登录
首页 » Others » 基于LMS 算法的多麦克风降噪

基于LMS 算法的多麦克风降噪

于 2020-11-28 发布
0 453
下载积分: 1 下载次数: 15

代码说明:

武汉理工大学 信息处理课设 基于LMS 算法的多麦克风降噪 给定主麦克风录制的受噪声污染的语音信号和参考麦克风录制的噪声,实现语音增强的目标,得到清晰的语音信号。2007控制科学与工程全国博士生学术论坛2007年8月其中日为语音信号与麦克风阵列所在平面的夹角,d为麦克风间距,c为声音传播速度,f为信号采样率。固定波束形成器通过延时求和单元产生参考语音信号y(n),y(n)与y(m)分别代表期望语音信号与噪声信号。y,(n)4x(m)=y(m)+y/(m(3)信号通过阻塞矩阵产生噪声参考信号用来估计波束形成输出信号中的噪声成分。选取B使其中任意行向量之和为零,即任意行向量线性无关。为了进一步降低噪声参考信号中的语音泄漏,参考文献“提出了用自适应阻塞矩阵替代固定阻塞矩阵的方法。ynly2nMM-[nJ]=BLun], u2n],umn自适应噪声抵消器ANC通过对输入噪声参考信号进行自适应滤波处理抵消了参考信号y,(m)中的噪声成分,得到增强的语音信号。em]=y[m-∑nnl3LMS自适应算法及改进31LMS自适应算法GSC架构中的自适应噪声抵消器ANC需要用增强的语音信号作为反馈对滤波器权值进行自适应更新。很多自适应算法基于LMS及其改进形式, Clark提出的块LMS算法使得滤波器的自适应逐块更新而非传统LMS滤波器逐点更新4, HOSHUYAMA、 Kellermann分别提出的基于范数约束自适应算法的权值更新,以及频域无约束实现。这些算法基本结构如图2所示y(n-1)(n-L+1)wo(ne(ny/(n)图2自适应横向滤波器结构图图2为图1中的M-1路L阶多通道自适应噪声对消器中某一路的展开形式,其抽头输入向量为[ym]yn-]yn-L+1],对应的抽头权向量为wmwn]w-]。LMS算法的梯度向量通过G2007控制科学与工程全国博士生学术论坛2007年8月计算抽头输入相关矩阵R和抽头输入与期望响应间互相关向量p得到VJ(n)=-2p+2Rv(m),将R和p的瞬态估计R(n)=y(m)y"(n),p(n)=y(n)y/(m)代入,得出梯度向量的瞬态估计:VJ(n)=-2y(n)y, (n+2y(n)y"(n)w(n)进而推出LMS算法权值更新公式为w(n+1)=w(n)+uy(n)Ly(n)-y"(n)w(n)32基于稳态噪声的自适应算法改进考查图2中具有L个抽头权值的LMS算法,抽头权值与抽头输入一一对应。在传统的逐点更新LMS算法中,每计算一个输出需要L次乘法,而更新一次抽头权值也需要L次乘法,故每次迭代需要2L次乘法。对于L个输出样值,所需要的乘法次数为2次。针对传统LMS算法复杂度高的缺点,Ca利用离散傅立叶变换在频域完成滤波器系数的自适应提出了快速块LMS箅法, Ann Spriet在此基础上通过改进LMS算法中的步长矩阵进一步降低了算法复杂度以上LMS算法改进均在图2的横向滤波器架构下进行,即抽头权值与抽头输入一一对应。考虑到稳态噪声的特点,本文提出了“一对多”的滤波器抽头权值更新算法,即L个输入样值共享一个滤波器权值。如此M路多麦克风语音增强系统中的ANC滤波器权值便由(M-1)×L维矩阵W[n=[w[η],n2[rl…wM-[r],其中H[n]=[won],w1[nw-r]退化为(M-1)×1维向量n]=[wryw2n],M-m]j。改进算法权值更新公式为w(n+D)=w(n)+uBu(nu"(n)[A-Bw(n)其中B为阻塞矩阵,A为固定波束形成器,为步长,U(n)为LxM维输入信号。与传统的“一对一”LMS滤波器相比,“一对多”结构在降低算法复杂度的同时,牺牲了前者具有的时间域严格对齐的特性。为降低这一缺点对系统降噪性能的影响,应在频域进行噪声对消,改进算法的多麦克风语音增强系统结构如图3所示。e(n)(n)B Yn图3改进的噪声消除算法结构图3中用虚线框表示可选滤波器权值w。由于实际应用中语音泄漏的存在,在参考语音信号中加入v能有效补偿由语音泄漏引起的语音崎变⑩。实际应用中由于阻塞矩阵输出不可避免的存在语音泄4642007控制科学与工程全国博士生学术论坛2007年8月漏,为了避免期望信号的消除,箅法中加入语音活动检测单元89,当前帧为噪声时更新滤波器系数,当前帧为语音信号时,滤波器系数不变33算法复杂度比较表1列出了本文算法与其他几种噪声消除算法之间算法复杂度的比较。我们采用实数乘法运算次数作为衡量算法复杂度的标准,每个N点傅立叶变换或其反变换需要Mlog2N次实数乘法运算。传统逐点LMS算法在时间域逐点更新滤波器权值。