Chen, et al.. Aipnet: Generative Adversarial Pre-training of Accent-invariant Networks for End-to-end Speech Recognition. 27 Nov. 2019.