When Evil Calls: Targeted Adversarial Voice over IP Network

Dependencies

• torch
• numpy
• scipy
• wave
• pyaudio
• google.cloud

To install all the dependencies:

pip install -r requirements.txt

Psychoacoustic

To determine the most proper source audio and the location to synthesize:

python sel_src.py --cmd_sound [cmd_audio_path] --src_sound_set [source_audio_path]

Run attack

To run TAINT attack:

python tx.py --env_sound [env_audio_path] --tar_sound [cmd_audio_path] --tar_cmd [target command] --output_path [output audio path]  --align_pos [aligned position] --ip [ip] --port [port number]

python3 rx.py --credential_path [google_credential_file] --record_time [time_length] --ip [ip]  --port [port number] 

Project website

https://sites.google.com/view/targeted-adversarial-voip

Citation

If you find our work useful, please cite:

@InProceedings{liu_evilcalls_2022,
      title = {When Evil Calls: Targeted Adversarial Voice over IP Network},
      author = {Liu, Han and Yu, Zhiyuan and Zha, Mingming and Wang, XiaoFeng and Yeoh, William and Vorobeychik, Yevgeniy and Zhang, Ning},
      booktitle = {Proceedings of the 2022 ACM SIGSAC conference on computer and communications security},
      month = {November},
      year = {2022}
}