A postfiltering approach for dual-microphone smartphones

Although beamforming is a powerful tool for microphone array speech enhancement, its performance with small arrays, such as the case of a dual-microphone smartphone, is quite limited. The goal of this paper is to study different postfiltering approaches that allow for further noise reduction. These postfilters are applied to our previously proposed extended Kalman filter framework for relative transfer function estimation in the context of minimum variance distortionless response beamforming. We study two different postfilters based on Wiener filtering and non-linear estimation of the speech amplitude. We also propose several estimators of the clean speech power spectral density which exploit the speaker position with respect to the device. The proposals are evaluated when applying speech enhancement on a dual-microphone smartphone in different noisy acoustic environments, in terms of both perceptual quality and speech intelligibility. Experimental results show that our proposals achieve further noise reduction in comparison with other related approaches from the literature.