混响环境下稳健麦克风阵列波束形成语音增强算法研究

发布时间：2018-03-26 19:07

本文选题：混响　切入点：麦克风阵列　出处：《南京信息工程大学》2017年硕士论文

【摘要】：麦克风阵列信号处理在助听设备、车载免提通讯系统、远程视频会议环境以及机器人听觉等语音信号处理中有着普遍的应用。但目前存在许多问题在麦克风阵列信号处理方面:首先,众所周知麦克风阵列接收到的语音是宽带信号,如果用传统的窄带波束形成方法来处理,会造成波束形成阻带发生畸变,无法抑制干扰信号;其次,通常存在麦克风通道的增益差异、相位偏移和位置抖动等不确定性造成的失配误差,而这些误差也会造成麦克风阵列信号处理的性能下降;最后,在相对封闭的几何空间内麦克风阵列与语音信号播放的位置存在一定距离时,麦克风阵列接收的信号不单仅是语音的直达信号,同时还有语音源在墙壁上多次反射的信号,从而导致语音的清晰度下降。为此,本文分析研究了混响环境下稳健麦克风阵列波束形成语音增强算法,并通过理论分析及计算机仿真和实验实测验证所提算法的有效性。主要工作如下:1.针对宽带信号波束形成器存在畸变的问题,给出了基于离散空间响应偏差函数的自适应加权宽带频率不变波束形成算法。该算法基于线性约束最小方差准则,首先通过离散空间响应偏差函数的二项式计算界定阵列空间响应偏差函数的均衡矩阵;其次将该均衡矩阵以系数加权的形式写入到该准则的波束形成算法目标优化函数中进行权值计算;最后再将系数加权频率不变波束形成算法中的加权系数定义为场点距离和信号频率的函数,该加权函数具有动态特性并采用自适应原理进行更新。通过仿真实验表明,该算法波束频率不变性能较好,阻带水平整体较低,在设置的干扰方向上形成了较深的零陷。2.针对现实环境中的麦克风通道特征误差导致宽带波束形成器性能下降的问题,给出了基于线性约束最小方差对角加载的稳健频率不变波束形成算法。该算法首先在线性约束最小方差准则的目标函数基础上以系数加权的形式结合离散空间响应偏差二项式矩阵,实现频率不变波束形成器;然后采用白噪声增益作为约束条件写入线性约束最小方差目标函数优化的准则中,分析白噪声增益约束值大小,以提高麦克风特征误差的稳健性,主要解决问题是麦克风的增益偏差、相位偏移和位置抖动等因素的不确定性造成低频信号全通滤波,从而导致波束形成器性能下降;最后通过拉格朗日乘子法计算固定权值和凸优化工具箱迭代求解得到最优权矢量。通过仿真实验表明,该算法波束稳健性能较好,稳健后的波束形成器的低频性能得到改善,阵列增益高、方向性较好。3.针对封闭环境中往往受到混响效应影响导致语音信号清晰度下降的情况,给出了基于混响环境下线性约束最小方差分频的改进维纳滤波后置波束形成算法。该算法基于延迟加权求和波束形成的维纳后置滤波结构上进行改进采用线性约束最小方差波束形成准则.改进的算法首先假设每个频段上混响时间不同的特性,在麦克风阵列接收信号的进行分频处理,将波束形成算法应用到高低频域的子带中,提高了混响抑制的精度;其次麦克风阵列接收到混响信号的直达波和反射波之间是不相关的,利用麦克风阵列接收信号的空间信息解决维纳滤波器的估计问题。通过仿真测验结果表明,该算法对混响抑制具有明显的改善,且提高了语音增强系统的评价指标得分。4.针对麦克风通道物理特性失配误差和封闭空间中引起的混响对语音清晰度的影响,研究了音频信号采集实验和数据分析。首先,在消声室中搭建实验平台,固定声源采集麦克风阵列数据,对采集的数据用本文提出的稳健算法进行验证;其次,在混响较强车库环境中搭建实验平台,固定声源采集混响阵列数据,对采集的数据进行本文提出的混响抑制算法验证。通过消声室和车库实际环境实测的算法输出信号的语谱图实验结果可以看出,所提出的算法与仿真结果预期一致。
[Abstract]:Microphone array signal processing in hearing aid equipment, hands-free communication system, has the widespread application of remote video conferencing environment and robot auditory processing of speech signal. But there are many problems in the microphone array signal processing: first of all, as everyone knows the received speech Mike wind array is a wideband signal, if using the traditional narrow-band beamforming the method to deal with, will cause the beamforming stopband distortion, can suppress the interference signal; secondly, there is the microphone gain difference channel, caused by phase shift and position jitter uncertainty such as mismatch errors, these errors will degrade the performance of microphone array signal processing; finally, playing with the microphone array speech signal in the geometric space is relatively closed position within a certain distance from the microphone array, signal receiving not only voice The direct signal, signal and speech source reflected many times on the wall, which leads to the speech articulation. Therefore, this paper studies the reverberation robust beamforming microphone array speech enhancement algorithm is effective, and through theoretical analysis and computer simulation and experimental verification of the proposed algorithm. The main work is as follows: 1. for wideband signal beamforming distortion problem, given the discrete spatial response adaptive weighted wideband frequency deviation function invariant beamforming algorithm. The algorithm based on linear constrained minimum variance criterion based on the first through the discrete space response function calculation deviation binomial response equilibrium matrix definition array spatial deviation function; secondly the equilibrium matrix the weighted coefficient of form written to the criteria of the beam forming algorithm to optimize the weights of target calculation function; Finally, the weighted coefficient of frequency invariant beamforming algorithm in the weighted