Collaborative Viseme Subword and End-to-End Modeling for Word-Level Lip Reading | IEEE Journals & Magazine | IEEE Xplore