Reproduce “Qwen-Audio-Jev: A Unified Interface for Audio Decisions” 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
Peixin Chen, Zhigao Chen, Zhiyun Fan, Ying Fang, Zhifu Gao, Qing Gu, Jiajun He, Kangqi Jing and 14 more
Qwen-Audio-Jev replaces generated answers with typed decisions for bounded audio questions. It reuses an audio-language model’s question-conditioned representation and applies separate readouts for category choices, yes-or-no propositions, and ordinal ratings. Across 18 dataset–task configurations, it reports 83.1366% accuracy on MMSU, 93.7595% on VoxCeleb1-O, and 0.123478 normalized MAE on NISQA overall. These are benchmark results, not comparisons establishing gains or measured speedups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.