/shallowfake

Exploring adversarial attacks against Speech Recognition and Deepfakes to protect privacy

Primary LanguagePython

This is a work-in-progress compilation of research efforts for ShallowFake: Adversarial Attacks on Speech Recognition and Audio Deepfakes. As of now, the code contains the following

  • Environment setup, training scripts, example output for PGD (projected gradient descent) attack on Whisper speech recognition
    • environment includes robust_speech and speechbrain submodules that are individually updated but path and package dependencies aren't
  • Sample script (sample.py) for NN preprocessing, training, and postprocessing
  • Initial project report in report.pdf