Update app.py

app.py

@@ -115,17 +115,18 @@ def get_user_ip():
         return request.headers.getlist("X-Forwarded-For")[0].split(',')[0].strip()
     return request.remote_addr
 
-# --- TTS HELPER FUNCTIONS (بدون حذف دیالوگ‌ها) ---
+# --- TTS HELPER FUNCTIONS (تمیزکننده و سازنده پرقدرت) ---
 def call_worker(index, chunk_payload):
     raw_text = chunk_payload.get("text", "")
-    clean_text = raw_text.strip()
     
-    # فقط اگر خود هوش مصنوعی متن خالی داده باشد (یعنی حرفی برای گفتن نباشد)، سکوت می‌گذاریم
+    # 🔴 پاکسازی متن از تمام کاراکترهای سمی (مارک‌داون، براکت و ...)
+    clean_text = re.sub(r'[*_`~#]', '', raw_text)
+    clean_text = re.sub(r'\[.*?\]|\(.*?\)', '', clean_text)
+    clean_text = clean_text.strip()
+    
     if not clean_text:
-        logging.warning(f"⚠️ [قطعه {index+1}] متن کاملا خالی بود.")
         return index, AudioSegment.silent(duration=500)
 
-    # زنده بودن مدل برای متون زیر 600 حرف
     use_live = True if len(clean_text) <= 600 else False
     
     target_speaker = chunk_payload.get("speaker")
@@ -141,7 +142,7 @@ def call_worker(index, chunk_payload):
         "fallback_to_live": True
     }
 
-    # افزایش تلاش به 50 بار برای اطمینان از ساخته شدن قطعی
+    # تلاش مستمر و بدون جا انداختن حتی یک کلمه
     max_attempts = 50 
     
     for attempt in range(max_attempts):
@@ -150,7 +151,7 @@ def call_worker(index, chunk_payload):
         
         for worker_url in workers:
             try:
-                logging.info(f"▶️ [قطعه {index+1}] ارسال درخواست به {worker_url} (تلاش {attempt+1}/{max_attempts})")
+                logging.info(f"▶️ [قطعه {index+1}] ارسال به {worker_url} (تلاش {attempt+1}/{max_attempts})")
                 response = requests.post(worker_url, json=worker_payload, headers=STANDARD_HEADERS, timeout=120)
                 
                 if response.status_code == 200:
@@ -164,16 +165,15 @@ def call_worker(index, chunk_payload):
                     continue
                     
             except Exception as e:
-                logging.error(f"❌ [قطعه {index+1}] قطعی ارتباط با کارگر: {str(e)[:40]}")
+                logging.error(f"❌ [قطعه {index+1}] قطعی ارتباط: {str(e)[:40]}")
                 continue
         
         sleep_time = 3 + (attempt * 1.5) + random.uniform(0.5, 2.0)
-        logging.warning(f"🔄 [قطعه {index+1}] تمام کارگرها شلوغ/خطا دادند. استراحت به مدت {sleep_time:.1f} ثانیه...")
+        logging.warning(f"🔄 [قطعه {index+1}] کارگرها شلوغ/خطا دادند. استراحت به مدت {sleep_time:.1f} ثانیه...")
         time.sleep(sleep_time)
     
-    # اگر پس از 50 بار تلاش ساخته نشد، ارور برمی‌گرداند تا فرآیند به صورت اصولی متوقف شود
-    logging.error(f"🚫 [قطعه {index+1}] پس از {max_attempts} بار تلاش ساخته نشد.")
-    return index, None
+    # اگر بعد از 50 بار ساخته نشد، ارور صادر میکنه تا پادکست ناقص نمونه
+    raise ValueError(f"قطعه {index+1} پس از 50 بار تلاش ساخته نشد.")
 
