The Research On SVM Speech Emotion Recognition Based On Ant Colony Optimization

To achieve real artificial intelligence need to implement the "emotional intelligence" of the computer.Due to voice signal,which is the main carrier of information in the human daily communication,contains a rich emotional resources,speech emotion recognition has an important research in artificial intelligence.Under the basic framework of speech emotion recognition,the selection and improvement of the extraction or selection of emotional features and the classification and recognition algorithm are the two key points to the performance of the speech emotion recognition model and the most important in the research of speech emotion.Aiming at feature optimization and classification model design,this paper proposes to construct a Multi-layer SVM classification model with genetic algorithm for parameter optimization,and take the advantage of the improved ant colony algorithm's global multiple optimization search capability to achieve feature reduction,which effectively improves the performance of the speech emotion recognition system.The specific work is as follows:(1)This thesis introduces the commonly used emotion models and corpus in speech emotion recognition,and selects the pure and noiseless Berlin database based on discrete model as corpus.The speech signal was pre-weighted,windowed,frame segmentation,endpoint detection and other front-end processing,and then the characteristics of various feature parameters were analyzed and extracted,and the 167-dimensional emotional feature set was constructed through the calculation of statistical variables.(2)The support vector machine with excellent performance in high dimensional small samples is selected as the classifier,and the Principal Components Analysis and Genetic Algorithm,as two optimization methods which are commonly used in support vector machines,are studied.Through the experimental study,Genetic Algorithm was selected to optimize the penalty factor C and the kernel parameter a to construct the SVM classifier.The degree of confusion between emotional will reduce emotion recognition accuracy,the multi-layer SVM is proposed to solve this problem and more complex class emotion classification problem.(3)Considering that the selected feature vectors have different ability to represent emotions and there are redundant and useless vectors,the ant colony algorithm which takes the weighted function of emotion recognition rate and emotion feature vector as the fitness function is adopted to perform feature optimization and dimensionality reduction.In view of the traditional ant colony algorithm is easily trapped in local optimal solution and the phenomenon of the stagnation of the search,make analysis to its parameter adjustment,put forward the greedy and local organic search strategy,and implementation of ant colony algorithm for the elegant dimension reduction of feature vector sets,combined with multilevel SVM to build multi-layer SVM classification model based on ant colony optimization on EMO-DB for emotion recognition.