Reproduce “ReplayFactor: Replay-Aware Dual-Attribute Learning for Practical 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
Miao He, Jiayi Zhou, Yu Zheng, Zhongjie Ba, Peng Cheng, Yuxi Zhou, Junjie Zhu, Changhao Zhang and 1 more
ReplayFactor separates generated-speech detection from replay detection by learning distinct source and replay representations, adding replay-sensitive acoustic features to the replay branch, and using gated fusion. On EchoFake’s open-set test, it improves macro-F1 from 62.27% to 76.44% and reduces equal error rate from 7.26% to 3.05% versus a same-backbone four-class baseline; closed-set performance is lower than the XLSR-Linear baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.