Reproduce “Measuring the Robustness of Audio Deepfake Detection under Real-World Corruption” 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
Xiang Li, Pin-Yu Chen, Wenqi Wei
The authors introduce AudioPerturber to test how 18 audio corruptions affect 10 deepfake detectors. On Wavefake, synthetic speech from six vocoders, many detectors are more vulnerable to modifications and compression than noise, especially neural codecs. In one Wave2Vec2 echo test, accuracy falls from 0.9874 to 0.558 while ViSQOL stays above 4; findings are limited to this dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.