The rise in demand of voice biometric systems also increases the threat from various kinds of spoofing attacks from unauthorized users. The latest ASVspoof 2021 challenge devotes to promote such research with very recent kinds of attacks based on logical access, physical access and speech deepfakes. In this work, we consider a few data augmentation methods to have a robust spoofing countermeasure based on the known information from the challenge evaluation protocol and with some unknown approaches that can be useful for each of the three tracks. The data augmentation strategies are carried over considering constant-Q transform based log power spectrum features with a light convolutional neural network framework. The studies show that our system benefits extensively with various known and unknown data augmentation methods and the result for each track outperforms the challenge baselines on the progress phase. We also carried out score level fusion of our systems with the challenge baselines that contributed to further enhance the performance to become one of the top performing systems on evaluation phase of ASVspoof 2021 challenge.
Cite as: Das, R.K. (2021) Known-unknown Data Augmentation Strategies for Detection of Logical Access, Physical Access and Speech Deepfake Attacks: ASVspoof 2021. Proc. 2021 Edition of the Automatic Speaker Verification and Spoofing Countermeasures Challenge, 29-36, doi: 10.21437/ASVSPOOF.2021-5
@inproceedings{das21_asvspoof,
author={Rohan Kumar Das},
title={{Known-unknown Data Augmentation Strategies for Detection of Logical Access, Physical Access and Speech Deepfake Attacks: ASVspoof 2021}},
year=2021,
booktitle={Proc. 2021 Edition of the Automatic Speaker Verification and Spoofing Countermeasures Challenge},
pages={29--36},
doi={10.21437/ASVSPOOF.2021-5}
}