Training VoiceFixer on Custom Dataset

[harshvardhan-truefan]

Hi,
I want to restore my speech audio samples. For that, I want to train a the voicefixer repo on my custom dataset because when I'm using the pretrained voicefixer, the voice of the speaker is getting changed slightly.

When I'm trying to train the data of custom dataset, I'm unable to do so. I'm getting the following error while training:
Runtime-error: The size of tensor a (301) must match the size of tensor b (934) at non-singleton dimension 2

This is coming during performing sanity check. The input and target tensor values are not same. Any idea how I can resolve this issue ? Also if you can mention the steps for training the model on custom data, that we be really helpful.

Thanks in advance !
Look forward to hearing from you.

Regards,
Harsh