面向高混响环境的欠定卷积盲源分离算法

doi:10.11959/j.issn.1000-436x.2023027

通信学报 ›› 2023, Vol. 44 ›› Issue (2): 82-93.doi: 10.11959/j.issn.1000-436x.2023027

面向高混响环境的欠定卷积盲源分离算法

解元¹, 邹涛¹, 孙为军², 谢胜利³

¹ 广州大学机械与电气工程学院，广东广州 510006
² 广东工业大学智能检测与制造物联教育部重点实验室，广东广州 510006
³ 广东工业大学物联网智能信息处理与系统集成教育部重点实验室，广东广州 510006

修回日期:2022-12-26 出版日期:2023-02-25 发布日期:2023-02-01
作者简介:解元（1989- ），男，安徽利辛人，博士，广州大学讲师，主要研究方向为盲信号分离、信号处理和机器学习等
邹涛（1975- ），男，辽宁沈阳人，博士，广州大学教授，主要研究方向为工业过程建模与仿真、模型预测控制、先进过程控制和实时优化技术研究与应用
孙为军（1975- ），男，安徽马鞍山人，博士，广东工业大学副教授，主要研究方向为模式识别和机器学习等
谢胜利（1956- ），男，湖北荆州人，博士，广东工业大学教授，主要研究方向为无线网络、自动控制、盲信号处理等
基金资助:
国家自然科学基金资助项目(62003095);国家自然科学基金资助项目(52171331);广东省重点领域研发计划基金资助项目(2019B01054002)

Algorithm of underdetermined convolutive blind source separation for high reverberation environment

Yuan XIE¹, Tao1 ZOU¹, Weijun SUN², Shengli XIE³

¹ School of Mechanical and Electrical Engineering, Guangzhou University, Guangzhou 510006, China
² Key Laboratory of Intelligent Detection and the Internet of Things in Manufacturing, Ministry of Education, Guangdong University of Technology, Guangzhou 510006, China
³ Key Laboratory of Intelligent Information Processing and System Integration of IoT, Ministry of Education, Guangdong University of Technology, Guangzhou 510006, China

Revised:2022-12-26 Online:2023-02-25 Published:2023-02-01
Supported by:
The National Natural Science Foundation of China(62003095);The National Natural Science Foundation of China(52171331);Research and Development Plan of Key Fields in Guangdong Province(2019B01054002)

摘要/Abstract

摘要：

为了解决高混响环境下欠定卷积混叠信号的分离问题，提出一种新的欠定卷积盲源分离算法。针对高混响环境的影响，设计全局脉冲响应网络削弱混响回声，提高信号质量。基于全局脉冲响应网络建立新的时频域混叠信号数学模型，采用全局脉冲响应矩阵缩短了传统脉冲响应的长度，降低了高混响带来的模型变换近似误差。基于非负矩阵分解理论设计模型参数的实时更新学习规则，将源信号分离问题转换为模型参数优化问题，实现混叠信号的盲源分离。实验结果表明，所提算法可以有效地实现中英文语音、音乐混叠信号的盲源分离，与现有比较流行的盲源分离算法的对比验证了所提算法的优越性。

关键词: 盲源分离, 欠定卷积混叠, 高混响环境, 全局脉冲响应网络, 非负矩阵分解

Abstract:

To separate the underdetermined convolutive mixture signals in the high reverberation environment, a novel algorithm of underdetermined convolutive blind source separation was proposed.Aiming at the influence of high reverberation environment, a global impulse response network was designed to weaken reverberation echo, improving signal quality.A new mathematical model of time-frequency mixing signals was established based on the global impulse response network.The global impulse response matrix which shortened the length of the traditional impulse response, reduced the approximation error of model transformation caused by high reverberation.The real-time update learning rules of model parameters were designed based on the theory of nonnegative matrix factorization, and the source signal separation problem was converted into the model parameter optimization problem, achieving blind source separation of mixing signals.Experimental results show that the proposed algorithm can effectively realize the blind source separation of Chinese and English speech and music signals, and the comparision with existing popular algorithms verified the superiority of the proposed algorithm.

