep research.In this paper, we firstly introduce the status of speech emotion recognition and its application in people’s life, then the difficulties of speech emotional recognition has been summarized. According to the research in different disciplines, different angles, we introduce many definition and classification of the emotions. In this paper Emotions are classified into four categories According to some known of emotional speech database. A small-scale emotion speech databases have been set up by cutting and recording. On this basis, eight features about energy, pitch and speech rate related features are extracted from speech signal.Speech emotion recognition is essentially classified as pattern recognition. Support vector machine classifiers in many of which also showed a strong classification ability, conditional random fields, a popular probabilistic method for structured prediction. It is not require by Hidden Markov Model of strong independence assumptions. This paper choose SVM, CRF and3categories to constitute sub-classifiers, the results of identify were fusion by Decision Templates. This paper compare the method based on single classifier and multi-classifiers. The experimental result shows that:multi-classifiers method’s recognition rate is better then single classifier, and the average recognition rate is much better then single classifier.