基于盲源分离的回声抵消算法对比研究  

A comparative study of echo cancellation algorithms based on blind source separation

作　　者：张哲 杨飞然[1,2] 杨军 ZHANG Zhe;YANG FeiRan;YANG Jun(Key Laboratory of Noise and Vibration Research,Institute of Acoustics,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院声学研究所,噪声与振动重点实验室,北京100190 [2]中国科学院大学,北京100049

出　　处：《中国科学：物理学、力学、天文学》2022年第4期46-57,共12页Scientia Sinica Physica,Mechanica & Astronomica

基　　金：国家自然科学基金(编号:62171438);中国科学院青年促进会(编号:2018027)资助项目。

摘　　要：通信系统中存在的声学回声和线路回声严重影响语音通信质量,回声抵消是消除回声干扰、提高语音质量的有效方法.近年来基于盲源分离的回声抵消算法展现出优势,在没有双讲检测情况下仍可取得较好的回声消除效果.本文系统比较了现有的基于盲源分离的回声抵消算法的性能.选取了NG-ICA,Aux-ICA和MICA三种时域算法进行线路回声抵消仿真,wRLS,batch-ICA,online-IVA和我们所提基于最大似然估计的方法等4种频域算法进行声学回声抵消仿真.在回声路径改变以及双端对讲情况下分析对比算法的收敛速度、跟踪速度以及应对近端语音干扰的稳定性.仿真结果表明,该类算法大都存在跟踪性能与近端干扰敏感性的冲突.在设定最优参数兼顾算法复杂度情况下,时域MICA算法在提高收敛速度的同时能够保证较高回声衰减量,频域算法中wRLS算法和基于最大似然估计的方法取得了较高的回声衰减量.The presence of acoustic and line echoes in the communication system seriously impacts speech communication quality,and echo cancellation is an effective method to eliminate echo interference and improve speech quality.Echo cancellation algorithms based on blind source separation have shown advantages in recent years,and they can achieve good performance even without double-talk detection.In this study,we systematically compare the performance of existing echo cancellation algorithms based on blind source separation.Three time-domain algorithms,i.e.,naturalgradient-based independent component analysis(NG-ICA),auxiliary-function-based independent component analysis(Aux-ICA),and multidimensional independent component analysis(MICA),were selected for line echo cancellation experiments;weighted recursive least square(wRLS),batch independent component analysis(batch-ICA),online independent vector analysis(online-IVA),speex,and our proposed method were used for the acoustic echo cancellation experiments.The convergence rate,tracking capability,and stability of the algorithms are analyzed for double-talk and echo-path change scenarios.Experimental results show that most of these algorithms have a tradeoff between the tracking capability and sensitivity to the near-end interference.For optimal parameters,the MICA algorithm can improve the convergence speed and ensure high echo cancellation,and the wRLS algorithm and maximum likelihood estimationbased method in the frequency domain achieved higher ERLE than other frequency-domain algorithms.

关 键 词：回声抵消 盲源分离 双端对讲 自适应滤波 

分 类 号：TN912.3[电子电信—通信与信息系统]