一种基于双通道自适应噪声对消的语音增强法  

A Method Based on Double-channel Adaptive Noise Cancelling for Speech Enhancement

作　　者：王振力[1] 张雄伟[1] 郑君杰[1] 

机构地区：[1]解放军理工大学通信工程学院,江苏南京210007

出　　处：《信号处理》2007年第5期786-790,共5页Journal of Signal Processing

基　　金：江苏省图像处理与图像通信实验室资助(资助号KJS03036);江苏省自然科学基金资助(资助号BK2006001)

摘　　要：在自适应噪声对消(ANC)中,本文根据主辅通道噪声间的相关性,提出了一种自适应滤波语音增强算法。在低信噪比(-10～0dB)白噪声条件下,文中研究了辅助通道采集的噪声有无"串音"两种情况下的语音增强效果。研究表明:在无串音和有串音两种条件下,本文算法增强语音的信噪比分别比NLMS算法提高约14dB和5dB,PESQ_MOS得分分别比NLMS算法提高约1.13和0.83,同时增强语音的听觉质量得到了极大地改善。By using the noisy correlations between the main and referenced channels in adaptive noise cancelling （ANC） ,a new adaptive filtering algorithm is proposed for speech enhancement. In an additive white Gaussian noise environment with low signal-to-noise ratio （SNR） ranging from 0 dB to -10dB ,speech enhancement is researched on the case of the reference input noise mixed with crosstalk whether or not. The result can be described as follows ： Assuming that the reference input containing erosstalk, the improved SNR and PESQ score by the new algorithm is more about 14dB and 1.13 than those of NLMS algorithm,respectively. Without the above assumption,the improved SNR and PESQ score become 5dB and 0. 83 , respectively. Moreover,the listening quality of the enhanced speech is remarkably improved.

关 键 词：自适应噪声对消 语音增强 NLMS 相关性 

分 类 号：TN912.35[电子电信—通信与信息系统]