The outcome of this research (carried out in cooperation with IIT's PAVIS department) is a novel technique that jointly solves the problem of microphone and audio source localization in indoor environments together with room geometry estimation as a finite set of planar surfaces. The method is fully uncalibrated since, differently from previous methods in the literature, no prior knowledge is required about microphone, source and surface positions, except a few assumptions on the number of planar surfaces and room convexity. In particular, no initialization at hand is required and no further assumption e.g. on microphone array compactness is assumed. As a remarkable feature, the method bypasses the NP-hard problem of “echo labeling” that forced most of the previous methods to rely on heuristics in order to avoid unfeasible computational burden. Results on synthetic experiments and on real data account for the effectiveness of the method and its practical usefulness.
Authors: Andrea Trucco, Marco Crocco, Alessio Del Bue