Update modeling_wavtokenizer.py

modeling_wavtokenizer.py

@@ -396,8 +396,8 @@ class Backbone(nn.Module):
                  num_bandwidths=4):
         super().__init__()
         
-        # Input projection: backbone.embed
-        self.embed = nn.Conv1d(input_dim, dim, kernel_size=3, padding=1)
+        # Input projection: backbone.embed (kernel_size=7 to match checkpoint)
+        self.embed = nn.Conv1d(input_dim, dim, kernel_size=7, padding=3)
         
         # Input normalization: backbone.norm
         self.norm = AdaLayerNorm(dim, num_bandwidths)