Update video2.py

video2.py

@@ -87,15 +87,16 @@ def detect_language(word):
 
 def calculate_pause(text_chunk):
     """
-    NATURAL PAUSE DURATIONS (Reduced for speed/flow)
+    MAX EFFICIENCY PAUSE DURATIONS
+    Only add a brief pause for meaningful punctuation.
     """
     t = text_chunk.strip()
-    # Quick breath for comma (was 250, now 100)
-    if t.endswith(',') or t.endswith(';'): return 100 
-    # Standard sentence end (was 650, now 350)
-    elif t.endswith('.'): return 350   
-    elif t.endswith('?'): return 400 
-    elif t.endswith('!'): return 350
+    # Micro-breath (70ms) for comma/semicolon
+    if t.endswith(',') or t.endswith(';'): return 70 
+    # Quick sentence stop (250ms)
+    elif t.endswith('.'): return 250   
+    elif t.endswith('?'): return 300 
+    elif t.endswith('!'): return 250
     return 0 
 
 def analyze_and_segment(text):
@@ -152,14 +153,14 @@ async def generate_chunk_with_retry(segment_data, semaphore):
 
     voice = VOICES["Tamil"] if lang_type == 'indic' else VOICES["English"]
     
-    # Keeping English slightly slower helps it blend with Tamil speed
-    rate = "-5%" if lang_type == 'english' else "+0%"
+    # Max efficiency: Neutral rate (+0%) for all.
+    rate = "+0%"
     pitch = "+0Hz"
 
     for attempt in range(MAX_RETRIES):
         async with semaphore:
             try:
-                await asyncio.sleep(random.uniform(0.1, 0.4)) # Jitter
+                await asyncio.sleep(random.uniform(0.1, 0.4)) 
                 
                 fd, path = tempfile.mkstemp(suffix=f"_{idx}.mp3")
                 os.close(fd)
@@ -187,6 +188,9 @@ def process_and_stitch(results):
     
     final_audio = AudioSegment.empty()
     
+    # 50ms silence pad to the start of the entire output to prevent clipping the first word
+    final_audio += AudioSegment.silent(duration=50)
+
     for i, item in enumerate(results):
         try:
             path = item['path']
@@ -201,29 +205,22 @@ def process_and_stitch(results):
             else:
                 prev_item = results[i-1]
                 
-                # --- FLOW LOGIC ---
+                # --- ZERO-GAP FLOW LOGIC ---
                 if prev_item['pause'] > 0:
-                    # If there was a comma/period, use the calculated pause
+                    # If there was punctuation, insert the micro-silence.
                     silence = AudioSegment.silent(duration=prev_item['pause'])
                     final_audio += silence + segment_audio
                 else:
-                    # If switching languages (Voltage -> nu), use CROSSFADE
-                    # Crossfade blends the end of English with start of Tamil.
-                    # 20ms is fast enough to sound connected but not overlapped.
-                    try:
-                        final_audio = final_audio.append(segment_audio, crossfade=20)
-                    except:
-                        # Fallback for tiny segments
-                        final_audio += segment_audio
+                    # If continuous speech (same language or language switch without punctuation), 
+                    # use direct append for 0ms gap.
+                    final_audio += segment_audio
             
         except Exception as e:
-            print(f"Error stitching segment {i}: {e}")
             continue
 
     return final_audio
 
 async def natural_tts_engine(full_text, output_file, native_lang_code):
-    print("Analyzing...")
     segments = analyze_and_segment(full_text)
     
     tasks = []
@@ -234,13 +231,11 @@ async def natural_tts_engine(full_text, output_file, native_lang_code):
     
     raw_results = await asyncio.gather(*tasks)
     
-    print("Stitching for Natural Flow...")
     final_audio = process_and_stitch(raw_results)
     
     if not final_audio: return None
         
-    print("Mastering...")
-    # Compression ratio 2.5 makes speech punchy and clear
+    # Final Mastering: Ensures volume is consistent and clear
     final_audio = compress_dynamic_range(
         final_audio, 
         threshold=-15.0, 
@@ -251,7 +246,6 @@ async def natural_tts_engine(full_text, output_file, native_lang_code):
     final_audio = normalize(final_audio)
     
     final_audio.export(output_file, format="mp3", bitrate="320k")
-    print(f"✅ Saved: {output_file}")
     return output_file
 
 async def generate_tts(id, lines, lang_input):
@@ -272,6 +266,8 @@ async def generate_tts(id, lines, lang_input):
 
 
 
+
+
 def audio_func(id, lines, lang):
     loop = asyncio.new_event_loop()
     asyncio.set_event_loop(loop)