Abstract: In order to improve the accuracy of speech endpoint detection in car environment, introduces a new measure of time series complexity, fuzzy entropy, and applies it to the characterization of speech. With sample entropy and fuzzy entropy respectively to the characterization of speech signals in car environments, and uses fuzzy C-means clustering algorithm and Bayesian information criterion algorithm, estimates the thresholds of the characteristics, then by using dual threshold method for endpoint detection. Experimental results demonstrate that, the fuzzy entropy method can distinguish the noise and speech signals better and has better performance of endpoint detection than sample entropy in car environments, the accuracy rate of fuzzy entropy method is superior to sample entropy method more than 16% in the same environments.

Key words: fuzzy entropy, sample entropy, speech endpoint detection, fuzzy C-means clustering algorithm, Bayesian Information Criterions

摘要： 为了提高车载噪声环境下语音端点检测的准确性，介绍了一种新的时间序列复杂性测度：模糊熵，并将其应用于语音信号的特征提取。分别以样本熵和模糊熵提取含噪语音信号的特征，使用双门限法对语音信号进行端点检测，特征门限值使用模糊C均值聚类算法和贝叶斯信息准则算法确定。仿真结果表明在车载噪声环境下与样本熵算法相比，模糊熵算法能更好地区分噪声信号和语音信号，具有更好的端点检测性能，相同环境下模糊熵算法的错误率比样本熵算法降低了16%以上。

关键词: 模糊熵, 样本熵, 语音端点检测, 模糊C均值聚类算法, 贝叶斯信息准则