Indoor Sound Source Localization based on Sparse Bayesian Learning and Compressed Data

In this paper, the problems of indoor sound source localization using a wireless acoustic sensor network are addressed and a new sparse Bayesian learning based algorithm is proposed. Using time delays for the direct paths from candidate source locations to microphone nodes, the proposed algorithm estimates the most likely source location. To reduce the amount of data that must be exchanged between microphone nodes, a Gaussian measurement matrix is multiplied on to each channel and the proposed method operates directly on the compressed data. This is achieved by exploiting sparsity in both the frequency and space domains. The performance is analysed in numerical simulations, where the performance as a function of the reverberation times in investigated, and the results show that the proposed algorithm is robust to reverberation.