摘要
文章提出了一种基于改进QR分解递归最小二乘法(RLS)的多通道线性预测(MCLP)语音去混响算法。RLS算法是一种常用的自适应滤波算法,然而声学应用中RLS算法的估计误差方差将大大增加。为了解决这个问题,提出了一种基于状态正则化(SR)的递归最小二乘法。它采用估计的系数作为先验信息来最小化指数加权的观测误差,改进后的算法能获得比传统RLS算法更低的方差。通过QR分解来实现所提出的的算法,与直接实现相比,该算法舍入误差更低。在多通道线性预测算法中用SR-QRRLS算法代替RLS算法,仿真实验结果表明,改进后的算法对混响的抑制效果更好。
In this paper,a multi-channel linear prediction(MCLP)speech de-reverberation algorithm based on improved QR decomposition recursive least squares is presented.In acoustic applications,the estimation error variance of the RLS algorithm will increase considerably.To resolve this problem,a RLS algorithm based on state regularization(SR)is presented in this article.It uses the estimated coefficients as a priori information to minimise the exponentially weighted observation error,and the improved algorithm is able to obtain lower variance than the conventional RLS algorithm.The proposed algorithm is implemented by QR decomposition,which results in lower rounding errors compared to a direct implementation.Replacing the RLS algorithm with the SR-QRRLS algorithm in the multichannel linear prediction,and through some simulation experiments,the SR-QRRLS algorithm is obtained to have a better reverberation suppression effect.
作者
于春和
孟璐瑶
YU Chun-he;MENG Lu-yao(School of Electronic Information Engineering,Shenyang Aerospace University,Shenyang 110136,China)
出处
《电脑与信息技术》
2022年第5期35-38,共4页
Computer and Information Technology
关键词
语音去混响
RLS算法
MCLP
de-reverberation
RLS algorithm
multichannel linear prediction