基于麦克风阵列的相容时延矢量声源定位方法

时延估计是常用的声源定位方法,传统的算法将定位分为两个步骤,即先估计麦克风阵列中每一对基元的接收信号时延,然后根据这些时延用几何的方法确定声源的位置。在低信噪比下,一对麦克风的时延估计误差较大,导致定位误差较大。相容时延矢量估计算法将两步合为一步,没有逐对估计时延,而是构造一个目标函数,通过搜索得到声源的位置。仿真结果表明,在低信噪比下,只需要较短的数据,该算法仍可得到较高的定位精度。

A consistent delay vector estimation algorithm for speech source localization based on microphone array

Time difference of arrive （TDOA） estimation is one of speech source localization algorithm. A traditional two-step procedure is adopted. TDOA estimates of the speech signal relative to pairs of spatially separated microphones are obtained and then applied to generate a source location estimate. However, in low SNR environment, TDOA estimates of a pair of microphones may have a large error. So source location estimation be significantly biased away from its true location. A consistent delay vector estimation algorithm is presented in this paper. Without estimating independently TDOA for every pair of microphones, it generates an objective function based on the all array. The maximum objective function value occurs at the true time delay vector. So the source location can be estimated by perform a search. The simulation results show that the algorithm can improves localization precision in case of low SNR and short signal time.