球形传声器阵列下基于主导声源检测MUSIC群时延算法的多声源定位*

针对混响环境中，多径效应、散射、衍射等原因导致声源定位失败或分辨能力不足的现象，提出一种基于主导声源检测MUSIC群时延的邻近多声源定位方法。该方法采用球形传声器阵列，相比平面阵列可以捕获3D声场信息，利用球谐域下信号的频率分量与角度分量解耦的优势，从而可直接利用频率平滑技术处理宽带语声信号而不需要构造聚焦矩阵，并在球谐域下通过设置阈值对一组时频段进行主导声源检测，从而选择出包含直达声的一组时频块来构造MUSIC群时延空间谱。上述举措在提升波达方向估计在高混响环境下定位鲁棒性的同时，也提高了多个邻近声源的分辨能力。仿真实验结果表明，所提出的主导声源检测MUSIC群时延算法，在高混响和低信噪比条件下，仍具有更好的定位精度与更优的邻近多声源分辨效果。

Multi-source localization based on MUSIC-group delay algorithm of direct path dominance test under spherical microphone array

To address the phenomenon that sound source localization fails or the ability is insufficient due to multipath effect, scattering, diffraction, and other reasons in reverberation environment, a novel multi-sound source localization method based on MUSIC-group delay of direct path dominance test is proposed. Compared to the planar array, the spherical microphone array can obtain 3D sound field information. It can take advantage of the decoupling of the frequency component and angle component of the signals in the spherical harmonic domain. Meanwhile, the frequency smoothing technology can be directly used to process broadband speech signals without constructing a focusing matrix. In the spherical harmonic domain, a set of time-frequency components containing direct sound is selected to construct the MUSIC group time delay spatial spectrum by setting a threshold to detect the dominant sound source among time-frequency bands. The measures mentioned above not only improve the localization robustness of direction of arrival estimation in a high reverberation environment, but also improve the resolution of multiple adjacent sound sources. The simulation results show that the novel MUSIC group delay algorithm of direct path dominance test proposed in this paper has better positioning accuracy and better resolution of adjacent multiple sound sources under the high reverberation or low signal-to-noise ratio environments.