Robotic Binaural and Monaural Information Fusion Using Bayesian Networks

Abstract

Auditory signal processing already starts outside the head. The external sound field has to couple into the ear canals. The relative positions of the two ear canals and the sound source lead to a coupling that is strongly depend on frequency. In this context not only the two pinnae but also the whole head have an important functional role, which is best described as a spatial filtering process. This linear filtering is usually quantified in terms of so-called head- related transfer functions, HRTFs, which can also be interpreted as the directivity characteristics of the two ears. In this work, the HRTF constitutes the cornerstone of a sound localization algorithm, which uses Bayesian information fusion to increase the localization resolution in a three-dimensional reverberant environment. The algorithm was tested in simulations as well as in a household environment. Compared to existing techniques, the method is able to localize, with higher accuracy, 3D sound sources under high reverberation conditions.