update loss

inference.py

@@ -35,12 +35,6 @@ class MasteringStyleTransfer:
                                                         STEMS=['mixture'], \
                                                         EFFECTS=['eq', 'imager', 'loudness'])
         # Loss functions
-        self.af_loss = AudioFeatureLoss(
-            weights=ito_config['af_weights'],
-            sample_rate=ito_config['sample_rate'],
-            stem_separation=False,
-            use_clap=False
-        )
         self.clap_loss = CLAPFeatureLoss(distance_fn='cosine')
 
     def load_effects_encoder(self):
@@ -81,6 +75,13 @@ class MasteringStyleTransfer:
         min_loss_step = 0
         all_results = []
 
+        af_loss = AudioFeatureLoss(
+            weights=ito_config['af_weights'],
+            sample_rate=ito_config['sample_rate'],
+            stem_separation=False,
+            use_clap=False
+        )
+
         for step in range(ito_config['num_steps']):
             optimizer.zero_grad()
 
@@ -89,7 +90,7 @@ class MasteringStyleTransfer:
 
             # Compute loss
             if ito_config['loss_function'] == 'AudioFeatureLoss':
-                losses = self.af_loss(output_audio, reference_tensor)
+                losses = af_loss(output_audio, reference_tensor)
             elif ito_config['loss_function'] == 'CLAPFeatureLoss':
                 if ito_config['clap_target_type'] == 'Audio':
                     target = ito_reference_tensor