.png){kind=icon}
返回
Multi-Scale Self-Supervised Learning for Efficient Audio Deepfake Detection
.png){kind=icon}
PRE
AI
DOI：10.1109/LSP.2025.3634032.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
神经声码器能够生成高度逼真的合成语音，对多媒体认证构成挑战；然而，现有的检测方法在应对未见过的合成方法时鲁棒性有限，且部署准备不足。我们提出MASD（多尺度人工痕迹感知自监督深度伪造检测器），结合多尺度SSL与手工特征。MASD将频谱图分解为三个频带，通过使用掩码重建、对比预测编码和对抗声码器分类进行预训练的编码器进行处理。特征通过交叉注意力与相位一致性、频谱通量和高频能量融合，并由温度缩放的SVM进行分类。在ASVspoof 2019 LA上的评估展示了最先进的表现。消融研究证实对抗增强是鲁棒性的主要驱动因素，将EER从1.52%提升至0.39%，而零样本跨数据集评估验证了泛化有效性，确立了MASD为实用解决方案。
Keyword:
Audio deepfake detection
adversarial augmen- tation
confidence calibration
self-supervised learning
期刊
I
IF:
3.9
论文数:
630
被引数:
0
机构
引用论文
Multi-Scale Permutation Entropy for Audio Deepfake Detection多尺度排列熵用于音频深度伪造检测.png)
ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
ICASSP 2024 - 2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)IF0
PREAI
分享收藏End-to-End anti-spoofing with RawNet2ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
OAAI
ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)IF0
OAAI分享收藏AASIST: Audio Anti-Spoofing Using Integrated Spectro-Temporal Graph Attention NetworksICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
OAAI
ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)IF0
OAAI分享收藏
查看更多内容