基于相关增强的单信道信号分离  

Single channel signal separation based on correlation enhancement

作　　者：刘威[1] 胡爱群[1] 

机构地区：[1]东南大学信息科学与工程学院,南京210096

出　　处：《东南大学学报（自然科学版）》2010年第3期464-470,共7页Journal of Southeast University：Natural Science Edition

摘　　要：提出了一种在单通道自适应干扰抑制框架下分离语音信号和伪随机信号的方法.该方法首先采用时间延迟叠加方法增强语音信号的相关性和周期性,并利用时延估计器提取周期性特征,从而提高估计精度.然后,采用自适应算法估计语音信号,并通过从混合信号中抵消语音信号,估计出伪随机信号.仿真结果表明,该方法既能分离清音信号,也能分离浊音信号.分离单音频信号时,规范化最小均方算法的性能优于其他基于最小均方的算法;使用递归最小二乘算法可有效分离出相关增强后的清音信号;与无增强的信号相比,使用相关增强策略可提高浊音信号的估计精度.An approach to separate a speech signal and a pseudo-random（PN） signal is presented in the framework of a single channel adaptive interference suppression model.First,the speech signal correlation and periodicities are enhanced using time-delay summation methods.And the periodicities are extracted by a time-delay estimator to improve the estimation precision.Then,the speech signal is estimated using adaptive algorithms.The PN signal is extrapolated by eliminating the speech signal.The simulation results show that the proposed method can separate unvoiced and voiced signals.When separating a single tone signal,the normalized least mean square（LMS） algorithm outperforms other LMS-based algorithms.In addition,the correlation-enhanced unvoiced signal can be effectively separated by using the recursive least squares algorithm.Compared with the non-enhanced signal,the estimation accuracy of the voiced signal can be improved by the correlation enhancement scheme.

关 键 词：单信道信号分离 自适应滤波 相关增强 时间延迟叠加 最小均方 递归最小二乘 

分 类 号：TN911.7[电子电信—通信与信息系统]