Binaural cues estimates based on Interaural Matching Filter for sound source localization

Abstract

Robot audition is a natural and convenient way for robot to interact with outside. Binaural sound source localization for robot audition plays an important role in human-robot interaction, speech capturing, enhancement and communication, etc. For the issue, interaural time difference (ITD) and interaural level difference (ILD) are two essential binaural (interaural) cues, yet they are still difficult to extract in complex environments. In this context, this paper proposes a new scenario to simultaneously estimate the ITD and ILD based on Interaural Matching Filter (IMF). The IMF is an optimal filter in view of eliminating the disparity of binaural signals so that it implies some information of binaural cues. Firstly, the IMF is decomposed into a minimum phase component and an all-pass component by homomorphic filtering. Subsequently, ITD is evaluated from the phase response of all-pass component, and ILD is yielded by the amplitude response of minimum phase component. Experiments verify the effectiveness of our method to preserve binaural cues, and it is robust for sound localization.