Upload main.py

main.py

@@ -1,344 +1,360 @@
-from fastapi import FastAPI, UploadFile, File, HTTPException
-from fastapi.middleware.cors import CORSMiddleware
-from fastapi.responses import JSONResponse
-import torch
-from transformers import pipeline
-import json
-import os
-from difflib import SequenceMatcher
-from typing import Dict, Any, Optional
-import tempfile
-import subprocess
-import shutil
-
-app = FastAPI(
-    title="Bayan AI بيان",
-    description="",
-    version="1.0.0"
-)
-
-app.add_middleware(
-    CORSMiddleware,
-    allow_origins=["*"],  # Allow all origins for local development
-    allow_credentials=True,
-    allow_methods=["*"],
-    allow_headers=["*"],
-)
-
-# CPU only on free tier
-device = -1
-
-# Load Whisper pipeline (model downloads/caches automatically on first run)
-pipe = pipeline(
-    "automatic-speech-recognition",
-    model="tarteel-ai/whisper-tiny-ar-quran",
-    device=device,
-)
-
-# Standard Surah names (1–114)
-surah_names = {
-    1: "Al-Fatiha (الفاتحة)",
-    2: "Al-Baqarah (البقرة)",
-    3: "Aal-E-Imran (آل عمران)",
-    4: "An-Nisa (النساء)",
-    5: "Al-Maidah (المائدة)",
-    6: "Al-An'am (الأنعام)",
-    7: "Al-A'raf (الأعراف)",
-    8: "Al-Anfal (الأنفال)",
-    9: "At-Tawbah (التوبة)",
-    10: "Yunus (يونس)",
-    11: "Hud (هود)",
-    12: "Yusuf (يوسف)",
-    13: "Ar-Ra'd (الرعد)",
-    14: "Ibrahim (إبراهيم)",
-    15: "Al-Hijr (الحجر)",
-    16: "An-Nahl (النحل)",
-    17: "Al-Isra (الإسراء)",
-    18: "Al-Kahf (الكهف)",
-    19: "Maryam (مريم)",
-    20: "Ta-Ha (طه)",
-    21: "Al-Anbiya (الأنبياء)",
-    22: "Al-Hajj (الحج)",
-    23: "Al-Mu'minun (المؤمنون)",
-    24: "An-Nur (النور)",
-    25: "Al-Furqan (الفرقان)",
-    26: "Ash-Shu'ara (الشعراء)",
-    27: "An-Naml (النمل)",
-    28: "Al-Qasas (القصص)",
-    29: "Al-Ankabut (العنكبوت)",
-    30: "Ar-Rum (الروم)",
-    31: "Luqman (لقمان)",
-    32: "As-Sajdah (السجدة)",
-    33: "Al-Ahzab (الأحزاب)",
-    34: "Saba (سبأ)",
-    35: "Fatir (فاطر)",
-    36: "Ya-Sin (يس)",
-    37: "As-Saffat (الصافات)",
-    38: "Sad (ص)",
-    39: "Az-Zumar (الزمر)",
-    40: "Ghafir (غافر)",
-    41: "Fussilat (فصلت)",
-    42: "Ash-Shura (الشورى)",
-    43: "Az-Zukhruf (الزخرف)",
-    44: "Ad-Dukhkhan (الدخان)",
-    45: "Al-Jathiya (الجاثية)",
-    46: "Al-Ahqaf (الأحقاف)",
-    47: "Muhammad (محمد)",
-    48: "Al-Fath (الفتح)",
-    49: "Al-Hujurat (الحجرات)",
-    50: "Qaf (ق)",
-    51: "Adh-Dhariyat (الذاريات)",
-    52: "At-Tur (الطور)",
-    53: "An-Najm (النجم)",
-    54: "Al-Qamar (القمر)",
-    55: "Ar-Rahman (الرحمن)",
-    56: "Al-Waqi'ah (الواقعة)",
-    57: "Al-Hadid (الحديد)",
-    58: "Al-Mujadila (المجادلة)",
-    59: "Al-Hashr (الحشر)",
-    60: "Al-Mumtahina (الممتحنة)",
-    61: "As-Saff (الصف)",
-    62: "Al-Jumu'ah (الجمعة)",
-    63: "Al-Munafiqoon (المنافقون)",
-    64: "At-Taghabun (التغابن)",
-    65: "At-Talaq (الطلاق)",
-    66: "At-Tahrim (التحريم)",
-    67: "Al-Mulk (الملك)",
-    68: "Al-Qalam (القلم)",
-    69: "Al-Haqqah (الحاقة)",
-    70: "Al-Ma'arij (المعارج)",
-    71: "Nooh (نوح)",
