Targeted Adversarial Examples for Black Box Audio Systems

Sample code to let you create your own adversarial examples! Paper linked here


pip3 install -r requirements.txt
git clone
tar -xzf deepspeech-0.1.0-models.tar.gz
rm deepspeech-0.1.0-models.tar.gz

Now create run an attack with the format:

python3 input_file.wav "target phrase"

For example,

python3 sample_input.wav "hello world"

You can also listen to pre-created audio samples in the samples directory. Each original/adversarial pair is denoted by a leading number, with model transcriptions as the title.