Network topology selection for distributed speech enhancement in wireless acoustic sensor networks

A wireless acoustic sensor network is envisaged where each node estimates a locally observed speech signal that has been corrupted by additive noise. The nodes perform noise reduction by means of the distributed adaptive node-specific signal estimation algorithm in a tree topology (T-DANSE). The T-DANSE algorithm inherently relies on a network that has been pruned to a tree topology where a single node has been designated as the root node. We will demonstrate that, due to the data-driven flow of the T-DANSE algorithm and unavoidable errors in the estimation of certain second-order statistics, the selection of the root node and the pruning of an adhoc network to a tree topology play an important role in the overall performance of the speech enhancement algorithm in terms of noise reduction as well as input-output delay. With this in mind we introduce the concept of eigenvector centrality with a weighted adjacency matrix that can be used to select a root node, as well as to prune an ad-hoc network to a specific tree topology that yields good speech enhancement performance when applying the T-DANSE algorithm.