Reproduce “REAL: Reward Modeling through Evidence-Linked Analysis of Lifelikeness in Generated Speech” 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

Yaodong Song, Jie Lian, Xusheng Yang, Genliang Zhao, Hongjie Chen, Jian Kang, Zhen-Hua Ling, Jie Li and 1 more

REAL-Judge evaluates which of two speech readings better matches human preference, linking its judgment to observations at specific audio times. Adapted from MiMo-Audio-Instruct, it reaches 84.43% accuracy on REAL-Eval-Standard and 65.36% on REAL-Eval-Fine, exceeding its backbone on both. Fine tests disjoint reference texts, speech, and source systems, but only Chinese; accuracy does not verify explanation faithfulness.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.