This is a project about the transfer attack on ASV spoofing systems. It is based on the code by https://github.com/nii-yamagishilab/project-NN-Pytorch-scripts, https://github.com/asvspoof-challenge, https://github.com/ghuawhu/end-to-end-synthetic-speech-detection, and https://github.com/Harry24k/adversarial-attacks-pytorch.

useage:

Training

To train the model, use

python main.py --config ./config/your_model.conf

to evaluate add --eval to evaluate the model itself.

Attacks

To do adversarial attacks and save the adversarial examples, use

python main.py --config ./config/you_attack.conf

Note that you should specify the blackbox and whitebox models, and the attack method in the config file

Add --eval to evaluate the adversarial examples, add --attack_eval to evaluate during attack without saving the examples.