A novel speech enhancement using forward-backward minima-controlled recursive averaging

Nonstationary noises may seriously damage speech intelligibility and reduce the performance of hearing aids. This paper proposes a novel speech enhancement approach, named forward-backward minima-controlled recursive averaging, that helps to solve this problem. A forward-backward estimation is developed to correctively track nonstationary noises, especially the beginnings of nonstationary noises. To suppress these nonstationary noises, a log-spectral amplitude estimator was selected as the gain function for our approach and used to effectively estimate the power spectrum of the noise. Moreover, harmonic regeneration was integrated to reduce the overestimation of noises generated by the forward-backward estimation. The experimental results demonstrated that our approach is able to improve performance compared with previous approaches using both objective and subjective measures.