ASR attack

[QingYun65]

Hello, thank you very much for your open source work! I am trying to reproduce SMACK to achieve attacks on ASR systems. For the given example, when I run the command python3 attack.py --audio "./Original_TheyDenyTheyLied.wav" --model iflytekASR --content "They deny they lied" --target "They did not have a light", I am able to obtain the corresponding adversarial sample. However, now I would like to attack using my own dataset, which consists of some audio files from the Librispeech dataset, targeting iFlytek ASR.

I used the following command: python3 attack.py --audio "./librispeech_test_clean_300_audio/121-127105-0033.flac" --model iflytekASR --content "IT WAS THE BEAUTY OF IT" --target "turn off the light", but no matter how I modify the number of iterations or change the original audio and target command, I am unable to obtain a successful adversarial sample. Could you give me some advice?