强混响环境下基于声矢量传感器的多源DOA估计

Multi-source DOA Estimation Based on Single Acoustic Vector Sensor with Strong Reverberation

声矢量传感器(Acoustic Vector Sensor,AVS)是一种轻量的、空间指向性强的麦克风阵列,广泛应用于声源定位领域中。混响环境下基于单个AVS的多源波达方向(Direction of Arrival,DOA)估计是声学领域的一个研究热点,然而随着混响程度增加,使用单个AVS进行多源DOA估计的精度显著下降。为突破单个AVS在强混响环境下多源DOA估计的性能局限,基于单个AVS设计了一种有效的算法来实现强混响环境下的多声源测向。分析了精确估计DOA的关键在于精确检测时频域上直达声占主导的单源时频点(Direct-Path-Single-Source-Point,DP-SSP);提出了一种两阶段的DP-SSP检测方法,在第一阶段采用频率平滑来减轻混响的影响,第二阶段采用基于子空间的方法来精确提取DP-SSP;对每个源对应的DP-SSP进行单源DOA估计,最终得到多源DOA估计结果。仿真结果表明,所提方法相比其他方法能更准确地估计声源数目并且大大提升多源DOA估计的精度。

The Acoustic Vector Sensor(AVS)is a lightweight microphone with strong spatial directivity,which is widely used in the field of sound source localization.Multi-source Directional of Arrival(DOA)estimation under reverberation utilizing AVS is a research focus in acoustic realm.Increasing reverberation brings huge challenges to the improvement in the accuracy of multi-source DOA estimation.An effective algorithm is proposed to address multi-source direction finding with a single AVS in highly reverberant cases.First,analysis is conducted to prove that the essence of accurate DOA estimation lies in the detection of Direct-Path-Single-Source-Point(DP-SSP)in the time frequency domain,which are dominated by a single actual source.Next,a two-step DP-SSP detection technique is designed where frequency smoothing is applied in the first step to eliminate the effects of reverberation and the DP-SSP are precisely identified in the second step by the subspace-based method.Finally,DOA estimation is separately implemented on a set of DP-SSP corresponding to each source.Experimental results demonstrate the effectiveness of the proposed method and the accuracy of DOA estimation is improved especially in highly reverberant cases compared with the state-of-the-art ones.