快速块LMS算法与多通道 Wiener算法通过FFT快速循环卷积特性实现LMS中的线性卷积运算,从而降低算法复杂度。本文算法在此基础上通过改进滤波器抽头权值更新算法进一步降低运算复杂度。由表1可见,当麦克风数目M4,L=32时,本文算法与多通道 Wiener滤波算法相比,R(3M+2)FT+8ML+2M63M+2)+4M2+6M_172(M+2)FFT+2ML6(M+2)+M40°文算法运算量降低了4倍左右。表1算法复杂度比较算法名称算法复杂度传统逐点LMS算法2ML快速块LMS算法(41(3M+2)FFT+16ML多通道 Wiener滤波算法53M+2)FFT+8M2+12M本文提出的算法(M+2)FF+2M…图4a)麦克风采集到的原始信号b)采用快速块LMS算法处理后的信号[4]c)采用多通道 Wiener滤波算法[10处理后的信号d采用本文算法处理后的信号4实验结果与分析实验采用线性排列的4个间距为4厘米的麦克风组成的语音采集系统,采样率为44KHZ,说话人位于阵列的正前方,噪声为稳态噪声,其与麦克风阵列法线所夹角度为50度。图4比较了麦克风采集到的信号、采用本文算法处理后的语音信号以及采用其他主流语音增强算法处理后的语音信号的时域波形。由4652007控制科学与工程全国博士生学术论坛2007年8月图4可见采用本文算法处理的语音信号背景噪声有明显降低。为进一步分析各种语音增强算法消噪能力,分别按照公式9计算各算法输出信号的信噪比,其中k代表帧序列号,N代表噪声,Y代表输出语音信号,L为帧长。∑(Y(k,2)2-|N(k,)SNRou(E)=10 log,o∑1MV6)图5釆用各箅法输出信号信噪比与输入信号信噪比之差来衡量噪声降低程度。由图5看出,在本文算法基础上在参考通道中加入可选滤波器权值能够进一步消除背景噪声,提高输出信噪比。苯文鲜法(使用权值w)木文好法未使用权值y块LMS算法Frame Number图5信噪比增强对比5结论本文在稳态噪声的前提下,提出了一种基于广义旁瓣消除器架构具有低算法复杂度的噪声消除算法,该算法通过改进LMS滤波器权值更新算法来达到降低算法复杂度的目的。实验结果证明,在稳态噪声环境下,该方法降噪性能优于传统LMS算法,同时有效降低了传统算法的算法复杂度。在现实生活中一些存在稳态噪声的场合,如发动机舱、厂房等该算法具有很强的实用价值。参考文献[U]LJ. Griffiths and C. W. Jim []. "An altemative approach to linearly constrained adaptive beamforming, IEEE Trans. AntennasProcess., voL. AP-30, no. I, pp 27-34, Jan. 1982.[2]0. Hoshuyama, A Sugiyama, and A Hirano [J]. "A robust adaptive beamformer for microphone arrays with a blocking matrixusing constrained adaptive filters, "IEEE Trans. Signal Process. vol 47, pp. 2677-2683, Oct. 1999[3]W. Herbordt and W Kellermann [J]. " Frequency-domain integration of acoustic echo cancellation and a generalized sidelobecanceller with improved robustness, "Eur. Trans. Telecommun., voL. 13, no 2, pp 123-132, Mar. -Apr. 2002.[4]Clark. G.A., S K Mitra, and S.R. Parker [J]. Block implementation of adaptive digital filters, "IEEE Trans. Circuits Syst,voL. CAS-28,PP584-592.1981.[5]Ann Spriet, Jan Wouters, Simon Doclo, Marc Moonen, "Frequency-Domain Criterion for the Speech Distortion WeightedMultichannel Wiener Filter for Robust Noise Reduction", Ap: //ftp. esat kuleuven. ac, be/pub/SISTA/doclo/reports/04-240 pdf[6JH. Buchner, J. Benesty, W. Kellermann J]. Generalized multichannel frequencydomain adaptive filtering: efficient realizationand application to hands free speech communication", Signal Processing 85(3), PP 549-570. 