基于卡尔曼滤波的STFT域回声抵消算法  

Kalman filter based acoustic echo cancellation in the STFT domain

作　　者：刘杨 杨飞然[1,2,3] 梁兆杰 杨军 LIU Yang;YANG Feiran;LIANG Zhaojie;YANG Jun(Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China;IOASONIC(Su zhou)Technologies Co.Ltd.,Suzhou 215163,Jiangsu,China;Shenzhen Power Supply Bureau Co.Ltd,Shenzhen 518000,Guangdong,China)

机构地区：[1]中国科学院声学研究所,北京100190 [2]中国科学院大学,北京100049 [3]中科新声(苏州)科技有限公司,江苏苏州215163 [4]深圳供电局有限公司,广东深圳518000

出　　处：《声学技术》2022年第5期757-762,共6页Technical Acoustics

基　　金：南方电网科技项目(090000KK52190062)。

摘　　要：提出了一种低复杂度的短时傅里叶变换域卡尔曼滤波算法来解决声学回声抵消问题。首先在短时傅里叶变换域建立了基于频域卷积传递函数的观测方程,并利用一阶马尔科夫模型对频域回声路径进行建模,给出了精确的卡尔曼滤波方程,并讨论了过程噪声和观测噪声的估计问题。为降低算法计算复杂度,提出了低复杂度卡尔曼滤波算法。另外,在更新滤波器时加入远端信号邻近频点的信息来进一步提高回声抵消性能。实验结果表明,所提算法对近端干扰不敏感,不需要额外的双端对讲检测算法,且比传统的频域自适应滤波算法具有更快的收敛速度。In this paper,a low-complexity Kalman filter algorithm in the short-time Fourier transform(STFT)domain is proposed for acoustic echo cancellation.The observation equation is obtained based on the convolutive transfer function in the STFT domain,and the echo path in the frequency domain is modeled by using the first-order Markov process.The exact Kalman filter is derived,and the estimation of the process noise and observation noise is discussed.A fast implementation method is presented to reduce the complexity of the exact Kalman filter.In addition,more adjacent frequency points of the far-end signal are introduced to further improve the echo cancellation performance.Experimental results show that the proposed algorithm is insensitive to the near-end interference and unnecessary to have an explicit double-talk detection algorithm,and it converges faster than the frequency domain adaptive filtering algorithm.

关 键 词：回声抵消 短时傅里叶变换 卡尔曼滤波 低复杂度 

分 类 号：TN912[电子电信—通信与信息系统]