repr-seq2seq-audio Representations in seq2seq based Automatic Speech Recognition Models. Download the TIMIT dataset and run, python preprocess.py