Reproduce “Text Dominance in Paralinguistic Conflict: Tracing Suppression Through Audio LLM Circuits” 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
Ankur Banerjee, Vishnu Guntakala, Danity Pike, Advay Roongta, Kevin Yang, Ruizhe Li
The authors test whether audio-language models use vocal cues when transcripts conflict with speech. In six-model benchmark trials, conflicting text dominated outputs at a mean rate of 0.78, or 0.86 excluding sarcasm and pacing. In Qwen2-Audio-7B-Instruct, speaker gender was decodable at 1.000 accuracy. Blocking claim influence at intermediate layers reduced false-claim outputs, but evidence covers one model and two categories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.