2012文献Generalized State Coherence Transform for Multidimensional TDOA Estimation of Multiple Sources.pdfVIP

246 IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, VOL. 20, NO. 1, JANUARY 2012 Generalized State Coherence Transform for Multidimensional TDOA Estimation of Multiple Sources Francesco Nesta, Member, IEEE, and Maurizio Omologo, Member, IEEE Abstract—According to the physical meaning of the fre- Phase (CSP)[5]. If the activities of the sources are sufficiently quency-domain blind source separation (FD-BSS), each mixing sparse over time and the TDOA measurements associated to all matrix estimated by independent component analysis (ICA) con- the sources have sufficient temporal density, multiple TDOA es- tains information on the physical acoustic propagation related to timation can be performed by short-term spatio-temporal clus- each source and then can be used for localization purposes. In this paper, we analyze the Generalized State Coherence Transform tering [6], [7]. However, in true multiple competitive source sce- (GSCT) which is a non-linear transform of the space repre- narios, these methods fail since the activity of the sources is not sented by the whole demixing matrices. The transform enables sufficiently disjoint, and some of them would tend to dominate an accurate estimation of the propagation time-delay of multiple over the others, at different frequencies. Extensions of the tra- sources in multiple dimensions. Furthermore, it is shown that ditional methods to the multiple source case have been recently with appropriate nonlinearities and a statistical model for the proposed [8]. However, most of them still require microphone reverberation, GSCT c