Research Of Selective Ada Boost SVM Speech Emotion Recognition Algorithm

As an important member of the human-computer interaction techniques, speech emotionrecognition technology is widely used in education, health, communication, computer,automation and other industries. Meanwhile, the knowledge involved in speech emotionrecognition is very broad, including computer science and technology, pattern recognition,phonetics, psychology, statistics and signal processing. It has good research foundations andbroad developing prospects.Currently, the research of speech emotion recognition has made a lot of achievements. Butthere are also many difficulties. By improving the accuracy of the classification algorithm,we can improve the performance of products and systems using speech emotion recognitiontechniques. So they can provide better service and user experience. This will also improvethe quality of work and efficiency of some industries which are important for promoting thedevelopment of many industries.In the research of speech emotion recognition, the SVM classification algorithm has agood performance, while the AdaBoost algorithm can further improve classification accuracybase on SVM algorithm.This article is based on SVM and AdaBoost algorithm. Proposing a new ensemble learningalgorithm, namely the Selective AdaBoostSVM algorithm. The idea of the algorithm is: First,use AdaBoost algorithm to train a number of SVM classifiers. Then use the Kmeansalgorithm to cluster these classifiers and obtain some representatives classifiers. For eachtesting sample, use the Knn algorithm to find its nearest neighbor training samples from thetraining set. Then get the training samples into representative classifiers and test them.Finally, choose the classifier reaching highest classification accuracy to be the final classifierof the current test sample.This article tests the accuracy of the algorithm in EMO-DB German speech database,CASIA Chinese speech library and SAVEE English speech library. First, find the optimalSVM parameters of these three speech libraries in five-fold and ten-fold cross-validation. Then start five-fold and ten-fold cross-validation to test these three speech libraries.The experimental results show that the algorithm proposing by this article can improve theclassification accuracy of these three speech libraries.In five-fold cross-validation, compare with single SVM algorithm, the SelectiveAdaBoostSVM algorithm improves the classification accuracy of three speech libraries by1.86%,1.51%and3.77%. And compare with AdaBoostSVM algorithm, the SelectiveAdaBoostSVM algorithm improves the classification accuracy by0.35%0.78%and0.13%.The final classification accuracy of Selective AdaBoostSVM algorithm in three speechlibraries are87.56%,81.75%and76.75%.In ten-fold cross-validation, compare with single SVM algorithm, the SelectiveAdaBoostSVM algorithm improves the classification accuracy of three speech libraries by1.46%,0.74%and1.86%. And compare with AdaBoostSVM algorithm, the SelectiveAdaBoostSVM algorithm improves the classification accuracy by0.21%0.37%and1.86%.The final classification accuracy of Selective AdaBoostSVM algorithm in three speechlibraries are87.29%,81.20%and76.51%.The results show that Selective AdaBoostSVM algorithm proposed in this article is feasibleto improve the classification accuracy of speech emotion recognition.