A Study On Negative Emotion Intensity Model Based On Speech

Speech emotion recognition is an important branch of artificial intelligence,has important research and application value in human-computer interaction to achieve harmony.The speech emotion calculation is widely concerned in the field of human-computer interaction,and the model of emotional intensity of speech is the core content of the study.However,due to the existence of a large difference in the way of describing emotion,and the social and multi-modal nature of emotion itself,both of them have caused great obstacle to the study of emotion intensity model of speech.In view of the above reasons,this thesis models and emotional speech recognition system were described in detail on the emotional description.And then establish the focus of training on the speech emotion intensity model for a detailed study and discussion,according to the Plutchik "emotion wheel" theory by the rank of the emotional strength to establish emotional intensity model using similarity algorithm.The division of the basic negative emotional intensity of emotional speech.Finally,the design experiment is used to identify the segmentation results of the emotion intensity model established in this thesis,and the average recognition rate is 90.4%.The main contents of this thesis are as follows:(1)The spectral clustering algorithm for optimization.This thesis measure similarity density sensitive,by reducing and enlarging the inter class data of within class data,construct a similarity matrix;Bagging algorithm,according to the different constraint conditions,select the feature vector optimal combination in Laplasse matrix,which solves the multi scale clustering and feature vector selection the problem,in order to achieve the optimization of the spectral clustering algorithm.(2)The spectral clustering algorithm based on the optimized training,and the establishment of the three order model.The emotional intensity of spectral clustering algorithm based on similarity,with basic negative emotional speech as data source,select and establish affective feature vector matrix,each column clustering of feature matrix,training and the establishment of three order emotion intensity model three order,emotional intensity division and use the model to realize the voice.Finally,the average recognition rate of 90.4% is obtained by SVM.This thesis is based on the basic negative emotional speech data,the emotional intensity of three order model established on the emotional intensity of training and the voice of the division and application of the model can be used not only in the field of speech emotion recognition,but also provides new ideas and new possibilities for the diagnosis of mental illness,and has the value of application and research.