Perceptual speech enhancement system based on non-uniform analysis

This paper addresses the problem of single speech enhancement in adverse environment. We propose a new speech enhancement system based on perceptual properties of human auditory ear using non-uniform and multi-band analysis. The noisy signal is divided into a number of sub-bands using a gammatone filter bank with non-linear Equivalent Rectangular Bandwidth (ERB) resolution, this sub-bands are individually estimated according the Log spectral magnitude minimum mean spectral error (MMSE) estimator. Evaluation tests demonstrate significant improvements results with the proposed method over classical speech enhancement algorithm, when tested with speech signals corrupted a posterior by various noises at different noise signal ratios.