/defense_FakeBob

Source code for the paper "Defending Against FakeBob Adversarial Attacks in Speaker Verification Systems With Noise-Adding"

Primary LanguageJupyter Notebook

defense_FakeBob

Source code for the paper "Defending Against FakeBob Adversarial Attacks in Speaker Verification Systems With Noise-Adding"

Main content

Jupyter notebook code used to gnerate Figures 3 and 5 in the paper. Moreover, the source code of denoising and noise-adding methods is provided.

File structure

  • ./Demo.ipuynb: Jupyter notebook code used to gnerate Figures 3 and 5 in the paper. Contain the code that shows an audio in time domain and mel spectrogram.
  • ./noise_adding.py: python code for the noise-adding method.
  • ./denoising.py: python code for the denoising method. It calls the method in "./noisereducev1.py" file.
  • ./noisereducev1.py: This file is downloaded and modified from Noise reduction in python using spectral gating by Tim Sainburg.
  • ./237-126133-0017-att-org.wav: Original clean audio that was used to generate an adversarial audio.
  • ./237-126133-0017-att-adv.wav: Adversarial audio.
  • ./237-126133-0017-att-denoised.wav: Denoised adversarial audio.
  • ./237-126133-0017-att-noise-added.wav: Noise-added adversarial audio.

Before running "Demo.ipynb", please make sure to create a "./figs/" folder that will contain generated figures.

If you have any questions, please feel free to comment or contact.