中文摘要：

阐述了支持向量机的分类机理，采用改进的MFCC语音特征参数，用基于不同核函数的支持向量机（SVM）作为语识别网络，对SVM多类分类问题采用“一对一”分类算法，实现了一个孤立词非特定人中等词汇量的抗噪语音识别系统。通过实验，得到了不同核函数下的识别结果；分析了核参数和误差惩罚参数对SVM推广能力的影响，并将实验结果同基于RBF神经网络的识别结果进行了比较。

英文摘要：

The classification principle of support vector machine was elucidated. Using improved MFCC speech characters and taking different kernel function based support vector machine as the recognition network for speech recognition system, a one-against-one method for multi-class support vector machine was adopted to realize a noise-robust speech recognition system for isolated words, non-specific person and middle glossary quantity. By experiments, the recognition results based on different kernel functions were obtained, the influences of the kernel parameter and the error penalty parameter on support vector machine＇s generalization ability were analyzed, and the different kernel based SVM speech recognition correct rates were compared with these obtained using RBF network in different SNRs.