why do you minus 2 in preprocessing ?

[TianYuan]

I would like to know why do you minus 2 in preprocessing ?

WaveRNN/utils/dataset.py

Line 69 in 3595219

 max_offsets = [x[0].shape[-1] -2 - (mel_win + 2 * hp.voc_pad) for x in batch] 

and you have + 2 * hp.voc_pad in mel_win, why + 2 * hp.voc_pad again here?
I am very confused and looking forward to your reply.