2011文献Approximated kernel density estimation for multiple TDOA detection.pdfVIP

APPROXIMATED KERNEL DENSITY ESTIMATION FOR MULTIPLE TDOA DETECTION Francesco Nesta and Maurizio Omologo Fondazione Bruno Kessler - irst Email: {nesta, omologo}@fbk.eu ABSTRACT but it becomes computationally infeasible when a large microphone spacing is used. In the following it is shown that an efficient approx- The Generalized State Coherence Transform (GSCT) has been re- imation of the kernel may be obtained through a GSCT based on cently proposed as an efficient tool for the estimation of multidi- a novel frequency-dependent non-linearity g (·) that is derived from mensional TDOA of multiple sources. The transform defines a mul- the underlying distribution. Simulations for the univariate and bi- tivariate likelihood of the TDOA through a non-linear integration of variate cases show that the proposed non-linearity dramatically im- complex-valued states, representing the acoustic propagation of mul- proves both smoothness and resolution of the resulting likelihood tiple sources. In the previous works the non-linearity was heuristi- and the detection of the modes gets more accurate for a wide range cally motivated leading to a difficult interpretation of the resulting