作者: Reinhold Haeb-Umbach , Sven Peschke , Ernst Warsitz
DOI:
关键词:
摘要: While the main objective of adaptive Filter-and-Sum beamforming is to obtain an enhanced speech signal for subsequent processing like recognition, we show how speaker localization information can be derived from filter coefficients. To increase accuracy, tracking performed by non-linear Bayesian state estimation, which realized sequential Monte Carlo methods. Improved acquisition and performance was achieved even in highly reverberant environments, comparison with both a Kalman Filter recently proposed Particle operating on output nonadaptive Delay-and-Sum beamformer.