The picture of network is not same as your code? about wavernn

[fatchord]

@hwong39 There's a connection coming from the resnet into fc2.

Anyway it turns out my 'throw in everything but the kitchen sink' approach to this model has made it more complicated than it needs to be. I'm currently training a much simpler version and will upload soon.

[fatchord]

@hdmjdp I don't see the discrepancy you point out. There are three dense layers at the end and the last two flow directly through without concat or residual between them.

[hdmjdp]

@fatchord

            x = x + h2
            x = torch.cat([x, a3_t], dim=1)
            x = F.relu(self.fc1(x))
            
            x = torch.cat([x, a4_t], dim=1)
            x = F.relu(self.fc2(x))
            x = self.fc3(x)

in your picture, there is a skip operation("+") between the second dense layer(fc1) and the third dense layer(fc2). But your code did not has .

[fatchord]

@hdmjdp If you take a look at the diagram again you will see three dense layers of which there is only a skip connection between the first (fc1) and the second (fc2). Not sure what I can do to convince you!

[hwong39]

@hdmjdp I have the same question with you.
@fatchord As you said "there is only a skip connection between the first (fc1) and the second (fc2)", but i did not find a skip connection between the first (fc1) and the second (fc2) in your NB4b.

[hdmjdp]

@fatchord “ a much simpler version and will upload soon.”， this version can run in realtime in cpu? I expected it very much.