Svm-based Speech Emotion Recognition

With the development of computer technology, human-computer interaction obtains more and more attention; as a result, researchers need to acquire higher communication skills. There is no doubt that emotions in communication is very important, and voice is overall the most direct way, so the emotion information in voice signal also gained researchers’ attention, especially on voice emotion recognization and voice emotion synthesis. In the research of voice emotion recognition, there are two main approaches to improve recognition rate, one is to improve the extraction or selection method in speech characteristics, and the other is to improve, or select better classification methods.This paper briefly introduced voice emotion recognization workflow, also done research about those two approaches to some extent.The paper focused on the voice feature extraction and voice emotion reorganization based on SVM. Main contents are as follows:1. Pre-processing of voice signal. The sentences were aggravated; window was added and endpoint detection processing was also applied. Mainly focused on comparison and analysis of varies results due to different ways to add window, preparing for experiments later on.2. Extraction of voice emotion characteristic parameters. Studied plenty of literature, and compared the choices of all kinds of emotional specification, the experiments are done for the purpose to compare the pitch extraction method and the resonance peak extraction method. We improved our research plan according to the methods proposed by previous researchers, and we get decent results.3. Study of classification methods. The main goal of this paper is to research on SVM classification method, analyzes the structure, principal and weakness of that, studied the improvement of SVM method. The experiments proved the effectiveness of SVM method in voice emotion reconginzation.