Closed alexdemartos closed 3 years ago
If you're changing the hop size, make sure you also increase n_fft
to 2048.
For a hop size of 300, you'll want something like this:
self.conv1 = ConvTranspose2d(1, 1, [3, 30], stride=[1, 15], padding=[1, 8], output_padding=[0, 1])
self.conv2 = ConvTranspose2d(1, 1, [3, 40], stride=[1, 20], padding=[1, 10])
Thanks very much for your help and fast response ;)
Hi,
I'm having trouble with the upsampling of the mel spectrograms. How should I change the ConvTranspose2d kernel, stride and padding to match a hop size of 300? Thank you in advance.