Microphone Array Acoustic Source Localization system based on Deep Learning

This paper presents a microphone array sound source localization system based on deep learning algorithms. Currently, the most popular acoustic source localization algorithms are based on the traditional array signal processing methods. These methods have good localization performance in the ideal acoustic environment. However, the performances degraded significantly in low signal-to-noise ratio (SNR) and strong reverberation environments. To deal with this problem, this study developed a deep neural network (DNN) based system. Unlike the traditional algorithms with poor adaptability to environmental conditions, the proposed system can automatically learn the spatial information of sound sources under various conditions through training a large amount of data. Furthermore, it can fully utilize all the information of the original data without additional feature extraction. A set of experiments are carried out to evaluate the performance of the proposed system in comparison with the generalized cross correlation phase transform (GCC-PHAT) method. Results verify that the DNN based system achieves higher accuracy under low SNR conditions.