Key words: blind source separation, underdetermined convolutive mixture, high reverberation environment, global impulse response network, nonnegative matrix factorization

中图分类号:

TN911.23

解元, 邹涛, 孙为军, 谢胜利. 面向高混响环境的欠定卷积盲源分离算法[J]. 通信学报, 2023, 44(2): 82-93.

Yuan XIE, Tao1 ZOU, Weijun SUN, Shengli XIE. Algorithm of underdetermined convolutive blind source separation for high reverberation environment[J]. Journal on Communications, 2023, 44(2): 82-93.

图/表 12

图1

图2

表1

图3

图4

表2

图5

图6

表3

图7

图8

图9

参考文献［27］

［1］	张贤达, 保铮．盲信号分离［J］．电子学报, 2001,29(S1): 1766-1771．
	ZHANG X D , BAO Z ． Blind source separation［J］． Acta Electronica Sinica, 2001,29(S1): 1766-1771．
［2］	马昊, 郑翔, 张煜 ,等．盲源分离技术研究:有监督与无监督［J］．计算机时代, 2022(11): 59-64．
	MA H , ZHENG X , ZHANG Y ,et al． Research on blind source separation techniques:supervised vs.unsupervised［J］． Computer Era, 2022(11): 59-64．
［3］	刘秋红, 许漫坤, 李天昀 ,等．基于互补对称滤波器的 APCMA 信号的盲分离算法［J］．电子学报, 2020,48(12): 2394-2401．
	LIU Q H , XU M K , LI T Y ,et al． Blind separation of APCMA signal based on complementary symmetric filters［J］． Acta Electronica Sinica, 2020,48(12): 2394-2401．
［4］	XIE Y , XIE K , YANG Q Y ,et al． Reverberant blind separation of heart and lung sounds using nonnegative matrix factorization and auxiliary function technique［J］． Biomedical Signal Processing and Control, 2021,69 102899.
［5］	MIETTINEN J , NITZAN E , VOROBYOV S A ,et al． Graph signal processing meets blind source separation［J］． IEEE Transactions on Signal Processing, 2021,69: 2585-2599．
［6］	马宝泽, 张天骐, 安泽亮 ,等．基于张量分解的卷积盲源分离方法［J］．通信学报, 2021,42(8): 52-60．
	MA B Z , ZHANG T Q , AN Z L ,et al． Convolutive blind source separation method based on tensor decomposition［J］． Journal on Communications, 2021,42(8): 52-60．
［7］	王泽林, 陈锴, 卢晶．车载场景结合盲源分离与多说话人状态判决的语音抽取［J］．声学学报, 2020,45(5): 696-706．
	WANG Z L , CHEN K , LU J ． Speech extraction based on blind source separation and multi-talker status tracking in automobile environment［J］． Acta Acustica, 2020,45(5): 696-706．
［8］	MERTINS A , MEI T M , KALLINGER M ． Room impulse response shortening/reshaping with infinity- and p-norm optimization［J］． IEEE Transactions on Audio,Speech,and Language Processing, 2010,18(2): 249-259．
［9］	KRISHNAN L , BETLEHEM T , TEAL P D ． Fast algorithms for acoustic impulse response shaping［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2019,27(2): 392-403．
［10］	JUNGMANN J O , MAZUR R , KALLINGER M ,et al． Combined acoustic MIMO channel crosstalk cancellation and room impulse response reshaping［J］． IEEE Transactions on Audio,Speech,and Language Processing, 2012,20(6): 1829-1842．
［11］	MERTINS A , MAASS M , KATZBERG F ． Room impulse response reshaping and crosstalk cancellation using convex optimization［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2021,29: 489-502．
［12］	邹亮, 张鹏, 陈勋．基于三阶统计量的欠定盲源分离方法［J］．电子与信息学报, 2022,44(11): 3960-3966．
	ZOU L , ZHANG P , CHEN X ． Underdetermined blind source separation based on third-order statistics［J］． Journal of Electronics ＆ Information Technology, 2022,44(11): 3960-3966．
［13］	田宝平, 应昊蓉, 杨文境 ,等．结合 ICA 和复数神经网络的双麦阵列盲源分离方法［J］．信号处理, 2021,37(11): 2185-2192．
