Reduction of Computational Cost Using Two-Stage Deep Neural Network for Training for Denoising and Sound Source Identification.