Merge pull request #1188 from IAHispano/formatter/main

chore(format): run black on main

Author: blaisewf

rvc/lib/algorithm/discriminators.py

@@ -21,9 +21,14 @@ class MultiPeriodDiscriminator(torch.nn.Module):
             Defaults to False.
     """
 
-    def __init__(self, use_spectral_norm: bool = False, checkpointing: bool = False, version: str = "v2"):
+    def __init__(
+        self,
+        use_spectral_norm: bool = False,
+        checkpointing: bool = False,
+        version: str = "v2",
+    ):
         super().__init__()
-        
+
         if version == "v1":
             periods = [2, 3, 5, 7, 11, 17]
             resolutions = []
@@ -33,12 +38,15 @@ def __init__(self, use_spectral_norm: bool = False, checkpointing: bool = False,
         elif version == "v3":
             periods = [2, 3, 5, 7, 11]
             resolutions = [[1024, 120, 600], [2048, 240, 1200], [512, 50, 240]]
-            
+
         self.checkpointing = checkpointing
         self.discriminators = torch.nn.ModuleList(
             [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
             + [DiscriminatorP(p, use_spectral_norm=use_spectral_norm) for p in periods]
-            + [DiscriminatorR(r, use_spectral_norm=use_spectral_norm) for r in resolutions]
+            + [
+                DiscriminatorR(r, use_spectral_norm=use_spectral_norm)
+                for r in resolutions
+            ]
         )
 
     def forward(self, y, y_hat):
@@ -160,6 +168,7 @@ def forward(self, x):
         x = torch.flatten(x, 1, -1)
         return x, fmap
 
+
 class DiscriminatorR(torch.nn.Module):
     def __init__(self, resolution, use_spectral_norm=False):
         super().__init__()
@@ -170,11 +179,49 @@ def __init__(self, resolution, use_spectral_norm=False):
 
         self.convs = torch.nn.ModuleList(
             [
-                norm_f(torch.nn.Conv2d( 1, 32, (3, 9),                padding=(1, 4),)),
-                norm_f(torch.nn.Conv2d(32, 32, (3, 9), stride=(1, 2), padding=(1, 4),)),
-                norm_f(torch.nn.Conv2d(32, 32, (3, 9), stride=(1, 2), padding=(1, 4),)),
-                norm_f(torch.nn.Conv2d(32, 32, (3, 9), stride=(1, 2), padding=(1, 4),)),
-                norm_f(torch.nn.Conv2d(32, 32, (3, 3),                padding=(1, 1),)),
+                norm_f(
+                    torch.nn.Conv2d(
+                        1,
+                        32,
+                        (3, 9),
+                        padding=(1, 4),
+                    )
+                ),
+                norm_f(
+                    torch.nn.Conv2d(
+                        32,
+                        32,
+                        (3, 9),
+                        stride=(1, 2),
+                        padding=(1, 4),
+                    )
+                ),
+                norm_f(
+                    torch.nn.Conv2d(
+                        32,
+                        32,
+                        (3, 9),
+                        stride=(1, 2),
+                        padding=(1, 4),
+                    )
+                ),
+                norm_f(
+                    torch.nn.Conv2d(
+                        32,
+                        32,
+                        (3, 9),
+                        stride=(1, 2),
+                        padding=(1, 4),
+                    )
+                ),
+                norm_f(
+                    torch.nn.Conv2d(
+                        32,
+                        32,
+                        (3, 3),
+                        padding=(1, 1),
+                    )
+                ),
             ]
         )
         self.conv_post = norm_f(torch.nn.Conv2d(32, 1, (3, 3), padding=(1, 1)))
@@ -183,7 +230,7 @@ def forward(self, x):
         fmap = []
 
         x = self.spectrogram(x).unsqueeze(1)
-        
+
         for layer in self.convs:
             x = F.leaky_relu(layer(x), self.lrelu_slope)
             fmap.append(x)
@@ -195,13 +242,17 @@ def forward(self, x):
     def spectrogram(self, x):
         n_fft, hop_length, win_length = self.resolution
         pad = int((n_fft - hop_length) / 2)
-        x = F.pad(x, (pad, pad), mode="reflect",).squeeze(1)
+        x = F.pad(
+            x,
+            (pad, pad),
+            mode="reflect",
+        ).squeeze(1)
         x = torch.stft(
             x,
             n_fft=n_fft,
             hop_length=hop_length,
             win_length=win_length,
-            window=torch.ones(win_length, device=x.device), 
+            window=torch.ones(win_length, device=x.device),
             center=False,
             return_complex=True,
         )

rvc/train/train.py

@@ -429,7 +429,9 @@ def run(
     )
 
     net_d = MultiPeriodDiscriminator(
-        config.model.use_spectral_norm, checkpointing=checkpointing, version=disc_version,
+        config.model.use_spectral_norm,
+        checkpointing=checkpointing,
+        version=disc_version,
     )
 
     if torch.cuda.is_available():

tabs/train/train.py

@@ -343,7 +343,7 @@ def train_tab():
                     info=i18n(
                         "Choose the vocoder for audio synthesis:\n- **HiFi-GAN**: Default option, compatible with all clients.\n- **MRF HiFi-GAN**: Higher fidelity, Applio-only.\n- **RefineGAN**: Superior audio quality, Applio-only."
                     ),
-                    choices=["HiFi-GAN", "RefineGAN"], #"MRF HiFi-GAN", ],
+                    choices=["HiFi-GAN", "RefineGAN"],  # "MRF HiFi-GAN", ],
                     value="HiFi-GAN",
                     interactive=True,
                     visible=True,