跳到正文
arXiv cs.AI· Yu-Xiang Lin, Yun-Man Hsu, Bo-Ru Lu, Kuan-Po Huang, Wenze Ren, Ming-To Chuang, Chun-Yi Kuan, Jiawei Du, Byeonggeun Kim, Qingming Tang, Chieh-Chi Kao, Ankur Gandhe, Chao Wang, Hung-yi Lee·· 4 小时前AI 评分34

CoTTAGE:面向文生音频的可解释自动评估器

CoTTAGE: An Explainable Judge for Text-to-Audio Generation

AI 导读

台澳等地研究者提出 CoTTAGE,一个针对文生音频(TTA)的推理式自动评估器,评估文本相关性(REL)与音频质量(OVL)。它遵循“先给理由再打分”范式,基于超过 10K 条新采集的人类 OVL 标注,先监督微调、再用基于 Thurstone 的偏好目标做强化学习。在覆盖 REL 和 OVL 的六个基准上,CoTTAGE 与人类判断的相关性达到 SOTA,并生成人类偏好的解释。

正文 · 原文

Authors:Yu-Xiang Lin, Yun-Man Hsu, Bo-Ru Lu, Kuan-Po Huang, Wenze Ren, Ming-To Chuang, Chun-Yi Kuan, Jiawei Du, Byeonggeun Kim, Qingming Tang, Chieh-Chi Kao, Ankur Gandhe, Chao Wang, Hung-yi Lee

View PDF HTML (experimental)

Abstract:Rapid progress in text-to-audio (TTA) generation demands reliable and explainable automatic evaluation. However, existing metrics fail to simultaneously deliver high correlation with human judgments and meaningful, audio-grounded explanations, forcing reliance on labor-intensive human evaluation that bottlenecks model development iteration. To address this, we propose CoTTAGE, a reasoning-based evaluator for text relevance (REL) and audio quality (OVL). Following a reason-then-predict paradigm, CoTTAGE first generates an audio-grounded rationale to diagnose errors before predicting scores. Supported by over 10K newly collected human OVL annotations, CoTTAGE is first trained via supervised fine-tuning and then refined with reinforcement learning using a Thurstone-based preference objective that converts sampled score distributions into pairwise preference probabilities and rewards agreement with rankings derived from human judgments. Across six benchmarks spanning REL and OVL, CoTTAGE achieves state-of-the-art correlation with human judgments while producing human-preferred explanations to accelerate TTA development.
Subjects: Sound (cs.SD); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.17644 [cs.SD]
  (or arXiv:2611.17644v1 [cs.SD] for this version)
  https://doi.org/10.48550/arXiv.2611.17644

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Yu-Xiang Lin [view email]
[v1] Fri, 13 Nov 2026 01:28:45 UTC (2,905 KB)

来源:arXiv cs.AI · arxiv.org