search

RaphaelOlivier / robust_speech

Apache License 2.0
38 stars 14 forks source link

I want to test using transformer model with cw algorithm. #9

Closed kyoungmingo closed 1 year ago

kyoungmingo commented 1 year ago

Hello~

I want to test using transformer model with cw algorithm.

When I used "python fit_attacker.py attack_configs/LibriSpeech/universal/ctc_5000bpe_fit.yaml", I met this error image If possible, can you tell me how to use the CW algorithm attack method using the transformer model in this git,robust_speech?

RaphaelOlivier commented 1 year ago

I cannot see the image, so I'm not certain. However, an error would occur if you try to use fit_attacker.py with a CW config. fit_attacker.py is only required for specific attacks that need training, like Universal. Just use evaluate.py directly with one of the /cw/ configs. If you want to compute the adversarial examples once and save them, it's not fit_attacker.py you should use but the save_audio parameter in the attack config.

RaphaelOlivier commented 1 year ago

Presumably fixed by answerabove