Efficient Implementation of Single-Channel Noise Reduction for Hearing Aids Using a Cascaded Filter-Bank

In this paper we present an efficient method to increase the frequency resolution for speech enhancement algorithms in hearing aids. Since the analysis-synthesis filter-bank applied in digital hearing aids needs to deliver a high stop band attenuation to enable large frequency dependent amplification gains and a low overall delay, speech enhancement often suffers from the resulting low frequency resolution. This leads to residual noise artifacts in the processed speech, since the noise between the harmonics can not be removed. To satisfy the constraints of both high stop-band attenuation and high frequency resolution, we propose a cascaded, two stage filter-bank system to increase the frequency resolution and improve noise reduction performance. Furthermore, we make use of an efficient and low delay implementation of the second filterbank stage to satisfy strict design constraints for hearing aids and achieve a complexity reduction by exploiting redundancies between sub-bands.