Backend
CoTTAGE: An Explainable Judge for Text-to-Audio Generation
arXiv cs.SD0 views
arXiv:2611.17644v1 Announce Type: new 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.
Source:https://arxiv.org/abs/2611.17644PreviousLazyWeights: Decode-on-Demand Edge LLMsNextAny Channel Count, One Token Stream: Incorporating Spatial Audio Understanding and Reasoning into Large Audio-Language Models
Related posts
Clustering App Ad Hooks in Go: TF-IDF, Cosine Similarity and a Hand-Labelled Sample, and the Walls You'll Hit (2026)
0dev.to
Building a Simple REST API with FastAPI: A Step-by-Step Guide for Beginners
0dev.to
What It Takes to Trust an AI Agent With Real Work
0dev.to
How to check that a PDF converter does not upload your file anywhere
0dev.to
One TPU v6e chip holds 927 agent sessions with host-held state and 5 with a kept chat (memory only)
0dev.to
How to redesign a website without losing Google rankings
0dev.to UX