基于ICA变换和MAP估计的语音增强算法  

Speech Enhancement Algorithm Based on ICA Transform and MAP Estimation

作　　者：高留洋[1] 李邵梅[1] 

机构地区：[1]信息工程大学国家数字交换系统工程技术研究中心,河南郑州450002

出　　处：《信息工程大学学报》2010年第5期553-556,共4页Journal of Information Engineering University

基　　金：国家863计划资助项目(2008AA011001)

摘　　要：提出了一种基于ICA变换和MAP估计的单通道语音增强算法,该方法依据ICA变换后语音信号服从拉普拉斯分布的特点,运用MAP估计纯音信号,然后进行逆ICA变换,获得时域纯音信号。仿真分析表明:在输入信噪比较低情况下,该算法增强效果好,同时增强后的语音信号失真小。This paper presents and analyses a new speech enhancement algorithm based on ICA transform and MAP Estimation.According to the characteristic that the ICA transformed speech signal obeys Laplacian distribution,the new algorithm accurately estimates the pure speech signal in ICA domain employing MAP Estimation,then obtains the pure speech signal in time domain by inverse ICA transform.The emulational analysis indicates the algorithm in this paper has good peformance as well as little distortion when the SNR of input noisy signal is low.

关 键 词：语音增强 ICA MAP 拉普拉斯分布 

分 类 号：TN912.3[电子电信—通信与信息系统]