5 layers

Author: MaxMax2016

config/maxgan.yaml

@@ -26,9 +26,9 @@ audio:
   mel_fmax: 8000.0
 #############################
 gen:
-  upsample_rates: [10,4,2,2]
-  upsample_kernel_sizes: [16,16,4,4]
-  upsample_initial_channel: 384
+  upsample_rates: [5,4,2,2,2]
+  upsample_kernel_sizes: [15,12,4,4,4]
+  upsample_initial_channel: 512
   resblock_kernel_sizes: [3,7,11]
   resblock_dilation_sizes: [[1,3,5], [1,3,5], [1,3,5]]
 #############################

model/generator.py

@@ -8,7 +8,8 @@
 from torch.nn.utils import remove_weight_norm
 
 from .nsf import SourceModuleHnNSF
-from .bigv import init_weights, AMPBlock
+from .bigv import init_weights, SnakeBeta, AMPBlock
+from .alias import Activation1d
 
 
 class SpeakerAdapter(nn.Module):
@@ -72,6 +73,7 @@ def __init__(self, hp):
             # spk
             self.adapter.append(SpeakerAdapter(
                 256, hp.gen.upsample_initial_channel // (2 ** (i + 1))))
+            # print(f'ups: {i} {k}, {u}, {(k - u) // 2}')
             # base
             self.ups.append(nn.ModuleList([
                 weight_norm(ConvTranspose1d(hp.gen.upsample_initial_channel // (2 ** i),
@@ -106,6 +108,8 @@ def __init__(self, hp):
                 self.resblocks.append(AMPBlock(hp, ch, k, d))
 
         # post conv
+        activation_post = SnakeBeta(ch, alpha_logscale=True)
+        self.activation_post = Activation1d(activation=activation_post)
         self.conv_post = weight_norm(Conv1d(ch, 1, 7, 1, padding=3))
 
         # weight initialization
@@ -120,6 +124,7 @@ def forward(self, spk, x, pos, f0):
         har_source = self.m_source(f0)
         har_source = har_source.transpose(1, 2)
         # pre conv
+        # x = x + torch.randn_like(x)       # for last train
         x = self.cond_pre(x)                # [B, L, D]
         p = self.cond_pos(pos)
         x = x + p
@@ -145,7 +150,7 @@ def forward(self, spk, x, pos, f0):
             x = xs / self.num_kernels
 
         # post conv
-        x = nn.functional.leaky_relu(x)
+        x = self.activation_post(x)
         x = self.conv_post(x)
         x = torch.tanh(x)
         return x

utils/train.py

@@ -142,7 +142,10 @@ def train(rank, args, chkpt_path, hp, hp_str):
 
             score_loss = score_loss / len(res_fake + period_fake)
 
+            # for fast train
             loss_g = score_loss + stft_loss + mel_loss
+            # for last train
+            # loss_g = score_loss + stft_loss
 
             loss_g.backward()
             optim_g.step()