Update app.py

app.py

@@ -482,7 +482,7 @@ def solve_optimal_alignment(original_segments, generated_durations, total_durati
             )
 
     return original_segments
-
+    
 def ocr_frame_worker(args):
     frame_idx, frame_time, frame = args
     ocr = PaddleOCR(use_angle_cls=True, lang="ch")  # Initialize OCR inside worker
@@ -491,7 +491,17 @@ def ocr_frame_worker(args):
     combined_text = " ".join(texts).strip()
     return {"time": frame_time, "text": combined_text}
 
-def extract_ocr_subtitles_parallel(video_path, interval_sec=0.5, num_workers=4):
+def frame_is_in_audio_segments(frame_time, audio_segments, tolerance=0.2):
+    """
+    Check if the frame_time falls within any audio segment (plus/minus tolerance).
+    """
+    for segment in audio_segments:
+        start, end = segment["start"], segment["end"]
+        if (start - tolerance) <= frame_time <= (end + tolerance):
+            return True
+    return False
+
+def extract_ocr_subtitles_parallel(video_path, transcription_json, interval_sec=0.5, num_workers=4):
     cap = cv2.VideoCapture(video_path)
     fps = cap.get(cv2.CAP_PROP_FPS)
     frames = []
@@ -501,7 +511,8 @@ def extract_ocr_subtitles_parallel(video_path, interval_sec=0.5, num_workers=4):
     while success:
         if frame_idx % int(fps * interval_sec) == 0:
             frame_time = frame_idx / fps
-            frames.append((frame_idx, frame_time, frame.copy()))
+            if frame_is_in_audio_segments(frame_time, transcription_json):
+                frames.append((frame_idx, frame_time, frame.copy()))
         success, frame = cap.read()
         frame_idx += 1
     cap.release()
@@ -517,6 +528,7 @@ def extract_ocr_subtitles_parallel(video_path, interval_sec=0.5, num_workers=4):
                     ocr_results.append(result)
             except Exception as e:
                 print(f"⚠️ OCR failed for a frame: {e}")
+
     return ocr_results
 
 def collapse_ocr_subtitles(ocr_json, text_similarity_threshold=90):
@@ -550,9 +562,13 @@ def post_edit_transcribed_segments(transcription_json, video_path,
     use OCR subtitles to post-correct and safely insert missing captions.
     """
 
-    # Step 1: Extract OCR subtitles
-    ocr_json = extract_ocr_subtitles_parallel(video_path, interval_sec=interval_sec, num_workers=num_workers)
-
+    # Step 1: Extract OCR subtitles (only near audio segments)
+    ocr_json = extract_ocr_subtitles_parallel(
+        video_path, 
+        transcription_json, 
+        interval_sec=interval_sec, 
+        num_workers=num_workers
+    )
     # Step 2: Collapse repetitive OCR
     collapsed_ocr = collapse_ocr_subtitles(ocr_json, text_similarity_threshold=90)