Deep Learning: Using TensorFlow LSTM seq2seq Neural Nets to build Pronunciation System - Tech Talk