改进的自适应特征值分解声源定位算法研究

Study on improved adaptive eigenvalue decomposition algorithm for acoustic source localization

麦克风阵列室内声源定位,常用声达时间差定位技术。本文针对估计时延的自适应特征值分解算法收敛速度慢,时延估计精度较差,麦克风较多等问题,提出一种改进的自适应特征值分解时延估计算法,采用单源多元混响模型,将混响效应描述为室内冲激响应滤波器对信号的滤波过程,估计不同阵元的冲激响应抑制混响,根据冲激响应峰值计算时延。通过引入最小均方牛顿算法,加快了AED算法的收敛速度。给出了对声源进行三维定位的三麦克风阵列结构,实际测试结果表明,改进算法与三麦克风阵列的定位方法对声源的定位更加准确。

Microphone array indoor sound source localization commonly adopts acoustic time differential positioning technology. This paper presents an improved adaptive eigenvalue decomposition time delay estimation algorithm ai- ming at the problems of slow convergence rate, poor time delay estimation accuracy, more microphones and other is- sues of the adaptive eigenvalue decomposition time delay estimation algorithm. In the proposed algorithm, the single- source multi-reverb model is adopted, the reverb effect is described as the signal filtering process of the indoor im- pulse response filter;the impulse responses of different elements in suppressing reverberation are estimated;and ac- cording to the impulse response peak the time delay is calculated. The convergence rate of the AED algorithm is ac- celerated through introducing LMS-Newton algorithm. The three-microphone array structure for the three-dimensional positioning of sound source is given. Actual test was performed, and the test results show that the improved algorithm with three-microphone array sound source localization method is more accurate in the aspect of positioning.