基于麦克风阵列的多声源测向方法研究 - 图文 下载本文

分类号 密级 UDC注1

学 位 论 文

基于麦克风阵列的多声源测向方法研究

(题名和副题名)

(作者姓名)

指导教师姓名

申请学位级别 硕 士 专业名称 电路与系统 论文提交日期 2014.02

论文答辩日期 2014.03

京 理 工 大 学

答辩委员会主席

评阅人

2014年 2月 24 日

学位授予单位和日期 南 注1:注明《国际十进分类法UDC》的类号。

声 明

本学位论文是我在导师的指导下取得的研究成果,尽我所知,在本学位论文中,除了加以标注和致谢的部分外,不包含其他人已经发表或公布过的研究成果,也不包含我为获得任何教育机构的学位或学历而使用过的材料。与我一同工作的同事对本学位论文做出的贡献均已在论文中作了明确的说明。

研究生签名: 年 月 日

学位论文使用授权声明

南京理工大学有权保存本学位论文的电子和纸质文档,可以借阅或上网公布本学位论文的部分或全部内容,可以向有关部门或机构送交并授权其保存、借阅或上网公布本学位论文的部分或全部内容。对于保密论文,按保密的有关规定和程序处理。

研究生签名: 年 月 日

硕士论文 基于麦克风阵列的多声源测向方法研究

摘 要

基于麦克风阵列的多声源测向技术通过对麦克风阵列接收的多声源混合信号进行处理,从而确定各个声源的方位。它在很多领域都具有广泛的应用前景和实际意义,如在民用方面的视/音频会议、语音识别及增强等领域中,常利用声源测向技术精确估计出说话人位置来控制摄像头,使其自动对该位置的语音信号进行增强。在军事方面声源测向技术被广泛地应用在飞机,火炮、狙击手探测等方面。因此,该技术成为了语音信号处理领域的研究热点之一。

本课题针对基于麦克风阵列多声源测向问题展开研究,归纳总结并比较了传统的几类声源测向方法。本文以典型的双阵元麦克风阵列为研究对象,针对远场多声源模型,将基于语音信号时频正交特性的退化分离估计技术(DUET)应用于声源信号测向。该算法利用了语音信号特有的时频稀疏和短时正交特性(W-Disjoint Orthogonality,W-DO),基于此特性的时延估计算法计算量小,实现简单,仅用两个麦克风就可以实现多个声源的方位测向。但是当声源存在波长小于两倍阵元间距的高频成分时,此类声源测向方法将出现相位卷绕模糊问题,而阵元间距因物理尺寸限制也不可能无限缩小,因此限制了该类方法的实际应用领域。针对上述问题,本文提出了一种基于迭代时频掩蔽的宽间距麦克风阵列多声源测向方法,该方法通过迭代消去过程,显著抑制了相位卷绕产生的影响。此外,结合基于能量的语音端点检测技术,本文进一步给出了上述方法的实时处理算法步骤。针对上述方法,本文进行了仿真实验和相关外场实验,实验结果表明:针对宽间距麦克风阵列多声源测向,本文所述方法明显优于常规DUET类方法,具有一定的实际应用价值。

关键词:麦克风阵列 多源测向 时延估计 相位模糊 实时处理

I

Abstract 硕士论文

Abstract

Measuring multiple source direction in a microphone-array refers to that the direction of acousitc source is determined by mixed-signal imformation received from microphones. It is widely used in many areas, such as video / audio conferencing, speech recognition and speech enhancement etc. By estimating the speaker position to control the microphone and camera, then the automatic speech signal of the speaker can be enhanced.In the military area, the technology is widely used in the aspects of sniper detection and target detection of aircraft, artillery etc.Therefore, multiple sound source direction measurement is a new hotspot in acoustic signal processing field.

This thesis mainly focuses on a study based on multiple sound source direction measurement technology. And the several categories of traditional sound source direction measurements are summarized and compared firstly. In this paper, the typical dual-microphone array is studied, focused on far-field multiple sound source mode, the degenerate unmixing estimation technique (DUET) based on W-Disjoint Orthogonality (W-DO) of the source signals is applied to the acousitc source direction measurement. The time delay estimation algorithm based on this characteristic has simple implementation, little computation.And it can measurement the directions of multiple acoustic sources with only two microphones.However, when the wavelength is less than twice the spacing of the two microphones, this kind of algorithm is prone to phase wrap-around aliasing, which often leads to artifacts. However the spacing can not be infinitely reduced, thus the practical applications is limited of such methods. In response to these problems, an approach to correct the phase wrap-around aliasing based on an iterative time-frequency masking process is presented in this paper. By iteratively clustering in the masked time-frequency plane and the artifacts due to the phase wrap-around aliasing can be extremely suppressed. In addition, combined with the speech endpoint detection technology that based on energy, the paper puts forward a real-time processing algorithm. For the above method, simulation and outdoor experiments are taken. The experimental results show that the method is superior to conventional DUET method, which proves that the method has a great practical application value.

KeyWords:Microphone array, Multiple source localization, Time delay estimation,

Disambiguity, Real-time processsing

II