Reproduce “GDBack: A Group-Delay Backdoor Attack on Speaker Identification via Analytic All-Pass Filtering” 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
Bicheng Xiong, Tao Li, Chenhui Cui, Zedong Xing, Qien Li, Weiping Wen, Rubing Huang, Tsong Yueh Chen
GDBack tests whether a speaker-identification backdoor can use a shared transformation while preserving utterance-specific output phases. It applies the same all-pass filter to poisoned recordings and assigns them to a target speaker. Across six dataset-model settings at 5% poisoning, mean attack success rate was 97.00%, with benign accuracy within 1.54 percentage points of clean models; over-the-air results varied by model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.