Deep Neural Networks for Speech Separation With Application to Robust Speech Recognition