Problem in changing sample rate into 16k Hz

[xuexidi]

Hi! Thanks for your code!

When I train the model in the sample rate of 22.05k Hz，and keep the other params the same with your given params. Every thing goes well.

But，when I tried to modified the sample rate to 16k Hz
the configration is:
sr：16k Hz
n_fft：800
hop_legth：200
win_length：800

when I started trainning， I got this warning:
train.py:187: UserWarning: Using a target size (torch.Size([64, 80, 51])) that is different to the input size (torch.Size([64, 80, 40])). This will likely lead to incorrect results due to broadcasting. Please ensure they have the same size.

And an error occured：
Traceback (most recent call last):
File "train.py", line 284, in
train_melgan()
File "train.py", line 187, in train_melgan
s_error = F.l1_loss(s_t, s_pred_t).item()
File "/home/pxdevice/anaconda3/envs/oneshot/lib/python3.6/site-packages/torch/nn/functional.py", line 2616, in l1_loss
expanded_input, expanded_target = torch.broadcast_tensors(input, target)
File "/home/pxdevice/anaconda3/envs/oneshot/lib/python3.6/site-packages/torch/functional.py", line 65, in broadcast_tensors
return _VF.broadcast_tensors(tensors)
RuntimeError: The size of tensor a (40) must match the size of tensor b (51) at non-singleton dimension 2

I have no idea to fix this error，should I modify the structure params of Generator network? Could you please help me, please!

[allenhung1025]

Hi,
you can make target and input last dimension(dim=2) the same size

[RussellSB]

Hi,
you can make target and input last dimension(dim=2) the same size

@allenhung1025 Where is this adjustment required to be made? I'm experiencing the same issue

[allenhung1025]

@RussellSB Hi, for Example, if the target size is (torch.Size([64, 80, 51])) and the input size is (torch.Size([64, 80, 40])), you can compute loss between target[:, :, :40] and input.