Paper Title
Context-Awareideal Ratio Mask Estimation for Speech Denoising using Deep Neural Networks

In this paper, we propose a context-aware approachfor speech denoising based on deep learning. In most studies, single deep neural network(DNN) is trained in various noisy environments. However, large variations of noisy environments can cause performance degradation. We develop environmentally-optimized models trained in each noisy environment, which is effective in restoring clean speech spectra under different noisy conditions.In this research work, we first classify noisy spectra through DNN models trained for classifying noise environment before noisy speech spectra are processed. We apply the proposed method to the estimation of ideal ratio mask (IRM) that is used for denoising in spectral domain. Experimental results showed that the proposed context-aware method performs better than that using non-context-awareness model. Keywords - Speech denoising, Context-awareness,Deep learning, Deep neural network,Ideal ratio mask