Update app.py

app.py

@@ -293,7 +293,15 @@ async def transcribe_voice_with_openai(file_path: str) -> str:
         system_prompt = """
 You are transcribing voice messages for Apex Biotical Veterinary WhatsApp Assistant. This is a professional veterinary products chatbot.
 
-CONTEXT: Users can speak product names, menu selections, numbers, and general queries in English or Urdu.
+IMPORTANT: ONLY TRANSCRIBE ENGLISH OR URDU SPEECH. IGNORE ALL OTHER LANGUAGES.
+
+CONTEXT: Users can speak product names, menu selections, numbers, and general queries in English or Urdu ONLY.
+
+LANGUAGE RESTRICTION: 
+- ONLY English (en) or Urdu (ur) are allowed
+- If you detect any other language, force it to English
+- Never transcribe in German, French, Spanish, Italian, or any other language
+- Always assume English or Urdu speech patterns
 
 PRODUCT NAMES (Veterinary Products):
 - Hydropex (electrolyte supplement)
@@ -339,13 +347,15 @@ English: search, browse, download, catalog, contact, availability, main menu, op
 Urdu: تلاش, براؤز, ڈاؤن لوڈ, کیٹلاگ, رابطہ, دستیابی, مین مینو, آپشن, نمبر, اختیار
 
 TRANSCRIPTION RULES:
-1. Transcribe product names exactly as listed above
-2. Convert spoken numbers to digits (1, 2, 3, etc.)
-3. Handle both English and Urdu speech
-4. Preserve exact spelling for product names
-5. Convert menu selections to numbers
-6. Handle common transcription errors (opium->option, numara->number)
-7. Maintain context for veterinary domain
+1. ONLY transcribe English or Urdu speech
+2. Transcribe product names exactly as listed above
+3. Convert spoken numbers to digits (1, 2, 3, etc.)
+4. Handle both English and Urdu speech
+5. Preserve exact spelling for product names
+6. Convert menu selections to numbers
+7. Handle common transcription errors (opium->option, numara->number)
+8. Maintain context for veterinary domain
+9. If unsure about language, default to English
 
 EXAMPLES:
 - "hydropex" -> "hydropex"
@@ -3408,13 +3418,14 @@ async def handle_voice_message_complete(from_number: str, msg: dict):
             logger.info(f"[Voice] Applied corrections: '{transcribed_text}' -> '{corrected_text}'")
             transcribed_text = corrected_text
         
-        # Detect language of transcribed text
+        # Detect language of transcribed text - FORCE ENGLISH OR URDU ONLY
         detected_lang = 'en'  # Default to English
         try:
             detected_lang = detect(transcribed_text)
-            logger.info(f"[Voice] Detected language: {detected_lang}")
+            logger.info(f"[Voice] Raw detected language: {detected_lang}")
             
-            # Map language codes to supported languages
+            # FORCE LANGUAGE TO ENGLISH OR URDU ONLY
+            # Map all languages to either English or Urdu
             lang_mapping = {
                 'ur': 'ur',  # Urdu
                 'ar': 'ur',  # Arabic (treat as Urdu for Islamic greetings)
@@ -3422,9 +3433,18 @@ async def handle_voice_message_complete(from_number: str, msg: dict):
                 'hi': 'ur',  # Hindi (treat as Urdu)
                 'bn': 'ur',  # Bengali (treat as Urdu)
                 'pa': 'ur',  # Punjabi (treat as Urdu)
-                'id': 'ur',  # Indonesian (often misdetected for Urdu/Arabic)
-                'ms': 'ur',  # Malay (often misdetected for Urdu/Arabic)
-                'tr': 'ur',  # Turkish (often misdetected for Urdu/Arabic)
+                'id': 'ur',  # Indonesian (treat as Urdu)
+                'ms': 'ur',  # Malay (treat as Urdu)
+                'tr': 'ur',  # Turkish (treat as Urdu)
+                'de': 'en',  # German -> English
+                'fr': 'en',  # French -> English
+                'es': 'en',  # Spanish -> English
+                'it': 'en',  # Italian -> English
+                'pt': 'en',  # Portuguese -> English
+                'ru': 'en',  # Russian -> English
+                'ja': 'en',  # Japanese -> English
+                'ko': 'en',  # Korean -> English
+                'zh': 'en',  # Chinese -> English
             }
             
             # Check if text contains Urdu/Arabic characters or Islamic greetings
@@ -3438,15 +3458,17 @@ async def handle_voice_message_complete(from_number: str, msg: dict):
                 detected_lang = 'ur'
                 logger.info(f"[Voice] Overriding language detection to Urdu due to Arabic/Urdu characters or Islamic greeting")
             
+            # Force language to English or Urdu only
             reply_language = lang_mapping.get(detected_lang, 'en')
-            logger.info(f"[Voice] Language '{detected_lang}' mapped to: {reply_language}")
+            logger.info(f"[Voice] Language '{detected_lang}' FORCED to: {reply_language}")
+            
+            # Additional safety check - if still not English or Urdu, force to English
+            if reply_language not in ['en', 'ur']:
+                logger.warning(f"[Voice] Language '{reply_language}' not in allowed list, forcing to English")
+                reply_language = 'en'
             
         except Exception as e:
-            logger.warning(f"[Voice] Language detection failed: {e}")
-            reply_language = 'en'
-        
-        if reply_language not in ['en', 'ur']:
-            logger.info(f"[Voice] Language '{reply_language}' not supported, defaulting to English")
+            logger.warning(f"[Voice] Language detection failed: {e}, defaulting to English")
             reply_language = 'en'
         
         # For Urdu voice notes, translate to English for processing