Real-Time Sound Source Localization for a Mobile Robot Based on the Guided Spectral-Temporal Position Method

Abstract

Sound source localization based on a microphone array for mobile robots faces great challenges because of factors such as uncertainty of the robot's movement, noise and reverberation, the requirements of a compact microphone array and so on. This paper studies a method for sound source localization in a dynamic environment using a cross-microphone array plane, which consists of four microphones on an intelligent mobile robot platform. Firstly, the method of spectral subtraction and campestral mean normalization is introduced to de-noise. Then GCC-PHAT-ργ and the guided spectral-temporal (ST) position method are proposed to suppress noise and reverberation based on the generalized cross-correlation method to estimate time delay. Finally, the sound source is positioned by adopting the geometric location method. This system is tested by a total of 2,016 sets of experiments. Even in an intensely noisy and reverberating environment, the guided ST position method achieves angle positioning accuracy of more than 95% with a less than 15 degrees localization error. Meanwhile, all the experimental data can be processed in real-time, within 0.4s.