Authors: zhujinyi(from Peking University)

Description: use CRNN to extract feature, deploy an attentional seq-to-seq model as the decoder
train da