1 paper
Zixiang Wan, Delin Chen, Wei Shi +3
Revision-capable streaming speech-to-text translation (S2TT) can correct earlier drafts, but process rewards based on visible text may credit content later withdrawn. Persistent De…