Update src/audio_preprocessing.py (#1)

- Update src/audio_preprocessing.py (9705702d0bf258a501d1601a8740dfad4515f169)


Co-authored-by: ahk-d <ahk-d@users.noreply.huggingface.co>

src/audio_preprocessing.py

@@ -100,4 +100,38 @@ def process_wav(wav_path, target_sr, do_trim_silences=True):
     # Loudness normalization to -20dB
     audio = match_target_amplitude(audio, -20.0)
 
-    return audio
+    return audio
+
+import numpy as np
+
+def assess_pronunciation_quality(dist_matrix, path):
+    # Extract distances along the alignment path
+    path_distances = [dist_matrix[i, j] for i, j in zip(*path)]
+    
+    # Calculate global min and max for normalization
+    global_min = dist_matrix.min()
+    global_max = dist_matrix.max()
+    
+    # Normalize distances
+    normalized_distances = [(d - global_min) / (global_max - global_min) for d in path_distances]
+    
+    # Analyze normalized distances
+    num_red_segments = sum(1 for d in normalized_distances if d >= 0.5)
+    total_segments = len(normalized_distances)
+    red_percentage = num_red_segments / total_segments if total_segments > 0 else 0.0
+    
+    # Calculate quality score and repetition need
+    quality_score = 1 - red_percentage
+    needs_repeat = red_percentage > 0.5
+    
+    # Print debug information
+    print(f"Raw distance stats:")
+    print(f"  Min distance: {min(path_distances):.4f}")
+    print(f"  Max distance: {max(path_distances):.4f}")
+    print(f"  Mean distance: {np.mean(path_distances):.4f}")
+    print(f"\nNormalized distance stats:")
+    print(f"  Number of red segments (>= 0.5): {num_red_segments}")
+    print(f"  Total segments: {total_segments}")
+    print(f"\nRed percentage: {red_percentage * 100:.2f}%")
+    
+    return quality_score, needs_repeat