Direction-of-arrival estimations based on a two-microphone array using two levels of Bayesian inference

Abstract

Due to a wide variety of potential applications, such as in video conferencing, mobile devices, and robotics, sound source localization using a small aperture consisting of only two microphones has been actively investigated. Based on the observed time-differences of arrival between sound signals, a probability distribution of the direction of the sources is derived to estimate the actual direction of sources. Many existing algorithms, however, assume a given number of sound sources. This paper describes a recent development in a model-based Bayesian probabilistic approach by Escolano et al. [J. Acoust. Am. Soc. 135, 742–751 (2014)], which allows both the number and direction of speech sources to be inferred. This paper will demonstrate that a unified framework encompassing two levels of Bayesian inference, model selection and parameter estimation, can be efficiently applied in this challenging task. This paper will also present different experimental setups and scenarios to investigate the performance of the proposed method.