⚗️ | Experimenting, again.

generator.py

@@ -1,32 +1,31 @@
 import torch.nn as nn
 
 class SISUGenerator(nn.Module):
-    def __init__(self, upscale_scale=4):  # No noise_dim parameter
+    def __init__(self):
         super(SISUGenerator, self).__init__()
-        layer = 32
-        # Convolution layers
+        layer = 16
+        # Convolution layers with BatchNorm and Residuals
         self.conv1 = nn.Sequential(
-            nn.Conv1d(1, layer * 2, kernel_size=7, padding=1),
+            nn.Conv1d(1, layer * 2, kernel_size=7, padding=3),
+            nn.BatchNorm1d(layer * 2),
             nn.PReLU(),
-            nn.Conv1d(layer * 2, layer * 5, kernel_size=5, padding=1),
+            nn.Conv1d(layer * 2, layer * 5, kernel_size=7, padding=3),
+            nn.BatchNorm1d(layer * 5),
             nn.PReLU(),
-            nn.Conv1d(layer * 5, layer * 5, kernel_size=3, padding=1),
-            nn.PReLU()
+            nn.Conv1d(layer * 5, layer * 5, kernel_size=7, padding=3),
+            nn.BatchNorm1d(layer * 5),
+            nn.PReLU(),
+        )
+        self.final_layer = nn.Sequential(
+            nn.Conv1d(layer * 5, layer * 2, kernel_size=5, padding=2),
+            nn.BatchNorm1d(layer * 2),
+            nn.PReLU(),
+            nn.Conv1d(layer * 2, 1, kernel_size=3, padding=1),
+            # nn.Tanh()  # Normalize audio... if needed...
         )
 
-        # Transposed convolution for upsampling
-        self.upsample = nn.ConvTranspose1d(layer * 5, layer * 5, kernel_size=upscale_scale,  stride=upscale_scale)
-
-        self.conv2 = nn.Sequential(
-            nn.Conv1d(layer * 5, layer * 5, kernel_size=3, padding=1),
-            nn.PReLU(),
-            nn.Conv1d(layer * 5, layer * 2, kernel_size=5, padding=1),
-            nn.PReLU(),
-            nn.Conv1d(layer * 2, 1, kernel_size=7, padding=1)
-        )
-
-    def forward(self, x, upscale_scale=4):
+    def forward(self, x):
+        residual = x
         x = self.conv1(x)
-        x = self.upsample(x)
-        x = self.conv2(x)
-        return x
+        x = self.final_layer(x)
+        return x + residual