Audio-deepfake detection: Adversarial attacks and countermeasures

Authors:
Mouna Rabhi, Spiridon Bakiras, Roberto Di Pietro

Where published:
ELSEVIER (Expert Systems with Applications)

 

Dataset names (used for):

  • H-VOICE: used to train a machine learning system to classify original and fake voice recordings obtained with the imitation and Deep Voice algorithms

 

Some description of the approach:
The study explores vulnerabilities of audio deepfake detection systems to adversarial attacks, demonstrating that current methods like Deep4SNet can be manipulated to nearly 0% detection accuracy using GAN-based attacks. A new, generalizable defense mechanism is proposed to enhance system resilience.

 

Some description of the data:
The study uses a dataset of 6,672 histograms from both original and fake audio samples for training their detection models.

 

Keywords:
Authentication, Adversarial attacks, Audio deepfake, Fake voice detection, GAN, Biometrics, Security

Instance Represent:
Audio histograms

Dataset Characteristics:
6,672 histograms from both original and fake audio samples

Subject Area:
Security of audio authentication systems

Associated Tools:
Deepfake audio detection

Feature Type:
Histograms

Main Paper Link

Code Link


License: This is an open access article distributed under the terms of the Creative Commons CC-BY license, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.


Last Accessed: 6/15/2024

NSF Award #2346473