On Training the Recurrent Neural Network Encoder-Decoder for Large Vocabulary End-To-End Speech Recognition

doi 10.1109/icassp.2016.7472641