Resaerch On Techniques Of Visualization Theory Based Speech Emotion Recognition

Speech Emotion Recognition research is an important subject in Emotion Computing, it has a wide range of applications in the intelligent human-computer interface designed to improve the accuracy of speech recognition and emotional translation. The visualization method as a new pattern recognition method is also paid more and more attention by researchers. The main work of this paper is to introduce the multivariate data graph to data representation and recognition of speech emotions, in order to enhance the visual effect, improve people's participation in the recognition process, and improve the final recognition performance through the optimization and improvement of multivariate data graphs.This paper firstly analyzes the acoustic characteristics of speech commonly used to analyze their ability to differentiate emotions. We finally selected short-term energy, pitch, speech speed, formant parameters and their derived emotional features. Next, we use multivariate data graph to visualize these data of emotion features. First we use radar plot, and research the ordering problem of radar plot. We compare two ordering algorithms and advance a new sorting algorithm, template-feature based optimization algorithm. And then use a more emotional image, Chernoff faces to represent the data of the features. In order to express the emotional meaning of the data, we retain the good nature of the Chernoff faces and also make some breakthroughs. We use least squares optimization to obtain the mapping between the data and Chernoff faces. Finally, in order to combine the advantages of radar plots and Chernoff faces, we advance a fusion method called dressed Chernoff faces, in which way we get the best results. In the process of using multiple data graphs, we optimize the expression of graphs and improve the algorithm. And last we make a fusion of these methods into dressed Chernoff faces, and get a complete system of multiple data graph. Finally, in the experiments on our corpus, the recognition results are much better than not using the method of multivariate data graphs, and achieve satisfactory results. The results confirm the effectiveness of the visualization method in speech emotion recognition.