Bayesian framework for direction of arrival analysis using spherical harmonics

Abstract

This work discusses a system of Direction of Arrival (DoA) analysis for an unknown number of sources using spherical beamforming and Bayesian inference with a spherical microphone array. In this system, sound pressure data is collected by the spherical array and processed using beamforming techniques in the spherical harmonic domain. A two-tiered framework of Bayesian inference is then used to evaluate analytic models against the beamformed data in order to determine the number, DoA(s), and strength(s) of the detected sources. Two series of tests are conducted: one sets out to determine the performance of this algorithm when two sources are close together and the other to evaluate its performance in a reverberant environment. Results showed that the performance of the system degrades when sources are closer than 45° to one another and that it fails to distinguish sources at 15°. Additionally, the algorithm failed to identify or locate a single sound source in an environment featuring significant reverberation.