Moving Sound Source Localization Based on Sequential Subspace Estimation in Actual Room Environments

Abstract

This paper presents a novel method for moving sound source localization and describes its performance evaluation in actual room environments. The method is based on MUSIC (MUltiple SIgnal Classification) which is one of the highest-resolution localization methods. When using MUSIC, computation of eigenvectors of correlation matrix is required for estimation, which often results in a high computational cost. In particular, in the case of a moving source, this is a crucial drawback, because estimation must be performed at every observation time. Furthermore, since the correlation matrix varies its characteristics due to spatial-temporal nonstationarity, the matrix must be estimated using only a few observed samples, which degrades the estimation accuracy. In this paper, PAST (Projection Approximation Subspace Tracking) is applied for sequential estimation of the eigenvectors spanning the subspace. In PAST, eigen-decomposition is not required, and therefore it is possible to reduce the computational costs. Several experimental results in actual room environments are presented in order to demonstrate the superior performance of the proposed method. © 2011 Wiley Periodicals, Inc. Electron Comm Jpn, 94(7): 17–26, 2011; Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/ecj.10333