On the Defense of Spoofing Countermeasures against Adversarial Attacks

Advances in speech synthesis have exposed the vulnerability of spoofing countermeasure (CM) systems. Adversarial attacks exacerbate this problem, mainly due to reliance most CM models on deep neural networks. While research adversarial anti-spoofing systems has received considerable attention, there is a relative scarcity studies focused developing effective defense techniques. In study, we propose strategy against such by augmenting training data with frequency band-pass filtering and denoising. Our approach aims limit impact perturbation, thereby reducing susceptibility samples. Furthermore, our findings reveal that use Max-Feature-Map (MFM) provides additional benefits suppressing different noise types. To empirically validate hypothesis, conduct tests using samples derived from ASVspoof challenge other well-known datasets. The evaluation results show mechanisms can potentially enhance performance