基于差分麦克风阵列的声源定位方法研究

发布时间：2018-02-20 22:47

本文关键词： 差分麦克风阵列声强估计时频稀疏性模糊聚类声源定位　出处：《南京航空航天大学》2016年硕士论文　论文类型：学位论文

【摘要】：麦克风阵列声源定位一直是语音信号处理领域研究的热点之一。经典的声源定位方法在小尺寸麦克风阵列条件下的性能变差,精度无法满足实际需要。因此,如何有效实现小尺寸阵列和混响噪声环境下的声源定位是麦克风阵列处理领域中有待进一步解决的关键问题。而基于差分麦克风阵列的声强估计方法为实现小尺寸阵列条件下的声源定位提供了一条重要技术途径。本文以差分麦克风阵列为研究对象,以声强估计作为主要定位手段,从单声源和多声源两个角度进行讨论,并在此基础上对基本方法进行改进,从而实现声源方位的准确定位。具体的工作如下:首先介绍了差分麦克风阵列定位的基本理论,然后详细介绍了基于声强估计、基于时延估计和基于可控波束形成的声源定位方法的基本原理,并对其性能进行仿真分析和比较。结果表明,基于声强估计的声源定位方法要优于其他两种方法,为接下来提出的算法奠定了基础。针对声强估计方法在高混响下性能变差的缺陷,本文结合局部标准差和瞬时方位估计值之间的关系提出了基于局部标准差加权的方法。仿真结果表明,此方法有一定的效果,但是改进效果并不明显。于是,本文提出了基于二次门限时频掩蔽的定位方法。该方法从时频域角度进行了改进,定位精度得到提升。此外,本文还从空域角度进行了算法改进,提出了基于波束指向性加权的方法,初步结果表明,此方法有效地提高了声源定位的性能。最后,本文结合时频域和空域改进方法的优点,提出了基于时频空域联合处理的方法,仿真和实测数据结果表明,该方法要优于只在时频域或空域处理的方法。本文还对多声源的声强估计声源定位方法进行研究,分析了现有的直方图方法的性能,提出了基于时频掩蔽和模糊聚类分析的多声源定位方法。从仿真结果和实验数据分析表明,本文提出的方法要优于直方图方法。
[Abstract]:Microphone array sound source localization has been one of the hotspots in the field of speech signal processing. The performance of the classical sound source localization method under the condition of small microphone array becomes worse, and the accuracy can not meet the actual needs. How to effectively realize sound source location in the environment of small array and reverberation noise is a key problem to be solved in the field of microphone array processing. The method of sound intensity estimation based on differential microphone array is to realize small scale. The sound source location under the condition of inch array provides an important technical approach. In this paper, the differential microphone array is taken as the research object. The sound intensity estimation is used as the main localization method, and the basic method is improved from the two angles of single sound source and multi sound source. The main work is as follows: firstly, the basic theory of differential microphone array positioning is introduced, and then the sound intensity estimation is introduced in detail. The principle of acoustic source location based on time-delay estimation and controllable beamforming is analyzed and compared. The results show that the sound source location method based on sound intensity estimation is superior to the other two methods. For the defects of the sound intensity estimation method under high reverberation, Based on the relation between local standard deviation and instantaneous azimuth estimation, a weighted method based on local standard deviation is proposed in this paper. The simulation results show that the method has some effect, but the improvement effect is not obvious. In this paper, a new location method based on quadratic threshold time-frequency masking is proposed. This method is improved from the angle of time-frequency domain, and the accuracy of location is improved. In addition, the algorithm is improved from the perspective of spatial domain. A beam-directivity weighted method is proposed. The preliminary results show that the proposed method can effectively improve the performance of sound source location. Finally, this paper combines the advantages of time-frequency domain and spatial domain improved methods. A method based on joint processing in time-frequency space domain is proposed. The simulation and measured data show that this method is better than that only in time-frequency domain or spatial domain. The performance of the existing histogram method is analyzed, and a multi-sound source location method based on time-frequency masking and fuzzy clustering analysis is proposed. The simulation results and experimental data show that the proposed method is better than the histogram method.
【学位授予单位】：南京航空航天大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.3

【参考文献】