Page 5 - 《应用声学》2022年第6期
P. 5

第 41 卷 第 6 期                                                                       Vol. 41, No. 6
             2022 年 11 月                         Journal of Applied Acoustics                 November, 2022

             ⋄ 研究报告 ⋄



                      应对说话人位置突变的鲁棒语声去混响                                                          ∗






                                     吴礼福      1,2†  孙帅恒     1   王 雷     1   孙芯年     1


                                         (1 南京信息工程大学电子与信息工程学院           南京  210044)
                                        (2 江苏省大气环境与装备技术协同创新中心           南京   210044)

                摘要：多通道线性预测算法是一种有效的语声去混响方法，但若房间内说话人位置发生变化，房间冲激响应也
                会随之改变，从而导致去混响效果变差。该文研究了基于递归最小二乘算法的自适应多通道线性预测语声去
                混响方法，提出了一种去混响过程中检测说话人位置突变的方法，它借助已有的多通道传声器信号，并未明显
                增加系统开销。实验表明该方法可以有效检测说话人位置的突变，并通过重置滤波器改善算法的收敛速度和
                稳定性。
                关键词：去混响；多通道线性预测；递归最小二乘
                中图法分类号: N912.35          文献标识码: A          文章编号: 1000-310X(2022)06-0851-09
                DOI: 10.11684/j.issn.1000-310X.2022.06.001






               Robust speech dereverberation considering sudden change of speaker position



                                 WU Lifu 1,2  SUN Shuaiheng 1  WANG Lei  1  SUN Xinnian 1

                    (1 School of Electronic & Information Engineering, Nanjing University of Information Science & Technology,
                                                    Nanjing 210044, China)
                       (2 Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,
                                                    Nanjing 210044, China)

                 Abstract: Multi-channel linear prediction (MCLP) algorithm is an effective speech de reverberation method,
                 but if the speaker position in the room changes, the room impulse response (RIR) will also change, resulting
                 in poor dereverberation effect. In this paper, an adaptive MCLP speech dereverberation method based on
                 recursive least square (RLS) algorithm is studied, and a method for detecting speaker position mutation in the
                 process of dereverberation is proposed. With the help of the existing multi-channel microphone signal, it does
                 not significantly increase the system overhead. Experiments show that this method can effectively detect the
                 mutation of speaker position, and improve the convergence speed and stability of the algorithm by resetting
                 the filter.
                 Keywords: Dereverberation; Multi-channel linear prediction; Recursive least square




             2021-10-08 收稿; 2022-02-15 定稿
             国家自然科学基金项目 (12074192)
             ∗
             作者简介: 吴礼福 (1981– ), 男, 安徽全椒人, 博士, 副教授, 研究方向: 声频信号处理。
              通信作者 E-mail: wulifu@nuist.edu.cn
             †
   1   2   3   4   5   6   7   8   9   10