Sound source localization method based on dual microarrays and deep learning

Abstract

In order to improve the localization accuracy in complex environments, a sound source localization method based on dual microarrays (DMA) and deep learning is studied. Generalized cross correlation-phase transform (GCCPHAT) sequence and the maximum value information of the sequence are used as localization cues, the three-dimensional coordinates of the sound source are used as the output of the network, and the mapping rules from input features to output are learned through the improved CNN network based on VGG16 network structure (referred to as V_CNN for short). Through simulation experiments, the sound source localization method based on circular array and V_CNN, the sound source localization method based on dual microarrays and ordinary convolutional neural network (CNN), and the sound source localization method based on dual microarrays and V_CNN are compared. The experimental results show that the sound source localization method in this paper has high localization accuracy under different noise and reverberation environments.