Automatic Speaker Verification Spoofing and Deepfake Detection Using Wav2vec 2.0 and Data Augmentation
Hemlata Tak, Massimiliano Todisco, Xin Wang, Jee-weon Jung, Junichi Yamagishi, Nicholas Evans
Abstract
The performance of spoofing countermeasure systems depends fundamentally upon\nthe use of sufficiently representative training data. With this usually being\nlimited, current solutions typically lack generalisation to attacks encountered\nin the wild. Strategies to improve reliability in the face of uncontrolled,\nunpredictable attacks are hence needed. We report in this paper our efforts to\nuse self-supervised learning in the form of a wav2vec 2.0 front-end with fine\ntuning. Despite initial base representations being learned using only bona fide\ndata and no spoofed data, we obtain the lowest equal error rates reported in\nthe literature for both the ASVspoof 2021 Logical Access and Deepfake\ndatabases. When combined with data augmentation,these results correspond to an\nimprovement of almost 90% relative to our baseline system.\n