【作者】 马源；

【导师】 赵力； 余华；

【作者基本信息】 东南大学 ， 电子与通信工程（专业学位）， 2021， 硕士

【摘要】 长期以来,从噪声环境中提取干净的目标语音是一个被广泛研究的课题。传统的单麦克风语音增强算法往往导致目标语音失真,声音质量得不到保证;而麦克风阵列技术可以实现对目标声源的定位以及定向增强,在保证目标语音不失真的前提下获得明显的干扰噪声抑制效果。基于麦克风阵列的声源定位及语音增强方案不仅可以应用于音视频电话会议以提升音频质量,还可以用在智能语音终端上,作为智能设备的声学前端处理器以获得更清晰的语音指令,进而提升人机交互效率。近年来,麦克风阵列技术被广泛研究并取得相当大的进步。本文主要研究了基于麦克风阵列的声源定位算法和语音增强算法,提出了适用于嵌入式设备的实时声源定位和语音增强系统方案,并以DSP芯片为核心搭建了基于均匀圆形麦克风阵列的拾音系统硬件设备作为方案落地平台。本文主要工作和创新点如下:(1)研究了麦克风阵列技术的背景以及国内外发展现状,特别是基于麦克风阵列的声源定位及语音增强技术的发展历程。研究了麦克风阵列信号处理基础知识。调研了语音信号的产生方式及其特性;学习并应用了针对语音信号特性进行的预处理手段;研究了麦克风阵列拓扑结构,以及构建麦克风阵列时需要考虑的阵元间距的选择问题。(2)提出了基于改进相关性时延估计的低复杂度声源定位方案。首先研究分析了基于高分辨率谱估计的声源定位算法、基于可控波束响应技术的声源定位算法和基于到达时间差(TDOA)的两步声源定位算法这三种经典的麦克风阵列声源定位技术,并以基于到达时间差的声源定位算法为基础,研究并提出了基于改进相关性时延估计的低复杂度声源定位方案。该声源定位方案主要有三点创新,分别是:在时延估计过程中,对广义互相关函数的加权函数进行低通滤波以保证噪声环境中的时延计算鲁棒性,同时进一步降低算法计算复杂度,更好的保证算法实时性;在利用时延估计结果确定声源方位时,借鉴了基于可控波束形成技术的声源定位算法的全局搜索策略,以最小均方误差为准则确定声源方位;针对声源方位动态变化场景下的动态帧数声源定位平滑策略,动态帧数平滑策略综合保证了定位结果的准确性以及对声源方位变化跟踪的敏捷性。仿真实验证明本文提出的基于改进相关性时延估计的低复杂度声源定位方案性能优于普通基于相位变换广义互相关时延估计定位算法。(3)提出了基于噪声建模和后滤波的麦克风阵列语音增强算法。首先研究了几种典型的多通道语音增强算法,包括固定波束形成、自适应波束形成以及后置滤波算法,具体分析了这些典型算法的实现原理、优缺点和适用场合。以后置滤波算法为基础,提出基于噪声建模和后滤波的麦克风阵列语音增强算法,出于对实时性与鲁棒性的考量,本文提出的语音增强算法采用固定波束形成器与后置滤波器级联的形式。该算法的创新点在于对声场中目标信号、方向性干扰信号以及环境噪声的综合建模分析以获得目标信号分量的最优估计。仿真实验证明本文提出的基于噪声建模和后滤波的麦克风阵列语音增强算法的性能相较于传统延时求和波束形成算法有显著的提升。(4)设计并搭建了麦克风阵列拾音系统,并实验证实了此麦克风阵列拾音系统的优异性能。拾音系统以嵌入式处理器TMS320C6748为核心,搭配TLV320AIC3106低功耗立体声音频编解码器及2559功率放大器;采用4个SPH0642HT5H-1硅麦构建均匀圆形麦克风阵列。基于此拾音系统硬件设备进行实验测试,测试结果表明本文设计的麦克风阵列拾音系统实际性能较优。更多还原

