Abstract: Speech endpoint detection is one of the important links of speech recognition system. In view of the question that the speech endpoint detection is difficult in noisy environment, this paper proposes a kind of speech endpoint detection method based on improved Support Vector Machine（SVM）. First, it extracts the feature vector of the noisy speech signal using the Wavelet Analysis（WA）. Then, it gets the optimal parameters of the SVM kernel function [γ] and penalty factor [C] using the Genetic Algorithm（GA）. Finally, it establishes speech endpoint detection model. Carried out the simulation experiments in the Matlab software platform, the results show that the average detection rate of GA-SVM is 94.5% under the condition of different noise. It is superior to the traditional double threshold algorithm and ordinary SVM algorithm.

Key words: Wavelet Analysis（WA）, Support Vector Machine（SVM）, Genetic Algorithm（GA）, speech endpoint detection

摘要： 语音端点检测是语音识别系统的重要环节之一。针对噪声环境下的语音端点检测困难，提出了一种改进的支持向量机的语音端点检测方法。利用小波分析（WA）提取含噪语音信号的特征向量。采用遗传算法（GA）得到最优的SVM核函数参数[γ]和惩罚因子[C]。建立语音端点检测模型。在Matlab软件平台下进行仿真实验，结果表明在不同的噪声条件下，GA-SVM算法的平均检测率达到94.5%，明显优于传统的双门限算法和普通的SVM算法。

关键词: 小波分析（WA）, 支持向量机（SVM）, 遗传算法（GA）, 语音端点检测