In paper 'Attention-Passing Models for Robust and Data-Efficient End-to-End Speech Translation',  all the models are based from architrecture proposed by another paper that you've read. Provide the full name of that paper.