2005[7]W.Herbordt and W. Kellermann [A]. " Efficient Frequency-domain realization of robust generalized sidelobe cancellers", IEEE4662007控制科学与工程全国博士生学术论坛2007年8月Fourth workshop, multimedia signal Processing, PP. 377-382 2001[8]S. Van Gerven, F. Xie [J. "A Comparative Study of Speech Detection Methods", Proc. EUROSPEECH, VoL 3, Rhodos, Greecepp.1095-1098.1997[9]J Sohn, N.S.Kim, W Sung [] A Statistical Model-Based Voice Activity Detection", IEEE Signal Processing Lett. 6(1)1-31999[10]A Spriet, M. Moonen, J Wouters[]. Robustness Analysis of Multi-channel wiener Filtering and generalized sidelobeCancellation for Multi-microphone Noise Reduction in Hearing Aid Applications", IEEE Trans. Speech and Audio Processing, 13(4)PP.487-503.2005[IlJFerrara, E R r [] Fast implementation of LMS adaptive filters", IEEE Trans. Acoust. Speech Signal Process,voL.ASSP-28pp474-475.1980[12]S. Doclo and M. Moonen[J]. " Multi-microphone noise reduction using recursive GSVD-based optimal filtering with ANCpostprocessing stage, "IEEE Trans. Speech Audio Process., vol. 13, no. 1,Pp 53-69, Jan. 2005[13]Philipos C Loizou [J]. "Speech Enhancement Based on Perceptually Motivated Bayesian Estimators of the MagnitudeSpectrum" IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL 13, NO 5, Pp.857-869, 2005种新的基于稳态噪声的噪声消除算法旧WANFANG DATA文献链接作者:董鹏宇,朱子元,林涛作者单位:同济大学超大规模集成电路研究所,上海20009本文链接http://d.g.wanfangdata.comcn/confereNce6584700.aspx

下载说明:请别用迅雷下载,失败请重下,重下不扣分!

发表评论


0 个回复

  • 心电信号HRV分析
    UI化的心电HRV分析工具,基于MATLAB,可直接使用。包括时域分析、频域分析、时频分析和非线性分析。
    2020-11-28下载
    积分:1
  • SIM868芯片详细资料
    希姆通公司sim868芯片的资料,下载后可以自己解压看.基本的详细资料
    2020-11-06下载
    积分:1
  • MATLAB 图像处理宝典(原版)
    MATLAB 图像处理宝典(原版),经典入门教材,大量理论,算法,demo,适合有一定英语基础的图像处理入门级打基础~
    2020-12-10下载
    积分:1
  • 模拟光耦器件HCNR201及其在高精度电压检测中的应用
    介绍了惠普公司的高精度线性模拟器件HCNR200D的基本原理给出了利用该器件对模拟电压进行检测与隔离的硬件电路.
