Audio source localization and tracking in reverberant environments

Abstract

Audio source localization and tracking (ASLT) in reverberant environment has been a challenging research goal for many years. The failure of conventional ASLT algorithms can be contributed to the improper modeling of reverberation as non-correlated Gaussian noise, as well as to the lack of consideration of previous estimations. Therefore, in this thesis, the Image Method based reverberant room impulse response simulation is discussed and improved to provide an accurate research basis for reverberation robustness ASLT techniques. The Particle Filtering (PF) algorithm, as an approximation to the Bayesian Filtering problem, is then studied in detail. The PF algorithm in theory provides an optimal solution to involve previous estimations to tackle the reverberation robustness problem. Two novel hybrid PF algorithms with the source height estimation based hypothesis systems are proposed in this thesis. The proposed hybrid PF algorithms outperform the existing PF algorithms in the reverberation robust ASLT implementation.