一种基于χ~2分布的子带噪声估计方法  

Subband Noise Estimation Based on χ~2 Distribution

作　　者：邹霞[1] 赵继勇[1] 张雄伟[1] 

机构地区：[1]解放军理工大学通信工程学院,南京210007

出　　处：《数据采集与处理》2008年第1期80-83,共4页Journal of Data Acquisition and Processing

基　　金：江苏省自然科学基金(BK2006001)资助项目

摘　　要：提出了一种基于χ2分布的子带噪声估计方法。带噪语音信号在临界带进行分解,并且假设子带信号服从χ2分布,然后在各个子带,采用基于χ2分布的改进最小统计量控制递归平均方法进行噪声估计。与传统的改进最小统计量控制递归平均噪声估计相比,该子带噪声估计方法可以利用人耳感知特性,并大大减少计算量。实验结果表明,提出的方法具有较好的噪声跟踪能力和较小的计算需求。采用该噪声估计的语音增强系统具有更强的噪声抑制性能和较好的增强语音信号质量。A subband noise estimation approach based on X^2 distribution is presented. The noisy speech signal is decomposed on the critical band and the subband signals are supposed as obey X^2 distributions. The subband noise estimation is updated by an improved minima controlled recursive averaging （IMCRA） approach and it is optimal for X^2 distribution of subband signals. Compared with the traditional minima controlled recursive averaging （MCRA） approach, the proposed subband noise estimation method can exploit the perceptual properties of human ear and lead to significant computational saving. Experimental results show that the proposed approach has noise track ability and lower computational requirements. The speech enhancement system using the proposed noise estimation yields less residual noise and better speech quality.

关 键 词：语音增强 噪声估计 子带 改进最小统计量控制递归平均 

分 类 号：TN912[电子电信—通信与信息系统]