噪声估计和谱熵结合的语音激活检测算法  

A Speech Enhancement Algorithm Based on Noise Estimation and Spectral Subtraction

作　　者：郑秋菊[1] 李强[1] 王岑[1] 

机构地区：[1]重庆邮电大学多媒体通信技术重点实验室

出　　处：《现代电信科技》2013年第12期36-40,共5页Modern Science & Technology of Telecommunications

基　　金：国家高技术研究发展863计划子课题5:信关站与组网关键技术研究(NO.2012AA01A508);信号与信息处理重庆市市级重点实验室建设项目(CSTC;2009CA2003);2012年重庆邮电大学大学生科研训练计划(A2012-68)

摘　　要：本文提出一种语音激活检测的改进算法。首先在传统噪声估计的基础上,用Bark子带代替了DFT频域变换,目的在于降低计算复杂度;其次将估计的噪声谱进行白化滤波,并运用于子带谱熵算法中的谱熵计算中。把谱熵值作为VAD算法提取的特征参数,通过门限设定与比较,得出最初的VAD判决结果,增加拖尾延迟保护机制得出最终的VAD判决结果。This paper puts forward an improved al- gorithm of voice activity detection. At first, on the basis of the traditional noise estimation, 129 DFI＇ bins are replaced by 18 Bark scaled frequency bands to reduce the computational complexity. Secondly, bleaching filtering is done to the estimated noise spectrum, and sub band spectrum entropy algorithm is applied to spectrum entropy calculation. Then with the spectral entropy as an extracted feature parame- ters of VAD algorithm, through the threshold setting and comparison, the initial VAD decision is ac- quired. At last, the trailing delay protection mecha- nism is introduced to get the final VAD decision re- suits.

关 键 词：语音激活检测 噪声估计 谱熵 最小统计量 

分 类 号：TN912.32[电子电信—通信与信息系统]