【Abstract】 For a long time,extracting clean target speech from a noisy environment has been a subject of extensive research.Traditional single-microphone speech enhancement algorithms often lead to distortion of the target speech,and the sound quality cannot be guaranteed;while the microphone array technology can realize the localization and directional enhancement of the target sound source,and obtain obvious interference noise under the premise of ensuring that the target speech is not distorted Inhibitory effect.The microphone array-based sound source location and voice enhancement solution can not only be applied to audio and video conference calls to improve audio quality,but also can be used on smart voice terminals as the acoustic front-end processor of smart devices to obtain clearer voice commands,and then Improve the efficiency of human-computer interaction.In recent years,microphone array technology has been extensively studied and made considerable progress.This paper mainly studies the sound source localization algorithm and speech enhancement algorithm based on the microphone array,proposes a real-time sound source localization and speech enhancement system solution suitable for embedded devices,and builds a uniform circular microphone array based on the DSP chip.The hardware equipment of the sound pickup system is used as the landing platform of the scheme.The main work and innovations of this paper are as follows:(1)The background of microphone array technology and the development status at home and abroad are studied,especially the development history of sound source localization and speech enhancement technology based on microphone array.The basic knowledge of microphone array signal processing is studied.The generation method and characteristics of the voice signal are investigated;the preprocessing methods for the characteristics of the voice signal are learned and applied;the topological structure of the microphone array and the selection of the array element spacing that need to be considered when constructing the microphone array are studied.(2)A low-complexity sound source localization scheme based on improved correlation time delay estimation is proposed.Firstly,three classical microphone array sound sources are studied and analyzed: a sound source localization algorithm based on high-resolution spectrum estimation,a sound source localization algorithm based on controllable beam response technology,and a two-step sound source localization algorithm based on time difference of arrival(TDOA).Localization technology,and based on the sound source localization algorithm based on the time difference of arrival,research and propose a low-complexity sound source localization scheme based on improved correlation time delay estimation.The sound source localization scheme has three main innovations,namely: in the time delay estimation process,the weighting function of the generalized cross-correlation function is low-pass filtered to ensure the robustness of the time delay calculation in the noisy environment,while further reducing the algorithm calculation Complexity,to better ensure the real-time performance of the algorithm;when using the time delay estimation result to determine the sound source position,the global search strategy of the sound source localization algorithm based on the controllable beamforming technology is used for reference,and the minimum mean square error is used as the criterion to determine the sound Source orientation: For the dynamic frame number sound source location smoothing strategy under the scene of the dynamic change of the sound source orientation,the dynamic frame number smoothing strategy comprehensively guarantees the accuracy of the positioning result and the agility of tracking the sound source orientation change.Simulation experiments prove that the low-complexity sound source localization scheme based on improved correlation time delay estimation is better than ordinary generalized cross-correlation time delay estimation localization algorithms based on phase transformation.(3)A microphone array speech enhancement algorithm based on noise modeling and postfiltering is proposed.Firstly,several typical multi-channel speech enhancement algorithms are studied,including fixed beamforming,adaptive beamforming,and post-filtering algorithms.The implementation principles,advantages and disadvantages and applicable occasions of these typical algorithms are analyzed in detail.Based on the post-filtering algorithm,a microphone array speech enhancement algorithm based on noise modeling and post-filtering is proposed.For the consideration of real-time and robustness,the speech enhancement algorithm proposed in this paper uses a fixed beamformer and post-filter The form of cascade.The innovation of this algorithm lies in the comprehensive modeling and analysis of the target signal,directional interference signal and environmental noise in the sound field to obtain the optimal estimation of the target signal component.The simulation experiment proves that the performance of the microphone array speech enhancement algorithm based on noise modeling and post-filtering proposed in this paper is significantly improved compared to the traditional delay-sum beamforming algorithm.(4)The microphone array pickup system was designed and built,and the experiment proved the excellent performance of the microphone array pickup system.The sound pickup system is based on the embedded processor TMS320C6748,with TLV320AIC3106 low-power stereo audio codec and 2559 power amplifier;4 SPH0642HT5H-1 silicon microphones are used to construct a uniform circular microphone array.Based on this sound pickup system hardware equipment for experimental testing,the test results show that the actual performance of the microphone array pickup system designed in this paper is better.更多还原