How to apply various attack functions during training

Thank you for providing the code.

However, I have one question.

According to resemble AI's blog, it is said to be robust (almost 100%) against various attacks (pitch shift, time stretch, high-pass, low-pass, compressor, delay, noise, resample). Can the '_attack' function inside '_perth_net.py' cover all the above attacks?

Do you plan to release the 'training script' later?