Request on details of the data augmentation module

[kur114]

Hello, I am in the process of training a model and find myself uncertain about the specifics of the data augmentation module. Here are two questions:

1. Is the module implemented using this code: https://github.com/revsic/torch-nansypp/blob/main/utils/augment?
2. Could you specify which hyperparameters in your configuration are passed to the augmentation module?

Thank you for your reply!

[hayeong0]

Hello, thank you for showing interest in our work. I have updated the augmentation code that I used for this work, so please refer to it:

• https://github.com/hayeong0/DDDM-VC/tree/master/augmentation

We also used the hyperparameter ratio proposed by NANSY, and observed that other values yielded worse or similar results, so we used the same hyperparameter. Specifically, it is as follows:

• Formant shifting: U(1, 1.4)
• Pitch randomization: U(1, 2)
• random frequency shaping: (1, 1.5)

[kur114]

Thank you very much for your detailed and helpful response. Your detailed implementation code has significantly clarified my understanding and will greatly assist me in my model training efforts.