Reproduce “CoTTAGE: An Explainable Judge for Text-to-Audio Generation” 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

Yu-Xiang Lin, Yun-Man Hsu, Bo-Ru Lu, Kuan-Po Huang, Wenze Ren, Ming-To Chuang, Chun-Yi Kuan, Jiawei Du and 6 more

CoTTAGE is an automatic judge for text-to-audio generation that explains audio evidence before scoring text alignment (REL) and overall audio quality (OVL). Initialized from Qwen2.5-Omni-7B, it is trained with supervised fine-tuning and preference-based reinforcement learning. On XACLE REL, its Pearson correlation with human ratings is 0.685, versus 0.551 for Gemini-3.1 Pro; this evidence does not establish performance beyond tested short-clip settings.

Summary by alphaXiv.

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