Reproduce “From Spectrograms to Shallow Speech SSL: Audio-Frontend Design for Generator-Disjoint and Low-Latency Audio--Visual Deepfake Detection” 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

Qiwu Wen, Jahangir Alam, Eric Charton

The authors test whether learned weighting of six frozen wav2vec 2.0 layers improves a fixed audio–visual deepfake detector. Macro AUROC averages AUROC across unseen generators. Weighted features reach 90.61% versus 88.08% for sixth-layer-only features across three runs, while six-layer execution is faster than twelve-layer execution. Gains do not extend to Hunyuan, and latency excludes media decoding.

Summary by alphaXiv.

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