Reproduce “What Survives the Codec Shift: Pooled No-Vocals Residuals for Speech 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
Jiajun Xu, Menglu Li, Xiao-Ping Zhang
MN-P combines speech features with a pooled residual derived after vocal separation to detect synthesized speech. It uses XLS-R features and a gated residual view. In screening, XLS-R had the best overall EER (4.843%), while no-vocals features led on held-out DAC F07 (8.619%). MN-P-AASIST reached 2.744% overall and 6.136% on F07; the residual’s useful source remains unidentified.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.