modify fx norm

inference.py

@@ -143,7 +143,9 @@ class MasteringStyleTransfer:
 
         # Apply fx normalization for input audio during mastering style transfer
         if normalize:
+            print(f"before normalization: {data.shape}")
             data = self.fx_normalizer.normalize_audio(data, 'mixture')
+            print(f"after normalization: {data.shape}")
 
         # Convert to torch tensor
         data_tensor = torch.FloatTensor(data).unsqueeze(0)
@@ -154,8 +156,8 @@ class MasteringStyleTransfer:
         print(f"input: {input_audio}")
         print(f"reference: {reference_audio}")
         input_tensor = self.preprocess_audio(input_audio, self.args.sample_rate, normalize=True)
-        reference_tensor = self.preprocess_audio(reference_audio, self.args.sample_rate)
         print(f"input_tensor: {input_tensor.shape}")
+        reference_tensor = self.preprocess_audio(reference_audio, self.args.sample_rate)
         print(f"reference_tensor: {reference_tensor.shape}")
 
         reference_feature = self.get_reference_embedding(reference_tensor)