Update main.py

main.py

@@ -46,16 +46,12 @@ def custom_dtw(X, Y, metric='euclidean'):
         D: Cost matrix
         wp: Warping path
     """
-    # Get sequence lengths
     n, m = len(X), len(Y)
-    
-    # Initialize cost matrix
     D = np.zeros((n + 1, m + 1))
     D[0, 1:] = np.inf
     D[1:, 0] = np.inf
     D[0, 0] = 0
     
-    # Fill cost matrix
     for i in range(1, n + 1):
         for j in range(1, m + 1):
             if metric == 'euclidean':
@@ -64,7 +60,6 @@ def custom_dtw(X, Y, metric='euclidean'):
                 cost = 1 - np.dot(X[i-1], Y[j-1]) / (np.linalg.norm(X[i-1]) * np.linalg.norm(Y[j-1]))
             D[i, j] = cost + min(D[i-1, j], D[i, j-1], D[i-1, j-1])
     
-    # Backtracking
     wp = [(n, m)]
     i, j = n, m
     while i > 0 or j > 0:
@@ -92,7 +87,6 @@ class QuranRecitationComparer:
         self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         print(f"Using device: {self.device}")
 
-        # Load model and processor once during initialization
         try:
             if token:
                 print(f"Loading model {model_name} with token...")
@@ -105,6 +99,8 @@ class QuranRecitationComparer:
                 
             self.model = self.model.to(self.device)
             self.model.eval()
+            # Set the configuration to always return hidden states
+            self.model.config.output_hidden_states = True
             print("Model loaded successfully!")
         except Exception as e:
             print(f"Error loading model: {str(e)}")
@@ -146,7 +142,8 @@ class QuranRecitationComparer:
             ).input_values.to(self.device)
             
             with torch.no_grad():
-                outputs = self.model(inputs, output_hidden_states=True)
+                # Call the model without passing output_hidden_states explicitly.
+                outputs = self.model(inputs)
             
             hidden_states = outputs.hidden_states[-1]
             embedding_seq = hidden_states.squeeze(0).cpu().numpy()
@@ -158,7 +155,6 @@ class QuranRecitationComparer:
 
     def compute_dtw_distance(self, features1, features2):
         """Compute the DTW distance between two sequences of features."""
-        # Make sure features are 2D arrays
         if features1.ndim == 1:
             features1 = features1.reshape(-1, 1)
         if features2.ndim == 1:
@@ -166,7 +162,6 @@ class QuranRecitationComparer:
             
         print(f"Feature shapes: {features1.shape}, {features2.shape}")
         
-        # Use a subsample if the sequences are too long to avoid memory issues
         max_length = 300
         if features1.shape[0] > max_length or features2.shape[0] > max_length:
             step1 = max(1, features1.shape[0] // max_length)
@@ -182,7 +177,6 @@ class QuranRecitationComparer:
             return normalized_distance
         except Exception as e:
             print(f"Error in compute_dtw_distance: {str(e)}")
-            # Fallback to a basic similarity measure if DTW fails
             mean_1 = np.mean(features1, axis=0)
             mean_2 = np.mean(features2, axis=0)
             euclidean_distance = np.sqrt(np.sum((mean_1 - mean_2) ** 2))
@@ -222,7 +216,6 @@ class QuranRecitationComparer:
             audio = self.load_audio(file_path)
             embedding = self.get_deep_embedding(audio)
             
-            # Store in cache for future use
             self.embedding_cache[file_path] = embedding
             print(f"Embedding shape: {embedding.shape}")
             
@@ -234,35 +227,30 @@ class QuranRecitationComparer:
     def predict(self, file_path1, file_path2):
         """
         Predict the similarity between two audio files.
-        This method can be called repeatedly without reloading the model.
-
+        
         Args:
             file_path1 (str): Path to first audio file
             file_path2 (str): Path to second audio file
-
+        
         Returns:
             float: Similarity score
             str: Interpretation of similarity
         """
         print(f"Comparing {file_path1} and {file_path2}")
         try:
-            # Get embeddings (using cache if available)
             embedding1 = self.get_embedding_for_file(file_path1)
             embedding2 = self.get_embedding_for_file(file_path2)
 
-            # Compute DTW distance
             print("Computing DTW distance...")
             norm_distance = self.compute_dtw_distance(embedding1.T, embedding2.T)
             print(f"Normalized distance: {norm_distance}")
 
-            # Interpret results
             interpretation, similarity_score = self.interpret_similarity(norm_distance)
             print(f"Similarity score: {similarity_score}, Interpretation: {interpretation}")
 
             return similarity_score, interpretation
         except Exception as e:
             print(f"Error in predict: {str(e)}")
-            # Return a fallback response in case of error
             return 0, f"Error comparing files: {str(e)}"
 
     def clear_cache(self):
@@ -273,7 +261,6 @@ class QuranRecitationComparer:
 # Global variable for the comparer instance
 comparer = None
 
-# Use the new lifespan API
 @app.on_event("startup")
 async def startup_event():
     """Initialize the model when the application starts."""
@@ -287,7 +274,6 @@ async def startup_event():
         print("Model initialized and ready for predictions!")
     except Exception as e:
         print(f"Error initializing model: {str(e)}")
-        # Don't raise here, let the app continue to load even if model fails
 
 @app.get("/")
 async def root():
@@ -319,7 +305,6 @@ async def compare_files(
     print(f"Created temporary directory: {temp_dir}")
     
     try:
-        # Save uploaded files to temporary directory
         temp_file1 = os.path.join(temp_dir, file1.filename)
         temp_file2 = os.path.join(temp_dir, file2.filename)
         
@@ -333,7 +318,6 @@ async def compare_files(
         
         print(f"Files saved to: {temp_file1} and {temp_file2}")
             
-        # Compare the files
         similarity_score, interpretation = comparer.predict(temp_file1, temp_file2)
         
         return ComparisonResult(
@@ -346,7 +330,6 @@ async def compare_files(
         raise HTTPException(status_code=500, detail=f"Error processing files: {str(e)}")
     
     finally:
-        # Clean up temporary files
         print(f"Cleaning up temporary directory: {temp_dir}")
         shutil.rmtree(temp_dir, ignore_errors=True)
 
@@ -360,4 +343,4 @@ async def clear_cache():
     return {"message": "Embedding cache cleared successfully"}
 
 if __name__ == "__main__":
-    uvicorn.run("main:app", host="0.0.0.0", port=7860, log_level="info")
+    uvicorn.run("main:app", host="0.0.0.0", port=7860, log_level="info")