	TIAN B P , YING H R , YANG W J ,et al． Blind source separation of binary array based on ICA and complex neural network［J］． Journal of Signal Processing, 2021,37(11): 2185-2192．
［14］	解元, 邹涛, 孙为军 ,等．面向卷积混叠环境下的盲源分离新方法［J］．自动化学报, 2022:doi．org/10．16383/j．aas．c211207．
	XIE Y , ZOU T , SUN W J ,et al． Novel blind source separation method for convolutive mixed environment［J］． Act Autom Sin, 2022:doi．org/10．16383/j．aas．c211207．
［15］	李帅, 刘宏清, 彭鹏 ,等．混响环境下基于卷积模型的欠定盲源分离［J］．信号处理, 2021,37(4): 624-632．
	LI S , LIU H Q , PENG P ,et al． Underdetermined blind source separation based on convolution model in reverberant environment［J］． Journal of Signal Processing, 2021,37(4): 624-632．
［16］	汤辉, 王殊．基于稳健联合分块对角化的卷积盲分离［J］．自动化学报, 2013,39(9): 1502-1510．
	TANG H , WANG S ． Robust joint block diagonalization based convolutive blind source separation［J］． Acta Automatica Sinica, 2013,39(9): 1502-1510．
［17］	ITO N , IKESHITA R , SAWADA H ,et al． A joint diagonalization based efficient approach to underdetermined blind audio source separation using the multichannel Wiener filter［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2021,29: 1950-1965．
［18］	DUONG N Q K , VINCENT E , GRIBONVAL R ． Under-determined reverberant audio source separation using a full-rank spatial covariance model［J］． IEEE Transactions on Audio,Speech,and Language Processing, 2010,18(7): 1830-1840．
［19］	KOUNADES-BASTIAN D , GIRIN L , ALAMEDA-PINEDA X , ,et al． A variational EM algorithm for the separation of time-varying convolutive audio mixtures［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2016,24(8): 1408-1423．
［20］	LEE D D , SEUNG H S ． Learning the parts of objects by non-negative matrix factorization［J］． Nature, 1999,401(6755): 788-791．
［21］	SAWADA H , KAMEOKA H , ARAKI S ,et al． Multichannel extensions of non-negative matrix factorization with complex-valued data［J］． IEEE Transactions on Audio,Speech,and Language Processing, 2013,21(5): 971-982．
［22］	SEKIGUCHI K , BANDO Y , NUGRAHA A A ,et al． Fast multichannel nonnegative matrix factorization with directivity-aware jointly-diagonalizable spatial covariance matrices for blind source separation［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2020,28: 2610-2625．
［23］	WANG J Y , GUAN S Z , LIU S P ,et al． Minimum-volume multichannel nonnegative matrix factorization for blind audio source separation［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2021,29: 3089-3103．
［24］	FENG F C , KOWALSKI M ． Underdetermined reverberant blind source separation:sparse approaches for multiplicative and convolutive narrowband approximation［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2019,27(2): 442-456．
［25］	WANG T H , YANG F R , YANG J ． Convolutive transfer function-based multichannel nonnegative matrix factorization for overdetermined blind source separation［J］． IEEE/ACM Transactions on Audio,Speech,and Language Processing, 2022,30: 802-815．
［26］	ALLEN J B , BERKLEY D A ． Image method for efficiently simulating small-room acoustics［J］． The Journal of the Acoustical Society of America, 1979,65(4): 943-950．
［27］	VINCENT E , GRIBONVAL R , FEVOTTE C ． Performance measurement in blind audio source separation［J］． IEEE Transactions on Audio,Speech,and Language Processing, 2006,14(4): 1462-1469．

