基于概率密度并联距离的话音检测算法  

Voice Activity Detection Using Parallel Distance of Probability Density

作　　者：吴其前[1] 张雄伟[1] 邹霞[1] 

机构地区：[1]解放军理工大学通信工程学院电子信息工程系,南京210007

出　　处：《电子与信息学报》2008年第12期2886-2889,共4页Journal of Electronics & Information Technology

基　　金：江苏省科学基金(BK2006001);江苏省"333"高层次人才培养工程专项资助课题

摘　　要：该文提出了一种基于概率密度并联距离的话音激活检测算法。算法根据语音信号和噪声信号的Mel域子带能量概率密度的不同特性,引入并联距离定义构造判决函数,通过判断该函数的值来进行语音激活检测。实验结果表明,在不同信噪比情况下,该文算法性能优于G.729BVAD算法。In this paper, a Voice Activity Detection algorithm is developed, which uses a distance measure parameter - parallel connection distance of probability density to improve detection. The algorithm is based on the determination of the speech/noise divergence by means of parallel distance of the Mel-field sub-band log-energies probability density. The proposed VAD is tested on speech signals with various noises under different SNRs. Experimental results show that the proposed VAD outperforms the standard algorithm G.729 VAD.

关 键 词：话音激活检测 概率密度 并联距离 

分 类 号：TN912.3[电子电信—通信与信息系统]