基于密度聚类的直达声波达方向估计

Direct-path DOA Estimation Based on Density Clustering

混响环境下的直达声波达方向(Direction of Arrival,DOA)估计采用基于阈值挑选时频点的方式,可以有效地提升直达声DOA估计精度。但用基于阈值挑选出的时频点,所得到的估计结果混有偏离较大的离群点,影响直达声DOA估计精度。为了削弱极少数偏离较大离群点的干扰,采用空间聚类(DBSCAN)算法,提出了基于密度的挑选时频点的直达声DOA估计。与基于阈值挑选时频点相比,利用基于密度挑选出的时频点,剔除了部分偏差较大的离群点,留下了含有直达声信息的时频点,提高了直达声DOA估计精度。

In order to reduce the interference of a few outliers,a density based spatial clustering(DBSCAN)algorithm is proposed to select time-frequency points for DOA estimation.Compared with time-frequency points selected based on the threshold value,the time-frequency points selected based on the density are used to eliminate some outliers with large deviation,leaving time-frequency points with direct sound information,which improves the DOA estimation results of direct sound.