We conducted outdoor sound-source-localization experiments in a 30 m × 50 m field, deploying an omnidirectional speaker and 13 microphones. In the proposed grid-based method, we first compute the cross-correlation function between the signals recorded by each microphone pair and take its peak value to estimate the time difference of arrival (TDOA) for that pair. Given this TDOA and the known coordinates of the two microphones, a hyperbolic function can be derived for the possible presence of a sound source. The estimation plane is divided into a grid, and the peak value of the cross-correlation function is assigned to each grid node that lies on the corresponding hyperbola. This process was repeated for all microphone pairs, and the maximum values of the cross-correlation were added across the entire grid. Finally, the grid point with the highest score is then taken as the estimated sound-source position. Microphone pairs with low value of the maximum cross-correlation function often produce unreliable TDOA estimation, which can affect the localization accuracy. Therefore, in order to select a microphone to be used for sound source localization, a threshold was set on the maximum value of the cross-correlation to see how it would affect the localization accuracy.
Morimoto et al. (Wed,) studied this question.