Real-Time Tracking of Multiple Acoustical Sources Utilising Rao-Blackwellised Particle Filtering

Abstract

This paper proposes a system for localising and tracking multiple simultaneous acoustical sound sources in the spherical harmonic domain, intended as a precursor for developing parametric sound-field editors and spatial audio effects. The real-time system comprises a novel combination of a direct-path dominance test, grid-less subspace localisation, and multiple target tracking based on Rao-Blackwellised particle filtering. It has robust multi-source performance and can adapt to sources that vary in number and direction over time. The proposed system was evaluated by using the framework established under the LOCATA 2018 localisation and tracking challenge, and comparing the results with the original submissions to the challenge. The results demonstrate that the proposed system yields the lowest angular error on the horizontal plane for both moving source(s) tasks, and provides results close to the winning submission for the static source(s) tasks. The proposed system also fares well in the other LOCATA evaluation metrics, and, more importantly, does so as a real-time system; i.e. with no requirement for offline post-processing of the tracker data and operating within reasonable computational constraints.