Adaptive Neuro-Fuzzy Inference System for Binaural Source Localisation

This paper shall present the effectiveness of a black box model to estimate the azimuth angle of a sound source using two microphones. Two electret microphones are placed at the ear positions of a polystyrene dummy head with human-like ear-to-ear separation. The interaural time difference (ITD) and interaural level difference (ILD) are two cues that are used for binaural source localisation together with the frequency content of the sound source. The black box structure proposed is a Sugeno type fuzzy inference system (FIS). The structure of the selected FIS is adaptively tuned by presenting it with training data consisting of ITD, ILD and frequency content. Training was successful and the results achieved on azimuth angle estimation for both bandlimited white noise and speech sound sources are presented. These results are comparable to what is currently state of the art.