Correlational Networks for Speaker Normalization in Automatic Speech Recognition