基于麦克风阵列的近场声源定位与跟踪

发布时间：2018-08-23 15:55

【摘要】：随着阵列信号处理技术的日渐成熟,基于麦克风阵列的声源定位与跟踪现已逐渐得到比较广泛的应用。由于室内声源为宽带非平稳信号,传统的窄带信号DOA估计和跟踪算法无法直接应用于此,且现有算法通常具有较高的算法复杂度,所以基于麦克风阵列的声源定位和跟踪仍然具有较大的改进空间。本文对基于麦克风阵列的近场声源DOA估计和跟踪相关算法进行了研究和改进,主要研究内容包括：第一、综合分析了语音信号的时频特性,结合阵列信号处理中远场均匀线阵平面波信号接收模型研究了均匀线阵、均匀圆阵和任意拓扑结构的近场球面波信号接收模型。第二、对麦克风阵列接收到的数据进行包括预滤波、预加重、归一化、加窗分帧和语音降噪等在内的各种预处理和语音端点检测。本文对语音降噪进行了重点研究,文中采用了自适应小波分解层数选取和改进型阈值函数相结合的方法来提高小波语音去噪的性能。第三、采用麦克风均匀圆阵近场模型,对近场3D-MUSIC算法和宽带聚焦3D-MUSIC算法进行了对比研究。针对均匀圆阵等传统麦克风阵列近场信号模型对声源俯仰角估计产在角度模糊的缺陷,建立了新的麦克风阵列模型；并在此基础上针对宽带聚焦3D-MUSIC算法中三维平均空间谱矩阵求解及谱峰搜索计算量大的问题,提出了分步降维估计法来减小算法计算量。最后,通过实验仿真验证了该方法在降低算法计算量的基础上,依然保持了良好的DOA估计性能。第四、将基于压缩投影逼近子空间(PASTd)算法的信号DOA跟踪,应用到三维近场声源跟踪,并利用第三点所述的分步降维估计法来减小每帧数据DOA估计时的计算量。针对该算法跟踪误差较大或收敛速度较慢的缺点,提出了可变遗忘因子PASTd算法,最后通过实验仿真验证了改进算法良好的DOA跟踪性能。
[Abstract]:With the development of array signal processing technology, sound source location and tracking based on microphone array has been widely used. Because the indoor sound source is a wideband non-stationary signal, the traditional narrowband signal DOA estimation and tracking algorithms can not be directly applied here, and the existing algorithms usually have high algorithm complexity. Therefore, the sound source location and tracking based on microphone array still has great improvement space. In this paper, the DOA estimation and tracking algorithms for near-field acoustic sources based on microphone array are studied and improved. The main research contents are as follows: first, the time-frequency characteristics of speech signals are analyzed synthetically. Combined with the plane wave reception model of far-field uniform linear array in array signal processing, the near-field spherical wave signal receiving model of uniform linear array, uniform circular array and arbitrary topology is studied. Secondly, all kinds of preprocessing and speech endpoint detection including pre-filtering, pre-weighting, normalization, windowed framing and speech denoising are carried out for the data received by the microphone array. This paper focuses on the research of speech denoising. In this paper, the adaptive wavelet decomposition layer selection and the improved threshold function are used to improve the performance of wavelet speech denoising. Thirdly, the near-field 3D-MUSIC algorithm and wideband focused 3D-MUSIC algorithm are compared by using the near-field model of microphone uniform circular array. A new microphone array model is proposed to solve the problem that the traditional near-field signal model of microphone array, such as uniform circular array, produces fuzzy angle to estimate pitch angle of sound source. On this basis, aiming at the problem of large amount of computation in solving the three-dimensional average spatial spectral matrix and searching the spectral peak in the wideband focused 3D-MUSIC algorithm, a fractional step reduced dimension estimation method is proposed to reduce the computational complexity of the algorithm. Finally, the experimental results show that the proposed method still maintains good DOA estimation performance on the basis of reducing the computational complexity of the algorithm. Fourthly, the signal DOA tracking based on compressed projection approximation subspace (PASTd) algorithm is applied to 3D near-field sound source tracking, and the fractional step dimensionality reduction method proposed in the third point is used to reduce the computational complexity of DOA estimation for each frame. A variable forgetting factor (PASTd) algorithm is proposed to overcome the disadvantages of large tracking error and slow convergence rate of the algorithm. Finally, the improved DOA tracking performance is verified by experimental simulation.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN912.3

【参考文献】