数据集	源信号	频率/kHz	时长/s
英文语音1	dev1_female3_src_1	16	10
英文语音2	dev1_female3_src_2	16	10
英文语音3	dev1_female3_src_3	16	10

数据集	源信号	频率/kHz	时长/s
音乐信号1	dev2_another_dreamer-the_ones_we_love_snip_69_94_guitar	44.1	25
音乐信号2	dev2_another_dreamer-the_ones_we_love_snip_69_94_vocals	44.1	25
音乐信号3	dev2_another_dreamer-the_ones_we_love_snip_69_94_drums	44.1	25

数据集	源信号	频率/kHz	时长/s
中文语音1	IC0936W0131	16	5
中文语音2	IC0936W0134	16	5
中文语音3	IC0936W0137	16	5

[1]	李骜, 冯聪, 牛宇童, 徐士彪, 张英涛, 孙广路. 面向视角非对齐数据的多视角聚类方法[J]. 通信学报, 2022, 43(7): 143-152.
[2]	马宝泽, 张天骐, 安泽亮, 邓盼. 基于张量分解的卷积盲源分离方法[J]. 通信学报, 2021, 42(8): 52-60.
[3]	邱飞岳,陈博文,陈铁明,章国道. 稀疏诱导流形正则化凸非负矩阵分解算法[J]. 通信学报, 2020, 41(5): 84-95.
[4]	陈雷,甘士忠,张立毅,王光艳. 基于样条插值与人工蜂群优化的非线性盲源分离算法[J]. 通信学报, 2017, 38(7): 36-46.
[5]	张天骐,马宝泽,强幸子,全盛荣. 带自适应动量因子的变步长盲源分离方法[J]. 通信学报, 2017, 38(3): 16-24.
[6]	常振超,陈鸿昶,黄瑞阳,于洪涛,刘阳. 基于非负矩阵分解的半监督动态社团检测[J]. 通信学报, 2016, 37(2): 132-143.
[7]	于淼,杨武,王巍,申国伟. 面向微博的多实体稀疏关系数据联合聚类[J]. 通信学报, 2016, 37(1): 151-159.
[8]	栗华,贾智平,王洪君,刘琚. 基于动态位隙分组盲分离的UHF RFID防碰撞算法[J]. 通信学报, 2012, 33(4): 47-53.
[9]	魏祥麟,陈鸣,张国敏,黄建军. NMF-NAD：基于NMF的全网络流量异常检测方法[J]. 通信学报, 2012, 33(4): 54-61.
[10]	王荣杰,詹宜巨,周海峰. 基于零空间表示和最大似然的欠定盲源分离方法[J]. 通信学报, 2012, 33(3): 183-190.
[11]	杨小牛,姚俊良,李建东,李钊,张琰. M-PSK/QAM系统中基于相位预编码和盲源分离的共道干扰消除[J]. 通信学报, 2010, 31(8A): 154-160.
[12]	姚俊良,杨小牛,李建东,付卫红,李钊,张琰. 基于最大比合并的超定盲源分离[J]. 通信学报, 2010, 31(7): 9-17.
[13]	张晋东,秦贵和,陈涛,金健. 基于FastICA-TDS的DS-CDMA盲源信号分离系统[J]. 通信学报, 2008, 29(8): 126-130.
[14]	林秋华,党杰,殷福亮. 盲源分离图像加密的相关运算解密法[J]. 通信学报, 2008, 29(1): 109-114.
[15]	郭文强,邱天爽,查代奉. 基于Borel测度峰值判定的欠定混合盲语音信号分离[J]. 通信学报, 2007, 28(9): 22-26.

面向高混响环境的欠定卷积盲源分离算法

Algorithm of underdetermined convolutive blind source separation for high reverberation environment

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献［27］

相关文章 15

Metrics

推荐阅读 0

面向高混响环境的欠定卷积盲源分离算法

Algorithm of underdetermined convolutive blind source separation for high reverberation environment

在线阅读

PDF下载

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 ［27］

相关文章 15

Metrics

推荐阅读 0

参考文献［27］