A high performance low complexity noise suppression algorithm

A high performance low complexity algorithm for noise suppression in hearing aids, using spatial information for estimating the required noise and speech power spectral densities (PSDs), is proposed. The main assumption of the scheme is that the target is directly in front of the hearing aid user while the noise comes from the back hemisphere. Furthermore, the goal is to develop a scheme that can be implemented in real time on a comercially available hearing aid. With the proposed approach, no statistical models are needed to estimate the speech and noise PSDs, which results in a robust and high performing noise suppression scheme. In a first step, the noise suppression filter is implemented in the FFT domain using a weighted overlap add scheme (WOLA), as is common for digital hearing aids. In a second step, to further reduce the computational complexity, a carefully selected wavelet decomposition is used instead of a WOLA. Hearing tests as well as objective performance measures show the excellent performance of the low complexity algorithm in the FFT as well as in the wavelet domain.