Fast Subnetwork Selection for Speech Enhancement in Wireless Acoustic Sensor Networks

Instead of using the entire wireless acoustic sensor network (WASN), selecting the most informative subnetwork can achieve a good speech enhancement performance while keeping a low resource cost. In this paper, a fast subnetwork selection method is derived for the delay-and-sum beamformer, which is also suitable for other existing speech enhancement algorithms. Inspired by the energy efficiency in wireless communication theory, the best subset of WASN is determined by maximizing a novel cost function that simultaneously considers the output signal-to-noise ratio (SNR) and the energy consumption. Then, a Dinkelbach-based solver is presented to obtain the optimal solution rapidly. Experimental results show that the proposed method achieves a better trade-off between performance and energy usage.