 # --- AI PODCAST SCRIPT LOGIC (Cohere Labs Space) ---
 def generate_podcast_in_background(task_id, system_prompt, safety_settings):
@@ -246,7 +246,7 @@ def generate_podcast_in_background(task_id, system_prompt, safety_settings):
     except Exception as e:
         with tasks_lock: tasks[task_id].update({'status': 'failed', 'error': str(e)})
 
-# --- FULL AUTO PODCAST LOGIC ---
+# --- FULL AUTO PODCAST LOGIC (پردازش دسته‌ای 7تایی) ---
 def generate_full_podcast_audio_background(task_id, prompt, speakers):
     try:
         logging.info(f"🚀 [پروژه {task_id}] عملیات ساخت پادکست آغاز شد.")
@@ -255,20 +255,21 @@ def generate_full_podcast_audio_background(task_id, prompt, speakers):
         
         spk_text = "\n".join([f"- {s['id']}: {s['name']}" for s in speakers])
         
-        # حفظ پرامپت قدرتمند برای کوتاه نوشتن و جلوگیری از ارور 500 در ریشه
+        # 🔴 پرامپت دقیقاً منطبق با سایت شد + دستور اکید برای نوشتن فقط به زبان فارسی تا کارگر ارور نده
         system_prompt = f"""Act as a Professional Podcast Producer.
 Topic: "{prompt}"
 Speakers Available:
 {spk_text}
 
-CRITICAL INSTRUCTIONS (MUST FOLLOW STRICTLY):
-1. Create a VERY LONG, in-depth, and highly detailed podcast script.
-2. STRICT RULE: Keep EVERY SINGLE dialogue line SHORT (Under 250 characters).
-3. If a speaker has a lot to say, BREAK their speech into multiple separate, consecutive turns for that SAME speaker. Do NOT write long paragraphs.
-4. NO stage directions, NO emojis, NO brackets like [laugh] or (sigh). Plain spoken text ONLY.
+CRITICAL INSTRUCTION: You must create a VERY LONG, in-depth, and highly detailed podcast script.
+- Do NOT write a short summary.
+- The conversation must deeply explore the topic. Make it a very long and detailed conversation, as long as necessary to fully cover the topic.
+- Make the dialogue engaging and informative.
+- 🔴 IMPORTANT: Write the ENTIRE script ONLY in Persian (Farsi) language. Do NOT use English characters.
 
 Output ONLY valid JSON.
-Format: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1", "dialogue": "..."}}]}}"""
+Format: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1", "dialogue": "..."}}]}}
+Dialogue rules: No stage directions like [laugh], (sigh). Just spoken words."""
 
         AYA_SPACE_URL = "https://coherelabs-aya-expanse.hf.space/gradio_api"
         MAX_ATTEMPTS = 50
@@ -319,13 +320,14 @@ Format: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1",
         for t in data["script"]:
             if "dialogue" in t: t["dialogue"] = re.sub(r'\[.*?\]|\(.*?\)', '', t["dialogue"]).strip()
 
-        script_turns = data["script"]
+        # فقط نوبت‌هایی که متن دارن رو پردازش می‌کنیم
+        script_turns = [t for t in data.get("script", []) if str(t.get("dialogue", "")).strip()]
         total_turns = len(script_turns)
         
-        logging.info(f"✅ سناریو با موفقیت استخراج شد! شامل {total_turns} نوبت کوتاه و استاندارد.")
+        logging.info(f"✅ سناریو استخراج شد! شامل {total_turns} نوبت گفتگو.")
         
         with tasks_lock:
-            tasks[task_id] = {'status': 'generating_audio', 'progress': f'در حال تولید همزمان صداها (0 از {total_turns} تکمیل شده)'}
+            tasks[task_id] = {'status': 'generating_audio', 'progress': f'در حال تولید صداها (0 از {total_turns} تکمیل شده)'}
         
         audio_results = [None] * total_turns
         completed_count = 0
@@ -339,22 +341,29 @@ Format: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1",
                 raise ValueError(f"خطا در تولید صدای نوبت {index+1}")
             return idx, audio_seg
 
-        logging.info("⚡ شروع پروسه ساخت موازی (ارسال تمام قطعات به صورت همزمان)...")
-        max_concurrent_workers = min(3, total_turns)
-        with concurrent.futures.ThreadPoolExecutor(max_workers=max_concurrent_workers) as executor:
-            future_to_index = {executor.submit(process_single_chunk, i, turn): i for i, turn in enumerate(script_turns)}
+        # 🌟 استراتژی طلایی دسته‌بندی (7 تا 7 تا) که پیشنهاد خودت بود 🌟
+        batch_size = 7
+        batches = [script_turns[i:i + batch_size] for i in range(0, total_turns, batch_size)]
+        
+        for batch_index, batch in enumerate(batches):
+            logging.info(f"🔄 شروع پردازش دسته‌ی {batch_index+1} از {len(batches)} (شامل {len(batch)} دیالوگ)...")
             
