Research on Sound Source Localization Algorithm Based on Multilayer Perceptron

Abstract

Sound source localization technology has been developing rapidly in the military and civil fields; however, it still has certain limitations in large-scale manufactory or industrial site with the noise and complex sound field environments. So, the single sound source localization algorithm based on the microphone array model of the multilayer perceptron was studied, firstly, a three-point microphone array model was built to collect and process the sound wave signals, then the Time-Differences-of-Arrival (TDOA) and sound pressure level of microphone signals were extracted to as feature vectors, and multilayer perceptron sound source localization algorithm with multiple features and multiple hidden layers was constructed. Based on the python machine learning toolbox-sklearn, the algorithm was implemented and simulated, and the result shows that it has better performance and robustness compared with other machine learning sound source localization algorithms.