Update app.py

app.py

@@ -76,16 +76,14 @@ def process_video(video_file, language="hindi", progress=gr.Progress()):
         video.close()
         print(f"Audio extracted: {audio_path} ({os.path.getsize(audio_path)} bytes)")
         
-        # Transcribe with word-level timestamps for better control
+        # Transcribe - keep chunk_length_s for full audio, but simplify everything else
         progress(0.4, desc="Transcribing with Whisper Turbo...")
         print(f"Starting transcription (language: {language})...")
         
         transcription_result = pipe(
             audio_path,
-            return_timestamps="word",  # Get word-level timestamps
-            chunk_length_s=30,
-            stride_length_s=(4, 2),
-            batch_size=8,
+            return_timestamps=True,
+            chunk_length_s=30,  # Keep this to process full audio
             generate_kwargs={
                 "language": language if language != "auto" else None,
                 "task": "transcribe"
@@ -94,53 +92,7 @@ def process_video(video_file, language="hindi", progress=gr.Progress()):
         
         print(f"Transcription result keys: {transcription_result.keys()}")
         
-        # Create optimal chunks from word timestamps
-        if 'chunks' in transcription_result and transcription_result['chunks']:
-            words_data = transcription_result['chunks']
-            
-            # Group words into readable subtitle chunks (max 10-12 words, max 4 seconds)
-            optimal_chunks = []
-            current_chunk_words = []
-            current_chunk_start = None
-            
-            for word_info in words_data:
-                word_text = word_info['text']
-                word_start, word_end = word_info['timestamp']
-                
-                if current_chunk_start is None:
-                    current_chunk_start = word_start
-                
-                current_chunk_words.append(word_text)
-                
-                # Check if we should end this chunk
-                chunk_duration = word_end - current_chunk_start
-                should_break = (
-                    len(current_chunk_words) >= 10 or  # Max 10 words
-                    chunk_duration >= 4.0 or  # Max 4 seconds
-                    word_text.strip().endswith(('.', '?', '!', '।'))  # Sentence end
-                )
-                
-                if should_break:
-                    optimal_chunks.append({
-                        'timestamp': (current_chunk_start, word_end),
-                        'text': ''.join(current_chunk_words)
-                    })
-                    current_chunk_words = []
-                    current_chunk_start = None
-            
-            # Add remaining words
-            if current_chunk_words:
-                last_word = words_data[-1]
-                optimal_chunks.append({
-                    'timestamp': (current_chunk_start, last_word['timestamp'][1]),
-                    'text': ''.join(current_chunk_words)
-                })
-            
-            transcription_result['chunks'] = optimal_chunks
-            print(f"Created {len(optimal_chunks)} optimal subtitle chunks from word timestamps")
-        else:
-            print("Warning: No word-level timestamps returned")
-        
+        # Verify we have chunks
         if not transcription_result.get('chunks') or len(transcription_result['chunks']) == 0:
             error_msg = f"Transcription returned no segments"
             print(error_msg)
@@ -151,17 +103,12 @@ def process_video(video_file, language="hindi", progress=gr.Progress()):
         print(f"Transcription complete: {len(transcription_result['chunks'])} segments")
         
         # Debug: Print first and last chunks
-        if 'chunks' in transcription_result:
-            print(f"\nFirst 3 chunks structure:")
-            for i, chunk in enumerate(transcription_result['chunks'][:3], 1):
-                duration = chunk['timestamp'][1] - chunk['timestamp'][0]
-                word_count = len(chunk['text'].split())
-                print(f"  Chunk {i}: {duration:.2f}s, {word_count} words - {chunk['text'][:50]}...")
-            print(f"\nLast 3 chunks structure:")
-            for i, chunk in enumerate(transcription_result['chunks'][-3:], len(transcription_result['chunks'])-2):
-                duration = chunk['timestamp'][1] - chunk['timestamp'][0]
-                word_count = len(chunk['text'].split())
-                print(f"  Chunk {i}: {duration:.2f}s, {word_count} words - {chunk['text'][:50]}...")
+        print(f"\nFirst 3 chunks:")
+        for i, chunk in enumerate(transcription_result['chunks'][:3], 1):
+            print(f"  {i}: {chunk}")
+        print(f"\nLast 3 chunks:")
+        for i, chunk in enumerate(transcription_result['chunks'][-3:], len(transcription_result['chunks'])-2):
+            print(f"  {i}: {chunk}")
         
         # Post-process: Split overly long chunks for better readability
         processed_chunks = []