基于语音端点检测和子空间方法的语音增强算法  

Speech enhancement algorithm based on VAD and subspace method

作　　者：张慧[1] 马建芬[1] 

机构地区：[1]太原理工大学计算机与软件学院,太原030024

出　　处：《计算机应用》2009年第B06期340-341,共2页journal of Computer Applications

摘　　要：利用子空间方法来实现语音的增强,在语音失真和残留噪声之间进行折中处理:既最小化语音失真,同时又使残留噪声保持在一个预先设定的值。传统的子空间法在平稳噪声环境下是有效的,但在非平稳环境下效果却不是很明显,因此利用语音端点检测(VAD)对噪声的协方差进行及时地更新。实验表明,采用基于VAD的子空间方法实现语音增强可以达到很好的效果。This paper proposed the subspace-based method to realize the speech enhancement. The authors made a compromise between the residua noise and the speech distortion to minimize the speech distortion while keeping the residual noise at a preset threshold. However, he traditional subspaee method is valid in the stationary noise environment, but is invalid in the non- stationary noise environment, so Voice Activity Detection （VAD） method was used to update the noise eovariance matrix . The experimental results prove that the subspace method based on VAD can efficiently enhance the speech signals.

关 键 词：语音增强 信号子空间 噪声子空间 奇异值分解/特征值分解 语音端点检测 

分 类 号：TN912[电子电信—通信与信息系统]