-            for future in concurrent.futures.as_completed(future_to_index):
-                try:
-                    idx, audio_seg = future.result()
-                    if audio_seg is not None:
+            with concurrent.futures.ThreadPoolExecutor(max_workers=len(batch)) as executor:
+                start_idx = batch_index * batch_size
+                future_to_index = {executor.submit(process_single_chunk, start_idx + i, turn): start_idx + i for i, turn in enumerate(batch)}
+                
+                for future in concurrent.futures.as_completed(future_to_index):
+                    try:
+                        idx, audio_seg = future.result()
                         audio_results[idx] = audio_seg
-                    completed_count += 1
-                    with tasks_lock:
-                        tasks[task_id]['progress'] = f'در حال تولید صداها ({completed_count} از {total_turns} تکمیل شده)'
-                except Exception as exc:
-                    logging.error(f"❌ خطای بحرانی در پردازش قطعه: {str(exc)}")
-                    raise Exception(str(exc))
+                        completed_count += 1
+                        with tasks_lock:
+                            tasks[task_id]['progress'] = f'در حال تولید صداها ({completed_count} از {total_turns} تکمیل شده)'
+                    except Exception as exc:
+                        logging.error(f"❌ خطای بحرانی: {str(exc)}")
+                        raise Exception(str(exc))
+            
+            logging.info(f"✔️ دسته‌ی {batch_index+1} با موفقیت تمام شد.")
 
         logging.info("✂️ تمام قطعات ساخته شد. در حال ادغام (میکس) فایل‌ها...")
         combined_audio = AudioSegment.empty()
@@ -492,7 +501,8 @@ def create_full_podcast():
         
         safety = [{"category": c, "threshold": "BLOCK_NONE"} for c in ["HARM_CATEGORY_HARASSMENT", "HARM_CATEGORY_HATE_SPEECH", "HARM_CATEGORY_SEXUALLY_EXPLICIT", "HARM_CATEGORY_DANGEROUS_CONTENT"]]
         spk_text = "\n".join([f"- {s['id']}: {s['name']}" for s in speakers])
-        sys_prompt = f"""Act as a Professional Podcast Producer.\nTopic: "{prompt}"\nSpeakers Available:\n{spk_text}\nCRITICAL INSTRUCTION: You must create a VERY LONG, in-depth, and highly detailed podcast script.\n- Do NOT write a short summary.\n- The conversation must deeply explore the topic.\n- Make the dialogue engaging and informative.\n\nOutput ONLY valid JSON.\nFormat: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1", "dialogue": "..."}}]}}\nDialogue rules: No stage directions like [laugh], (sigh). Just spoken words."""
+        
+        sys_prompt = f"""Act as a Professional Podcast Producer.\nTopic: "{prompt}"\nSpeakers Available:\n{spk_text}\nCRITICAL INSTRUCTION: You must create a VERY LONG, in-depth, and highly detailed podcast script.\n- Do NOT write a short summary.\n- The conversation must deeply explore the topic.\n- Make the dialogue engaging and informative.\n- 🔴 IMPORTANT: Write the ENTIRE script ONLY in Persian (Farsi) language. Do NOT use English characters.\n\nOutput ONLY valid JSON.\nFormat: {{"selected_speakers": ["id1", "id2"], "script": [{{"speaker_id": "id1", "dialogue": "..."}}]}}\nDialogue rules: No stage directions like [laugh], (sigh). Just spoken words."""
         
         threading.Thread(target=generate_podcast_in_background, args=(task_id, sys_prompt, safety)).start()
         return jsonify({"task_id": task_id}), 202