改进的基于交叉熵的语音端点检测算法  

Improved voice activity detection algorithm based on cross-entropy

作　　者：赵春晖[1] 冯冲[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,哈尔滨150001

出　　处：《黑龙江大学自然科学学报》2014年第3期392-397,421,共6页Journal of Natural Science of Heilongjiang University

基　　金：国家自然科学基金资助项目(61077079);黑龙江省自然科学基金重点资助项目(ZD201216);哈尔滨市优秀学科带头人基金资助项目(RC2013XK009003)

摘　　要：针对传统的语音端点检测方法在强噪声环境下的可靠性会急剧下降问题,提出了一种改进的基于交叉熵的语音端点检测方法。该算法以子带交叉熵为语音/非语音的区分特性,采用基于语音存在概率的算法对背景噪声进行估计,将每帧语音与估计噪声的频谱划分成若干个子带,计算出每个子带能量与背景噪声之间的交叉熵,根据各帧交叉熵的值对输入的带噪语音进行端点检测。实验结果表明,该算法能够有效地区分语音和非语音,在强噪声环境下依然能够保持较高的检测率,与传统算法相比,具有鲁棒性。Faced with the fact that traditional voice activity detection method would drop dramatical- ly in high noise environments, an improved voice activity detection algorithm based on cross-entropy is proposed. The algorithm makes use of the sub-band cross-entropy as the speech/non-speech discrimina- tion feature. The analyses first estimate the background noise based on the speech presence probability and then divide the speech spectrum and noise spectrum into several sub-bands. The cross-entropy is giv- en between the speech signal and the background noise. And then the speech and non-speech signal are classified based on the cross-entropy. Tests show that the algorithm effectively distinguishes speech from non-speech, and even in high noise environments, the algorithm has a high detection rate. The algorithm is robust compared with traditional methods.

关 键 词：端点检测 噪声估计 交叉熵 频谱划分 

分 类 号：TN912.3[电子电信—通信与信息系统]