fix flow

app.py

@@ -189,26 +189,85 @@ def preprocess_audio_ffmpeg(audio_data: bytes, target_sr: int = 16000) -> np.nda
         logger.error(f"FFmpeg preprocessing failed: {e}")
         raise HTTPException(status_code=400, detail="Audio preprocessing failed. Ensure ffmpeg is installed.")
 
+def _score_transcription_quality(text: str) -> float:
+    if not text or not text.strip():
+        return 0.0
+    
+    text_lower = text.lower()
+    score = 0.0
+    
+    if len(text.strip()) > 3:
+        score += 0.3
+    
+    if any(char.isalpha() for char in text):
+        score += 0.2
+    
+    if len(text.split()) > 1:
+        score += 0.2
+    
+    if not any(char in text for char in "[]{}()"):
+        score += 0.1
+    
+    if not text.endswith("..."):
+        score += 0.1
+    
+    if len(text.strip()) > 10:
+        score += 0.1
+    
+    return min(score, 1.0)
+
 def speech_to_text(audio_data: bytes) -> str:
     audio_array = preprocess_audio_ffmpeg(audio_data)
     
-    igbo_result = _get_igbo_asr()
-    if igbo_result[0] is not None and igbo_result[1] is not None:
-        igbo_model, igbo_proc = igbo_result
-        igbo_text = _run_whisper(igbo_model, igbo_proc, audio_array, language="igbo")
-        if igbo_text and igbo_text.strip():
-            logger.info("Using Igbo ASR result")
-            return igbo_text
+    mms_text = ""
+    igbo_text = ""
     
     mms_result = _get_mms()
     if mms_result and mms_result[0] is not None and mms_result[1] is not None:
         mms_model, mms_proc = mms_result
         mms_text = _run_mms(mms_model, mms_proc, audio_array)
-        if mms_text and mms_text.strip():
-            logger.info("Using MMS ASR result")
-            return mms_text
+        logger.info(f"MMS result: '{mms_text}'")
+    
+    igbo_result = _get_igbo_asr()
+    if igbo_result[0] is not None and igbo_result[1] is not None:
+        igbo_model, igbo_proc = igbo_result
+        igbo_text = _run_whisper(igbo_model, igbo_proc, audio_array, language="igbo")
+        logger.info(f"Igbo ASR result: '{igbo_text}'")
+    
+    if not mms_text and not igbo_text:
+        return ""
+    
+    if not mms_text:
+        logger.info("Using Igbo ASR result (MMS failed)")
+        return igbo_text
+    
+    if not igbo_text:
+        logger.info("Using MMS ASR result (Igbo ASR failed)")
+        return mms_text
+    
+    mms_score = _score_transcription_quality(mms_text)
+    igbo_score = _score_transcription_quality(igbo_text)
     
-    return ""
+    mms_lang = detect_language(mms_text)
+    igbo_lang = detect_language(igbo_text)
+    
+    logger.info(f"MMS: '{mms_text}' (score: {mms_score:.2f}, lang: {mms_lang})")
+    logger.info(f"Igbo: '{igbo_text}' (score: {igbo_score:.2f}, lang: {igbo_lang})")
+    
+    if igbo_lang == "ig" and mms_lang != "ig":
+        logger.info("Using Igbo ASR result (detected Igbo language)")
+        return igbo_text
+    
+    if mms_lang == "ig" and igbo_lang != "ig":
+        logger.info("Using MMS ASR result (Igbo ASR didn't detect Igbo)")
+        return mms_text
+    
+    if igbo_score > mms_score + 0.1:
+        logger.info("Using Igbo ASR result (higher quality score)")
+        return igbo_text
+    else:
+        logger.info("Using MMS ASR result (higher quality score)")
+        return mms_text
 
 
 def get_ai_response(text: str, response_language: str = None) -> str:
@@ -235,11 +294,26 @@ def get_ai_response(text: str, response_language: str = None) -> str:
 