    2021-05-06下载
    积分:1
  • 矩形件优化排样算法的研究与实现
    矩形件优化排样算法的研究与实现,矩形件优化排样算法的研究与实现
    2020-12-05下载
    积分:1
  • 不平衡数据集的几种处理序SMOTE等
    关于在不平衡数据集中需要用到的一些欠采样过采样的方法,SMOTE等算法的实现及其示例。
    2020-12-11下载
    积分:1
  • ARM Linux内核源码剖析.pdf
    ARM Linux内核源码剖析.pdfARM Linux内核源码剖析.pdfARM Linux内核源码剖析.pdfARM Linux内核源码剖析.pdf 完整书签
    2020-12-10下载
    积分:1
  • 智能车MT9V034摄像头调试上位机
    里面包含了十几种动态二值化阈值的算法,能够采集和保存智能车图像信息,也能在上面编写程序实现图像处理算法。
    2020-07-04下载
    积分:1
  • Vehicle Dynamics Theory and Application
    不错的汽车动力学教材,是参与汽车底盘电子开发的动力学基础。Reza n. jazarVehicle DynamicsTheory and ApplicationsSpringerReza n. jazarDept of Mechanical EngineeringManhattan collegeRiverdale. NY 10471ISBN:978-0-387-74243-4e-ISBN:978-0-387-74244-1Library of Congress Control Number: 200794219c 2008 Springer Science+ Business Media, LLCAll rights reserved. This work may not be translated or copied in whole or in part without thewritten permission of the publisher(Springer Science+Business Media, LLC, 233 SpringStreet, New York, NY 10013, USA), except for brief excerpts in connection with reviews orscholarly analysis. Use in connection with any form of information storage and retrievalelectronic adaptation, computer software, or by similar or dissimilar methodology now knownor hereafter developed is forbidden. The use in this publication of trade names, trademarksservice marks and similar terms, even if they are not identified as such, is not to be taken as anexpression of opinion as to whether or not they are subject to proprietary rightsPrinted on acid-free paper987654321springer. comKavoshmy daughter, Vazan,and my wife, MojganHappiness is when you win a race against yourselfPrefaceThis text is for engineering students. It introduces the fundamental knowledge used in vehicle dymamics. This knowledge can be utilized to developcomputer programs for analyzing the ride, handling, and optimization ofroad vehiclesVehicle dynamics has been in the engineering curriculum for more thana hundred years. Books on the subject are available, but most of themare written for specialists and are not suitable for a classroom applicationA new student, engineer, or researcher would not know where and howto start learning vehicle dynamics. So, there is a need for a textbook forbeginners. This textbook presents the fundamentals with a perspective onfuture trendsThe study of classical vehicle dynamics has its roots in the work ofgreat scientists of the past four centuries and creative engineers in thepast century who established the methodology of dynamic systems. Thedevelopment of vehicle dynamics has moved toward modeling, analysisand optimization of multi-body dynamics supported by some compliantmembers. Therefore, merging dynamics with optimization theory was anexpected development. The fast-growing capability of accurate positioninsensing, and calculations, along with intelligent computer programming arethe other important developments in vehicle dynamics. So, a textbook helpthe reader to make a computer model of vehicles, which this book doesLevel of the bookThis book has evolved from nearly a decade of research in nonlineardynamic systems and teaching courses in vehicle dynamics. It is addressedprimarily to the last year of undergraduate study and the first year graduatestudent in engineering. Hence, it is an intermediate textbook. It providesboth fundamental and advanced topics. The whole book can be coveredin two successive courses, however, it is possible to jump over some sections and cover the book in one course. Students are required to know thefundamentals of kinematics and dynamics, as well as a basic knowledge ofnumerical methodsThe contents of the book have been kept at a fairly theoretical-practicallevel. Many concepts are deeply explained and their application empha-sized, and most of the related theories and formal proofs have been explained. The book places a strong emphasis on the physical meaning andapplications of the concepts. Topics that have been selected are of highinterest in the field. An attempt has been made to expose students to aPrefacebroad range of topics and approachese There are four special chapters that are indirectly related to vehicle dy-amics: Applied Kinematics, Applied Mechanisms, Applied dynamics, andApplied vibrations. These chapters provide the related background to understand vehicle