Binaural voice activity detection for MWF-based noise reduction in binaural hearing aids

The Speech Distortion Weighted Multichannel Wiener Filter (SDW-MWF) is a powerful multimicrophone noise reduction technique, especially for binaural hearing aids where two devices are connected by a wireless link. As is the case for other single- and multi-microphone techniques, the SDW-MWF relies on a voice activity detection (VAD) algorithm, which classifies frames as noise-only or speech+noise frames. In this paper, two novel binaural fusion VADs are proposed, which are extensions of a fusion VAD originally proposed for a wireless sensor network application. By making use of the exchange of information over the wireless link of the binaural hearing aid, the binaural VADs perform a decision fusion of energy VADs calculated in the left and right device and, if the available bandwidth also allows for transmitting audio signals, a cross-correlation based VAD. The superior performances of the proposed binaural VADs are assessed (in terms of receiver operating characteristics and also by evaluating the impact on the SDW-MWF performance) by experiments with a binaural hearing aid setup.