load classifier weights

.gitignore

@@ -17,5 +17,3 @@ build/
 # VSCode
 .vscode/
 *.code-workspace
-
-behaviour_model/

app.py

@@ -24,7 +24,7 @@ def create_demo():
     emotion_model.eval()
 
     behaviour_model = get_behaviour_model(
-        behaviour_model_path="behaviour_model/",
+        behaviour_model_path="classifier_weights.bin",
         device=device,
     )        
 

src/audio_processor.py

@@ -104,10 +104,22 @@ class AudioProcessor:
             str: The predicted emotion label.
 
         """
+        try:
+            input_frames, _ = librosa.load(
+                audio_path, 
+                sr=SAMPLING_RATE
+            ) 
+        except Exception as e:
+            gr.Error(f"Error loading audio file: {e}.")
 
         print("Segmenting audio...")
         out = self.segmentation_model(
-            inputs=audio_path,
+            inputs={
+                "raw": input_frames,
+                "sampling_rate": SAMPLING_RATE,
+            }, 
+            chunk_length_s=30, 
+            stride_length_s=5,
             return_timestamps=True,
         )
 

src/model/behaviour_model.py

@@ -1,4 +1,3 @@
-import os
 import argparse
 import torch
 from .make_model import make_model
@@ -20,9 +19,10 @@ hparams_dict = {
 }
 hparams = argparse.Namespace(**hparams_dict)
 
-def get_behaviour_model(behaviour_model_path, device):
-    state_dict = torch.load(os.path.join(behaviour_model_path, 'pytorch_model.bin'), map_location=device)
+def get_behaviour_model(classifier_weights_path, device):
+    state_dict = torch.load(classifier_weights_path, map_location=device)
     model = make_model(hparams)
-    model.load_state_dict(state_dict)
+    model.classifier.load_state_dict(state_dict)
+    model.eval()
 
     return model

src/model/classifier_weights.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c77df5b5cd060698cf2ad93cef2f1b23795ef2faebc0860a5acddc6d87d47b3
+size 1846682