dynamics and its subsystemsOrganization of the bookThe text is organized so it can be used for teaching or for self-studyChapter 1"Fundamentals, "contains general preliminaries about tire andrim with a brief review of road vehicle classificationsPart I"One Dimensional Vehicle Dynamics, " presents forward vehicledynamics, tire dynamics, and driveline dynamics. Forward dynamics refersto weight transfer, accelerating braking, engine performance, and gear ratiodesignPart II"Vehicle Kinematics, presents a detailed discussion of vehiclemechanical subsystems such as steering and suspensionsPart IIT"Vehicle Dynamics, employs Newton and Lagrange methodsto develop the maneuvering dynamics of vehiclesPart Iv "Vehicle Vibrations, presents a detailed discussion of vehi-cle vibrations. An attempt is made to review the basic approaches anddemonstrate how a vehicle can be modeled as a vibrating multiple degreeof-freedom system. The concepts of the Newton-Euler dynamics and La-grangian method are used equally for derivation of equations of motionThe RMS optimization technique for suspension design of vehicles is intro-duced and applied to vehicle suspensions. The outcome of the optimizationtechnique is the optimal stiffness and damping for a car or suspended equipmentMethod of presentationThis book uses a fact-reason-application"structure. The "fact"is themain subject we introduce in each section. Then the reason is given as a" proof. The application of the fact is examined in some examples. Theexamplesare a very important part of the book because they show howto implement the facts. They also cover some other facts that are neededto expand the subjectPrerequisitesSince the book is written for senior undergraduate and first-year graduatelevel students of engineering, the assumption is that users are familiar withmatrix algebra as well as basic dynamics. Prerequisites are the fundamentals of kinematics, dynamics, vector analysis, and matrix theory. Thesebasics are usually taught in the first three undergraduate yearsPrefaceUnit SystemThe system of units adopted in this book is, unless otherwise stated, theinternational system of units(SI). The units of degree(deg)or radian(rad)are utilized for variables representing angular quantitiesSymbolse Lowercase bold letters indicate a vector. Vectors may be expressed inan n dimensional Euclidian space. ExamplerCUppercase bold letters indicate a dynamic vector or a dynamic matrix, such as force and moment. ExampleFo Lowercase letters with a hat indicate a unit vector. Unit vectors arenot bolded. ExampleLowercase letters with a tilde indicate a 3 x 3 skew symmetric matrixassociated to a vector. Examplea3211An arrow above two uppercase letters indicates the start and endpoints of a position vector. ExampleON = a position vector from point o to point Ne The length of a vector is indicated by a non-bold lowercase letterExampleCapital letter B is utilized to denote a body coordinate frame. ExampleB(ocgB(Oxyz)B1(o1x19121)ⅹ11PrefaceCapital letter G is utilized to denote a global, inertial, or fixed coordinate frame. ExampleG(XYZG(OXYZRight subscript on a transformation matrix indicates the departureframes. ExampleRB= transformation matrix from frame B(oxyz)Left superscript on a transformation matrix indicates the destinationframe. ExampleRBtransformation matrix from frame B(o cgz)to frame G(OxYZ)Capital letter R indicates rotation or a transformation matrix, if itshows the beginning and destination coordinate frames. Example0BSIn a0Whenever there is no sub or superscript, the matrices are shown in abracket. ExampleCOS asin a osIn aCOs O0e Left superscript on a vector denotes the frame in which the vectoris expressed. That superscript indicates the frame that the vectorbelongs to; so the vector is expressed using the unit vectors of thatEr= position vector expressed in frame G(OXYZ)Right subscript on a vector denotes the tip point that the vector isreferred to. ExamplePsition vector ofexpressed in coordinate frame G(OXYZ)Right subscript on an angular velocity vector indicates the frame thatthe angular vector is referred to. ExampleB= angularof the body coordinate frame B(oxyz)
    2020-12-12下载
    积分:1
  • 802.11 WLAN物理层仿真源代码 matlab
    802.11 WLAN物理层仿真源代码,使用matlab写的,比较新,比较全面,从发送端到信道到接收端都有
    2020-11-29下载
    积分:1
  • 696518资源总数
  • 104652会员总数
  • 48今日下载