Noise suppression based on loudness management for hearing aid

Abstract

It is well known that speech intelligibility does not improve even if SNR is improved and noise annoyance decreases by noise suppression for hearing aids, since the noise suppression decreases the level of the desired signal as well as the one of the noise. One solution for the problem is to minimize the speech distortion as much as possible even if a large quantity of residual noise exists. In this study, we propose a noise suppression algorithm based on loudness management to improve speech intelligibility for hearing aid wearer. Our algorithm not only minimizes the speech distortion but also maintains the speech loudness. In the algorithm, the speech spectrum is estimated from the observed signal first, and then the loudness of the estimated speech spectrum is calculated. Simultaneously, the loudness of the observed signal is also calculated and the gains in each frequency band are controlled in order to correspond between these two loudness. In our loudness management, even if the bands are almost filled by noise component, it is not suppressed as much as possible not to reduce barely audible speech component. As results, the noise suppressed signal with low speech distortion is obtained and speech intelligibility is improved comparing with commonly used method.