基于快速样本熵计算的清浊音判决与语音分割  

Discrimination and Segmentation of Voiced/Unvoiced Based on Fast Computation of Sample Entropy

作　　者：孙桂琪 庄晓东[1] 范珍艳 SUN Guiqi;ZHUANG Xiaodong;FAN Zhenyan(School of Electronic Information,Qingdao University,Qingdao 266000,China)

机构地区：[1]青岛大学电子信息学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2018年第4期98-103,共6页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：山东省自然科学基金项目(ZR2016FM11);山东省高等学校科技计划项目(J15LN41)

摘　　要：为了对语音帧的清浊音属性进行判断,本文提出了一种基于快速样本熵的清浊音判决和语音分割方法。通过计算英语单音素发音的样本熵,可以发现清浊音的信号复杂度有明显的区别,并根据复杂度的不同来进行清浊音的判决。同时,在快速算法中将数值二值化,由低维信号矢量的近邻矩阵递推高维信号矢量的近邻矩阵,可以快速有效的进行语音分割,并进行仿真实验。仿真结果表明,与其他传统方法相比,基于样本熵的方法可以得到较好的语音分割结果,而且快速样本熵算法将运算时间缩短了80倍,明显减少了运算时间,提高了运算效率。该研究在语音信号方面具有较好的应用前景。This paper presents a method of voiced/unvoiced decision and speech segmentation based on fast sample entropy algorithm to judge the voiced nature of the speech frame. By calculating the sample entropy of English tone pronunciation, it is found that the signal complexity of the voiced/unvoiced is obviously different. And the voiceless judgment is carried out according to the complexity. Meanwhile, the fast algorithm does the calculations in binary, the neighborhood matrix of the high-dimensional signal vector is deduced from the neighborhood matrix of the lower dimensional signal vector, therefore the segmentation is more efficient. Simulation experiments are conducted. The experimental results show that the method of sample entropy can get the better results of speech segmentation compared with other traditional methods, the computation time is reduced by 80 times and the operation efficiency is improved by using the fast sample entropy algorithm. This research has a good application prospect in speech signal.

关 键 词：样本熵 快速样本熵 清浊音判决 语音分割 

分 类 号：TN912.3[电子电信—通信与信息系统]