The paper presents a new dataset, the Large Replay Parallel Dataset (LRPD), designed to improve the generalization ability of deep neural networks (DNNs) in detecting replay attacks in voice anti-spoofing. The LRPD, which contains over 1 million utterances collected by 19 devices in 17 environments, outperforms traditional methods like GMM in presentation attack detection. The research also provides an example training pipeline in PyTorch and a baseline system. The LRPD dataset is freely available for research purposes.
Publication date: 4 Oct 2023
Project Page: https://ieeexplore.ieee.org/document/9746527
Paper: https://arxiv.org/pdf/2309.17298
