AnyAudio-Judge benchmark for audio instruction following

Arxiv CS.SD (Sound/Audio ML)

The paper introduces a dynamic rubric-based benchmark and evaluator for audio instruction following, addressing limitations of holistic scoring from general-purpose LLMs. It provides fine-grained, interpretable evaluation metrics.

benchmarks

Arxiv CS.SD (Sound/Audio ML)·Haitao LiTian TanYuguang YangShan YangXie Chen

8 days ago