coefficient is defined as a function of field distance and signal frequency, the weighting function is dynamic and updated by adaptive principle. The simulation results show that the algorithm performance is frequency invariant, the stopband level is low, in the direction of interference on the setting the formation of deep nulls for.2. microphone channels characteristic error in real environment leads to broadband beamformer performance degradation problem, presents a robust frequency linear constrained minimum variance diagonal loading beamforming algorithm based on invariant. Firstly, the objective function based on linear constrained minimum variance on the weighted coefficient of combination the discrete space response bias binomial matrix, to achieve frequency invariant beamformer; then the white noise gain as the constraint condition of writing The linear constrained minimum variance objective function optimization criterion, value analysis of white noise constraint, in order to improve the error robustness of microphone characteristics, mainly solves the problems of deviation of the microphone gain, phase shift and position jitter and other factors of uncertainty caused by the low frequency signal of all pass filter, which leads to the performance of the beamformer decreased; finally by Lagrange multiplier method to calculate the fixed weights and iterative convex optimization toolbox to obtain the optimal weight vector. The simulation results show that the algorithm robust performance, low frequency performance robust beamformer has the improved array, high gain, good direction for.3. closed environment is often caused by the reverberation effect of speech signal loss of clarity the improved Wiener filter, rear beam reverberation frequency based on linear constrained minimum variance is given The formation algorithm based on Wiener post filtering structure delay weighted sum beamforming was improved by linear constrained minimum variance beamforming criterion. The algorithm begins with the assumption that the characteristics of each band in different reverberation time, microphone array receiving signal frequency processing, beamforming algorithm is applied to the low frequency subband in improving the accuracy of reverberation suppression; second microphone array receives the reverberation signal between the direct wave and reflected wave is not related to the spatial information of a received signal using a microphone array to solve the estimation problem of the Wiener filter. The simulation test results show that this algorithm has obvious improvement of reverberation suppression and improve the.4. evaluation. The index score system according to physical characteristics of microphone channel mismatch and reverberation caused by the enclosed space of speech enhancement The clarity of the influence, on the audio signal analysis and data acquisition experiment. First, build in anechoic chamber experimental platform, fixed sound source collecting microphone array data to validate the robust algorithm, the data collected in this paper; secondly, to build a strong reverberation environment in the garage experiment platform, fixed source acquisition reverberation array the data on the data collected in this paper the reverberation suppression algorithm verification. Through the algorithm output signal of anechoic room and garage actual environment according to the measured spectrum experiment results show that the proposed algorithm and the simulation results as expected.

【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3

【参考文献】