https://arxiv.org/pdf/1710.09617.pdfhttps://www.isca-speech.org/archive_v0/Interspeech_2016/pdfs/0753.PDF
CTC에서 존재하던 Conditional independence assumption을 해소Streaming ASR Application에 사용 가능Modeling $ P(yt | x{1:t}, y\_{1:u-1}) $Predicting the current token