GitHub - sab747/adversarial_audio: Assessing differential vulnerability of original Adversarial Audio attack on different sexes + accents (English)

Adversarial Speech Commands

Fork of original work by github user nesl being used to validate and expand upon results.

Python 3.9 (3.9.7 used during our recent work) Tensorflow 2.6 (2.6.0 used during our recent work, original implementation version undocumented)

Download datset

bash download_dataset.sh

Download model:

bash download_checkpoint.sh

To run:

./run_attack.sh dataset_dir ckpts_dir limit max_iters test_size

Name		Name	Last commit message	Last commit date
Latest commit History 43 Commits
speech_commands		speech_commands
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
_config.yml		_config.yml
audio_attack.py		audio_attack.py
count_files.sh		count_files.sh
download_checkpoint.sh		download_checkpoint.sh
download_dataset.sh		download_dataset.sh
evaluate_attack.py		evaluate_attack.py
evaluate_attack.sh		evaluate_attack.sh
evaluate_realdata_accuracy.py		evaluate_realdata_accuracy.py
extract_times.sh		extract_times.sh
machine_deception2017.ipynb		machine_deception2017.ipynb
run_attack.sh		run_attack.sh
select_examples.py		select_examples.py
time_log.txt		time_log.txt
time_result.txt		time_result.txt
times_all.txt		times_all.txt