Investigating the Resilience of Audio Deepfake Detectors under Imperceptible Adversarial Attack

Hong-Hanh Nguyen-Le (ID: 23203495)

University College Dublin

Course: COMP47700 - Speech And Audio

Dataset

Wavefake ASVspoof 2019 ASVspoof 2021

Train deepfake detection

1. Change the dataset path in train_models.py
2. Change the config path in train_models.py which corresponds to the model that need to train in config folder
3. Run below: 'python train_models.py'

Evaluate deepfake detection

1. Change the dataset path in evaluate_models.py
2. Change the config path in evaluate_models.py which corresponds to the model that need to train in config folder
3. Run below: 'python evaluate_models.py'

Evaluate the adversarial examples

Run the jupytor file evaluate_adv with adversarial samples