Closed TTTJJJWWW closed 5 years ago
@TTTJJJWWW Hi, Yes we use the MFC preprocessing step (any input to the DeepSpeech model needs to be preprocessed as such). However, since our method is black box, there is no need for the reverse MFC.
@rtaori Hi, I found that "models/session_dump" was used in "run_audio_attack.py", why the project need use it? As a black-box-attack, why not use the the original model (from deepspeech) instead? Is the disturbance(or mutate) added before or after the MFCC process?
We used it simply for ease of use. It was much easier to use the TF session, especially at the rate of samples we were passing through the model. The disturbances is added before the MFCC process.
@rtaori @TTTJJJWWW While running run_audio_attack.py file, saver.restore(sess, "models/session_dump") this line giving me an error of models; No such file or directory
saver.restore(sess, "models/session_dump") File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 1536, in restore if not checkpoint_management.checkpoint_exists(compat.as_text(save_path)): File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/checkpoint_management.py", line 364, in checkpoint_exists if file_io.get_matching_files(pathname): File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/lib/io/file_io.py", line 342, in get_matching_files for single_filename in filename File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/framework/errors_impl.py", line 528, in exit c_api.TF_GetCode(self.status.status)) tensorflow.python.framework.errors_impl.NotFoundError: models; No such file or directory
Can anyone help me with how can I resolve this error?
Thanks in advance.
@rtaori Hi, thanks for your project. I noticed that your project is very similar to another one(audio_adversarial_examples from carlini). I believe you referred to that project. So compared with that project, is your project exactly the same in terms of audio preprocessing(MFCC and the reverse of MFCC)? It's about the research I'm doing. Look forward to your reply.