 HAUSA_WORDS = [
     "aikin","manoma","gona","amfanin","yanayi","tsaba","fasaha","bisa","noman","shuka",
-    "daji","rani","damina","amfani","bidi'a","noma","bashi","manure","tsiro","gishiri"
+    "daji","rani","damina","amfani","bidi'a","noma","bashi","manure","tsiro","gishiri",
+    "kasa","ruwa","iska","rana","wata","dare","gari","shinkafa","wake","gyada",
+    "aljihu","kudi","kasuwa","gida","makaranta","asibiti","motar","jirgi","keke","doki",
+    "kare","kaza","rago","tunkiya","shanu","kaji","kaza","kifi","kifi","kifi",
+    "abinci","ruwa","shayi","kofi","sugar","gishiri","mai","kayan","miya","tuwo",
+    "fura","koko","kunu","zobo","fura","koko","kunu","zobo","fura","koko",
+    "yaro","yarinya","mutum","mace","yara","mata","maza","dattijo","tsoho","sabon",
+    "babba","karami","tsufa","sabon","kyau","mugun","dadi","daci","mai","kyau",
+    "ina","wane","me","yaushe","yaya","dom","saboda","tare","da","kuma","amma",
+    "ko","idan","sai","har","tun","sai","kafin","bayan","cikin","kan","karkashin",
+    "gaban","bayan","tsakanin","tsaye","kwance","hau","sauka","tashi","zauna","tashi"
 ]
 
-YORUBA_WORDS = [
-    "ilé","ọmọ","òun","awọn","agbẹ","oko","ọgbà","irugbin","àkọsílẹ","omi","ojo","àgbàlá","irọlẹ"
+YORUBA_WORDS = ["ọkọ","aya","baba","iya","egbon","aburo","omo","ebi","ọrẹ","arakunrin",
+    "ọjọ","ọsẹ","osu","ọdun","owuro","ọsan","alẹ","oru","ile","ita",
+    "oja","aso","ounje","ata","isu","iresi","ewa","epo","iyọ","ọti",
+    "nkan","ti","ni","si","lati","ati","pelu","fun","ninu","lori",
+    "abẹ","iwaju","ẹhin","arin","wọle","jade","lo","wa","je","mu",
+    "sọ","ro","kọ","gbo","ri","mọ","fẹ","nifẹ","fẹran","dara",
+    "buburu","tobi","kekere","pupa","funfun","dudu","pọ","diẹ","gbogbo","kan"
 ]
 
 IGBO_WORDS = [
@@ -249,6 +323,22 @@ IGBO_WORDS = [
     "ọrụ","ugbo","mmiri","ala","nri","ahụhụ","ọhụrụ","ncheta","akụkọ","ugwu"
 ]
 
+ENGLISH_WORDS = [
+    "farm","farmer","farming","agriculture","crop","crops","plant","plants","seed","seeds",
+    "soil","water","rain","sun","weather","harvest","field","fields","tractor","plow",
+    "fertilizer","pesticide","organic","yield","yields","grain","wheat","corn","rice",
+    "vegetable","vegetables","fruit","fruits","livestock","cattle","cow","cows","sheep",
+    "goat","goats","chicken","chickens","pig","pigs","milk","meat","eggs","feed",
+    "barn","silo","greenhouse","irrigation","drought","flood","pest","disease","healthy",
+    "growth","mature","ripe","planting","sowing","weeding","pruning","spraying","fertilizing",
+    "cultivation","cultivate","cultivated","arable","fertile","barren","productive","yield",
+    "production","produce","grow","growing","grown","planted","harvested","harvesting",
+    "season","seasons","spring","summer","autumn","winter","climate","temperature","humidity",
+    "moisture","dry","wet","fertile","nutrients","nutrition","feed","feeding","grazing",
+    "pasture","meadow","grass","hay","straw","compost","manure","organic","chemical",
+    "sustainable","sustainability","biodiversity","ecosystem","environment","conservation"
+]
+
 def detect_language(text: str) -> str:
     text_lower = text.lower()
     if any(word in text_lower for word in HAUSA_WORDS):
@@ -257,6 +347,8 @@ def detect_language(text: str) -> str:
         return "yo"
     elif any(word in text_lower for word in IGBO_WORDS):
         return "ig"
+    elif any(word in text_lower for word in ENGLISH_WORDS):
+        return "en"
     lang = detect(text)
     if lang.startswith("ha"):
         return "ha"
@@ -273,8 +365,8 @@ def text_to_speech_file(text: str) -> str:
     
     supported_tts_languages = ["ha", "yo", "en"]
     if lang not in supported_tts_languages:
-        logger.warning(f"Language '{lang}' not supported for TTS, falling back to English")
-        lang = "en"
+        logger.warning(f"Language '{lang}' not supported for TTS, returning text response")
+        raise Exception(f"TTS not available for language '{lang}' - returning text response")
     
     global tts_ig
     if lang == "ha":