Research Of Speech Emotion Recognition

In recent years, more and more researchers pay attention to emotional information in thevoice, the human-computer emotional interaction is an important research topic in the speechrecognition field. Speech emotion recognition technology is not only used to enhance thecomputer’s humanization and intelligence, to improve human-computer interaction experience,but also to improve the accuracy of speech recognition system. The main purpose of thispaper is to conduct the speech emotion recognition model on the base of the depth analysis ofa variety of voice features, and to analyze the reliability of the model by the speech emotionexperimentsThis paper introduces the background of speech emotion recognition research andbuilding a model of speech emotion recognition technology, the important part of the speechsignal preprocessing, feature extraction and analysis of voice emotion, the type of speechemotion recognition method and the current domestic and international voice EmotionRecognition Research and development.A detailed description of the process of building and design ideas for the speech emotionrecognition system is followed. The main content of the paper are:(1) Introduce the definition and classification of voice emotion. And then briefly descriptionof the classification and acquisition of speech emotion database.(2) Based on emotional speech database, a detailed analysis of the anger, surprise, joy,disgust, fear, and five kinds of emotional states such as pitch, the variation of theemotional characteristics of the spectrum, speed, etc., and select the most suitablecharacteristics parameters.(3) Based on the characteristic parameters selected to build the HMM and ANN combination of speech emotion recognition model. Because the poor classification ability of theisolated HMM, considering the method of combining the HMM and a posteriori classifier,to overcome the inherent shortcomings of the HMM, the experiments show thatrecognition rate of the combination of HMM and ANN model improved obviouslyCompared with the isolation HMM(4) In the case of small samples, SVM has the stronger ability of classification, but the SVMalgorithm can only apply to the two types of classification. In this paper, the method ofusing the binary tree SVM of the multi-class classification algorithm to constructmulti-class SVM classification model is proposed. The experiments prove that thisalgorithm can effectively improve the recognition rate.