音乐噪声环境下的双声道语音活动检测  

Dual-channel Voice Activity Detection in Music Noise Environments

作　　者：廖逢钗[1] 李鹏[2] 徐波[2] 

机构地区：[1]三明学院数学与计算机科学系,福建三明365004 [2]中国科学院自动化研究所数字内容技术研究中心,北京100190

出　　处：《信号处理》2009年第11期1820-1824,共5页Journal of Signal Processing

基　　金：国家973资助项目(No.2004CB318105);福建省自然科学基金项目(2009J01296)

摘　　要：提出双声道的语音活动检测算法(DCVAD,Dual-Channel Voice Activity Detection)。该算法首先利用骨架互相关的方法估计目标语音和背景音乐的时延,然后在对目标语音进行时延补偿的基础上,利用通道间语音和背景音乐自身以及二者相互之间的相关性,实现了语音活动检测的目标。实验表明,在背景音乐环境DCVAD算法语音活动检测的平均正确率比单通道基于谱熵的方法高约22%。A dual-channel voice activity detection algorithm （DCVAD） is proposed. The proposed algorithm first applies skeleton cross-correlation method to estimate the time delays corresponding to the target speech and background music, and then based on the time delay compensation on the target speech, it uses the relationship of the self-correlation and cross-correlation of/between the target speech and the background music to realize the objective of the voice activity detection. Experiments show that the average accuracy of DCVAD algorithm is about 22% higher than single channel VAD method based on spectrum entropy in the environments of background music.

关 键 词：语音活动检测 时延估计 互相关 听觉滤波器 

分 类 号：TN912.3[电子电信—通信与信息系统]