Page 25 - 《应用声学》2022年第1期
P. 25

第 41 卷 第 1 期                                                                       Vol. 41, No. 1
             2022 年 1 月                          Journal of Applied Acoustics                   January, 2022

             ⋄ 研究报告 ⋄



              分布式无线声传感网加权预测误差语声去混响方法                                                                     ∗





                                  韩 哲     1,2  郑成诗      1,2  柯雨璇      1,2†   李晓东     1,2



                                     (1 中国科学院噪声与振动重点实验室 (声学研究所)            北京  100190)
                                                (2 中国科学院大学      北京  100049)

                摘要：室内混响会严重降低语声质量，因此在室内语声通信中对混响的抑制显得尤为重要。针对无线声传感
                网，该文提出一种基于加权预测误差的分布式自适应去混响算法。通过调整传统递归最小二乘算法，所提出的
                分布式加权预测误差算法仅需利用一路相同的参考信号和其他节点的本地输出而非全部信号，便可实现最优
                输出，从而大幅度降低节点间传输的通道数与各节点内滤波器维度。在无线声传感网的仿真实验中表明，该分
                布式算法可在提供更快的收敛速度和更高的鲁棒性的同时，实现与集中式算法相当的去混响性能。
                关键词：无线声传感网络；阵列信号处理；去混响
                中图法分类号: TN912.35           文献标识码: A          文章编号: 1000-310X(2022)01-0021-11
                DOI: 10.11684/j.issn.1000-310X.2022.01.003






                Distributed weighted prediction error method for speech dereverberation in
                                         wireless acoustic sensor networks



                            HAN Zhe  1,2  ZHENG Chengshi   1,2  KE Yuxuan 1,2   LI Xiaodong 1,2

                                 (1 Key Laboratory of Noise and Vibration Research, Institute of Acoustics,
                                         Chinese Academy of Sciences, Beijing 100190, China)
                                   (2 University of Chinese Academy of Sciences, Beijing 100049, China)

                 Abstract: Room reverberation may severely degrade speech quality, and thus it is critical to suppress the
                 reverberant components. In this paper, we propose a weighted prediction error (WPE)-based distributed
                 adaptive speech dereverberation method in the wireless acoustic sensor networks (WASNs). Through modifying
                 the traditional recursive least squares (RLS) algorithm, the proposed distributed WPE (DWPE) method can
                 achieve the optimal output by utilizing only a common reference signal and local outputs of other nodes
                 rather than all the received signals. Therefore, the proposed method can provide a significant reduction in
                 communication bandwidth and filter dimension at each node. Experiments in simulated WASNs validate that
                 the proposed method outperforms the centralized method in terms of convergence speed and robustness while
                 achieving almost the same performance.
                 Keywords: Wireless acoustic sensor networks; Array signal processing; Dereverberation


             2021-01-11 收稿; 2021-03-14 定稿
             青年英才计划项目 (QNYC201813)
             ∗
             作者简介: 韩哲 (1996– ), 男, 河南周口人, 硕士研究生, 研究方向: 阵列信号处理。
             † 通信作者 E-mail: keyuxuan@mail.ioa.ac.cn
   20   21   22   23   24   25   26   27   28   29   30