Research On Improved Support Vector Machine Classification Algorithm In Speech Recognition Application

With the rapid development of the information industry and the Internet, higher intelligent level is required by the information society. Speech recognition technology is a fast and convenient way of exchange of information, the object of which is speech. The final of it is to make sure that people can communicate with machine smoothly and naturally, providing facilitate life condition for people.Compared with the traditional method of speech recognition, Support Vector Machine (SVM) has better generalization performance and higher recognition efficiency, so it has been widely used in the field of pattern recognition. After decades of development, support vector machine theory has been greatly improved, on the basis of the original algorithm, new algorithms have been put forward. In this paper, support vector machine algorithm is studied especially. the main work is as follows:(1) First, briefly introduces the basic principles and methods of speech recognition. In the analysis of the inadequacies of traditional speech recognition methods, current popular machine learning method is brought in, which is SVM. Then, expounds the statistical learning theory. The nature of the support vector machine is to solve a quadratic programming problem, on the situation of linear inseparable, the original sample is mapped to a high dimensional nuclear space according to nonlinear mapping. As long as choosing the appropriate kernel function, the classification function of higher dimensional space can be obtained, so as to achieve linear separable.(2) For the shortcomings of the traditional one-against-one support vector machine algorithm in the forecast period, the improved algorithm has been proposed. In the middle of the testing phase, statistics votes of all the categories, the lower of which will be removed, and the sub-classifiers constituted by these categories is not caculated, so the recognition efficiency of algorithm can be improved. Finally, the improved method is applied to speech recognition system with noise immunity when in different vocabularies and different SNR. The experiment results show that the method has certain advantages.(3) Support vector machine has higher recognition rate in the condition of a small sample sets and higher SNR. Otherwise, the results is unsatisfactory. In order to solve these problems, training samples are pruned firstly by k-Nearest Neighbor algorithm. The purpose of pruning is to delete noise points or outliers, make sure that the separating hyperplane is as simple as possible, improving training speed. After pruning, training and recognition are carried out by standard Support Vector Machine. The result of experiment indicates that the number of training sets and support vectors have been reduced greatly after pruning. While maintaining higher recognition rate, training seed is speeded up distinctly.