On the robustness of distributed EM based BSS in asynchronous distributed microphone array scenarios

Distributed microphone array (DMA) processing has recently started attracting a lot of attention as a promising alternative to conventional microphone arrays with co-located elements. To perform efficient blind source separation (BSS) in distributed manner, we have recently proposed a clustering based BSS approach that utilizes a distributed expectation-maximization algorithm. In this paper, we further investigate the effectiveness of this proposed method in typical DMA situations with drift error, i.e., the inevitable sampling frequency mismatch between different nodes/devices, which is a very important aspect that has yet to be examined. We compare our method experimentally with some conventional methods and their variants newly proposed here, and confirm that our method is robust against drift error and achieves stable performance in various typical DMA scenarios, while the performance of the conventional approaches changes greatly depending on the DMA setup and amount of drift error.