Reproduce “Omni-JEV: Text and Dual-Stream Speech Decisions” 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
Shengzhou Kong
Omni-JEV frames response control as choosing an action after a collected user segment, conditioned on assistant state and history. It assigns probabilities to five actions from role-indexed audio or text. In matched single-seed continuation runs, audio accuracy is 84.08% versus 84.04%, while probability scores trade off. The fixed weak-label test does not establish reliable live control.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.