Robust Bayesian and Maximum a Posteriori Beamforming for Hearing Assistive Devices

Abstract

Multi-microphone speech enhancement systems often apply beamforming to enhance one or multiple desired signals in a noisy environment. Common for many beamforming methods, is that they require the direction-of-arrival (DOA) of the target sound source to be known in order to achieve optimal noise reduction performance. To improve robustness against DOA uncertainty, we propose maximum a posteriori (MAP) and Bayesian beamformers that are able to take advantage of prior information on the target direction. We compare the proposed MAP and Bayesian beamformers to state-of-the-art beamforming methods for noise reduction in hearing assistive devices. We evaluate the proposed beamformers in isotropic babble noise in terms of segmental SNR (SSNR) and extended short-time objective intelligibility (ESTOI). Results show that the proposed methods outperform current state-of-the-art beamformers used for noise reduction in hearing aids in most scenarios.