Application of Support Vector Machine with Modified Gaussian Kernel in A Noise-Robust Speech Recognition System

To improve the generalization ability of the machine learning and solve the problem that recognition rates of the speech recognition system become worse in the noisy environment, a modified Gaussian kernel function which may pay attention to the similar degree between sample space and feature space is proposed. In this paper, used the modified Gaussian kernel support vector machine to a speech recognition system for Chinese isolated words, non-specific person and middle glossary quantity and chose the improved noise-robust MFCC parameters as the speech feature, used "one-against-one" method for the multi-class classification problem of SVM, and analyzed the influence of Gaussian kernel parameter gamma and error penalty parameter C on SVM generalization ability. Experiments indicate that the recognition rates of SVM which chose the best parameters and modified Gaussian kernel are much better than those of traditional HMM model and RBF network. The robustness is better too.