over-the-air_speech_recogniztion_attack


Introduction

The course final project for UCLA EE209AS Winter 2018 - Special Topics in Circuits and Embedded Systems: Security and Privacy for Embedded Systems, Cyber-Physical Systems, and the Internet of Things by Professor. Mani Srivastava. In this project, we use deep learning (audio U-Net) to build model remove electronic noise and air noise during adversarial example transmission over the air. The contribution of this project are:

  • make the adversarial example attack can transmit over-the-air, which eventually be a practical attack.
  • found audio U-Net is also a possible defense for adversarial example attack due to strong ability to remove noise.

For more problem details, please go to my personal website.

Requirements and Dependencies

Recommended use Anaconda to create the individual environment for this project and use following code to install dependencies:

conda install -c conda-forge tensorflow 
conda install -c conda-forge tqdm
conda install -c conda-forge librosa

The following packages are required (the version numbers that have been tested are given for reference):

  • Python 2.7 or 3.6
  • Tensorflow 1.0.1
  • Numpy 1.12.1
  • Librosa 0.5.0
  • tqdm 4.11.2
  • SoX 14.4.2 (for Mac: brew install sox, for Ubuntu: apt-get install sox)
  • pysox 1.3.0 (pip install sox)