For the low accuracy of traditional speech signal endpoint detection in non -stationary noise environment , we propose an improved speech endpoint detection method which is based on dynamic characteristics analysis of speech signal .Different from the traditional method of speech signal which analyses the speech signals in the way of deeming them as short -time stationary signals , this method faces the characteristics of non-stationary signal of the speech signal directly , and carries out recursion quantitative analysis on audio signals .After the decision is made on dual threshold of recursion degree , the assisted deterministic judgement will also be done for improving the accuracy of sonant detection .Experiments prove that compared with the traditional short-term energy method and the cepstrum distance metric method , the accuracy is greatly improved by applying this method to the speech signal endpoint detection , in particular , in low SNR environment the accuracy improves 3%~8%.%针对非平稳噪声环境中，传统语音信号端点检测方法正确率不高，提出一种基于语音信号动力学特征分析的改进型语音端点检测方法。不同于传统方法将语音信号看作短时平稳信号进行分析，该方法直面语音信号的非平稳特性，对音频信号进行递归定量分析。在对递归度双门限判决后，为了提高浊音检测正确率再辅之以确定性判定。实验证明，与传统短时能量法、倒谱距离测度法相比，利用该方法进行语音信号端点检测，正确率大大提高，特别是在低信噪比环境中正确率提高3％～8％。
展开▼
