Speech Enhancement with LSTM Recurrent Neural Networks and its Application to Noise-Robust ASR