Perceptual Weighting Deep Neural Networks for Single-channel Speech Enhancement

Improving the perceptual quality of speech signals is a key yet challenging problem for many real world applications. In this paper, we propose a perceptually motivated approach based on deep neural networks (DNNs) for speech enhancement task. The proposed approach take into consider the masking properties of the human auditory system and reduces the perceptual effect of the residual noise. Given the good performance of deep learning in signal representation, a DNN is employed for accurately modeling the clean speech spectrum. In the DNN training stage, perceptual weighting matrix is used to adjust the weight of the error when using back propagation algorithm transfer the error from DNN output layer to the front layer. Evaluations on various real-world background noises show that the proposed method performs better than different competitive methods.