shithub: opus

Info  •  Files  •  Log  •  Branches

Also use previous frame

--- a/dnn/torch/fargan/fargan.py

+++ b/dnn/torch/fargan/fargan.py

@@ -9,7 +9,7 @@

 fid_dict = {}

 def dump_signal(x, filename):

-    #return

+    return

     if filename in fid_dict:

         fid = fid_dict[filename]

     else:

@@ -143,7 +143,7 @@

         gain_param = 1 if self.has_gain else 0

-        self.sig_dense1 = nn.Linear(3*self.subframe_size+self.passthrough_size+self.cond_size+gain_param, self.cond_size, bias=False)

+        self.sig_dense1 = nn.Linear(4*self.subframe_size+self.passthrough_size+self.cond_size+gain_param, self.cond_size, bias=False)

         self.sig_dense2 = nn.Linear(self.cond_size, self.cond_size, bias=False)

         self.gru1 = nn.GRUCell(self.cond_size, self.cond_size, bias=False)

         self.gru2 = nn.GRUCell(self.cond_size, self.cond_size, bias=False)

@@ -171,7 +171,8 @@

         idx = 256-torch.maximum(torch.tensor(self.subframe_size, device=device), period[:,None])

         rng = torch.arange(self.subframe_size, device=device)

         idx = idx + rng[None,:]

-        prev = torch.gather(exc_mem, 1, idx)

+        pred = torch.gather(exc_mem, 1, idx)

+        prev = torch.cat([pred, prev], 1) 

         #prev = prev*0

         dump_signal(prev, 'pitch_exc.f32')

         dump_signal(exc_mem, 'exc_mem.f32')

@@ -196,7 +197,7 @@

         dump_signal(sig_out, 'exc_out.f32')

         if self.has_gain:

             pitch_gain = torch.exp(self.gain_dense_out(gru3_out))

-            sig_out = (sig_out + pitch_gain*prev[:,:-1]) * gain

+            sig_out = (sig_out + pitch_gain*prev[:,:self.subframe_size]) * gain

         exc_mem = torch.cat([exc_mem[:,self.subframe_size:], sig_out], 1)

         dump_signal(sig_out, 'sig_out.f32')

         return sig_out, exc_mem, (gru1_state, gru2_state, gru3_state, passthrough)

--- a/dnn/torch/fargan/train_fargan.py

+++ b/dnn/torch/fargan/train_fargan.py

@@ -54,7 +54,7 @@

 sequence_length = args.sequence_length

 lr_decay = args.lr_decay

-adam_betas = [0.9, 0.99]

+adam_betas = [0.8, 0.99]

 adam_eps = 1e-8

 features_file = args.features

 signal_file = args.signal

@@ -92,7 +92,7 @@

 dataloader = torch.utils.data.DataLoader(dataset, batch_size=batch_size, shuffle=True, drop_last=True, num_workers=4)

-optimizer = torch.optim.Adam(model.parameters(), lr=lr, betas=adam_betas, eps=adam_eps)

+optimizer = torch.optim.AdamW(model.parameters(), lr=lr, betas=adam_betas, eps=adam_eps)

 # learning rate scheduler

-- 

⑨