Try the new ones here I still get 'ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 512, 1])' on the Tacotron Pipeline though