Reproduce “CRESiST: Self-Enhancing Explorative Policy Learning for Simultaneous Speech Translation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chenyang Le, Pei Zhang, Yuxiang Chen, Bing Han, Derek F. Wong, Baosong Yang, Yanmin Qian
CRESiST performs simultaneous speech translation by interleaving speech chunks and translated text, then learning when to read and write without external forced alignments. Its architecture supports KV-cache reuse and unsegmented inference. On ACL 60/60 English-to-German, it scores 85.59 XCOMET-XXL and 1.08 LongYAAL, versus InfiniSST’s 63.52 and 1.46; these benchmark results do not establish runtime gains or limitless operation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.