Reproduce “VocalGrad: Evaluating Acoustic Perception in Audio Language Models” 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

Koki Ryu, Hitomi Yanaka

VocalGrad tests whether audio language models can judge the direction of change in one acoustic feature while speech content is preserved. Across nine controlled categories, models answer whether a feature increases or decreases over time. Human annotators averaged 94.4% accuracy, versus 51.8% for the best model. Probes show category-dependent decodability; results concern controlled transformations, not natural variation.

Summary by alphaXiv.

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