MicEye: Locating Sound Source in 3D with Small-size Microphone Array

Indoor Sound Source Localization (ISSL) is under growing focus with the rapid development of intelligent voice assistants. The predominate approaches are to find multiple angles of arrival (AoAs) based on the array received signals and then retrace the source by triangulation method. The performance of these solutions is bounded by the size of array and they are rarely used to locate 3D sound sources with only a small-size microphone array. In this paper, we propose an ISSL system, named MicEye, a small-size microphone based solution that utilizes several time difference of arrival (TDOA) for the multipath signals to achieve sound source localization in 3D. By combining the cross correlation spectra of the array signals and sound propagation geometry, the proposed MicEye can efficiently generate a localization heatmap by picking and summing the power of correlation peaks at TDOA delays of multipaths in both 2D and 3D. Extensive experiment results show that the MicEye can effectively locate sound sources in different rooms, exhibiting about 2× improved accuracy and $\frac{1}{{10}}$ latency of the State-of-the-Art (SoTA) solutions.