-    72: "Al-Jinn (الجن)",
-    73: "Al-Muzzammil (المزمل)",
-    74: "Al-Muddathir (المدثر)",
-    75: "Al-Qiyamah (القيامة)",
-    76: "Al-Insan (الإنسان)",
-    77: "Al-Mursalat (المرسلات)",
-    78: "An-Naba (النبأ)",
-    79: "An-Nazi'at (النازعات)",
-    80: "Abasa (عبس)",
-    81: "At-Takwir (التكوير)",
-    82: "Al-Infitar (الإنفطار)",
-    83: "Al-Mutaffifin (المطففين)",
-    84: "Al-Inshiqaq (الإنشقاق)",
-    85: "Al-Buruj (البروج)",
-    86: "At-Tariq (الطارق)",
-    87: "Al-A'la (الأعلى)",
-    88: "Al-Ghashiyah (الغاشية)",
-    89: "Al-Fajr (الفجر)",
-    90: "Al-Balad (البلد)",
-    91: "Ash-Shams (الشمس)",
-    92: "Al-Lail (الليل)",
-    93: "Ad-Duha (الضحى)",
-    94: "Ash-Sharh (الشرح)",
-    95: "At-Tin (التين)",
-    96: "Al-Alaq (العلق)",
-    97: "Al-Qadr (القدر)",
-    98: "Al-Bayyina (البينة)",
-    99: "Az-Zalzalah (الزلزلة)",
-    100: "Al-Adiyat (العاديات)",
-    101: "Al-Qari'ah (القارعة)",
-    102: "At-Takathur (التكاثر)",
-    103: "Al-Asr (العصر)",
-    104: "Al-Humazah (الهمزة)",
-    105: "Al-Fil (الفيل)",
-    106: "Quraish (قريش)",
-    107: "Al-Ma'un (الماعون)",
-    108: "Al-Kawthar (الكوثر)",
-    109: "Al-Kafirun (الكافرون)",
-    110: "An-Nasr (النصر)",
-    111: "Al-Masad (المسد)",
-    112: "Al-Ikhlas (الإخلاص)",
-    113: "Al-Falaq (الفلق)",
-    114: "An-Nas (الناس)",
-}
-
-import re
-
-def normalize_text(text: str) -> str:
-    """Robust normalization for Arabic text."""
-    text = re.sub(r"[إأآاٱ]", "ا", text)
-    text = re.sub(r"ى", "ي", text)
-    text = re.sub(r"ؤ", "ء", text)
-    text = re.sub(r"ئ", "ء", text)
-    text = re.sub(r"g", "ة", text)
-    text = re.sub(r"ة", "ه", text)
-    text = re.sub(r"[\u064B-\u065F\u0670]", "", text) # Tashkeel
-    text = re.sub(r"[\u06D6-\u06ED]", "", text) 
-    text = re.sub(r"ء", "", text) # Remove Hamza to handle varying forms
-    return " ".join(text.strip().split())
-
-# Pre-load all verses at startup
-all_verses = []
-
-surahs_dir = "surahs_json_files"
-if not os.path.isdir(surahs_dir):
-    raise FileNotFoundError("Missing 'surahs_json_files/' folder.")
-
-for filename in sorted(os.listdir(surahs_dir)):
-    if filename.endswith(".json"):
-        try:
-            surah_number = int(filename.split("_")[0])
-        except:
-            continue
-        surah_name = surah_names.get(surah_number, f"Surah {surah_number}")
-        file_path = os.path.join(surahs_dir, filename)
-        
-        with open(file_path, "r", encoding="utf-8") as f:
-            data = json.load(f)
-        
-        verses = [ayah["text"] for ayah in data.get("ayahs", []) if "text" in ayah]
-        
-        for ayah_number, verse_text in enumerate(verses, start=1):
-            verse_norm = normalize_text(verse_text)
-            all_verses.append({
-                "surah_number": surah_number,
-                "surah_name": surah_name,
-                "ayah_number": ayah_number,
-                "verse_text": verse_text,
-                "verse_norm": verse_norm
-            })
-
-print(f"Loaded {len(all_verses)} verses from {len(os.listdir(surahs_dir))} surahs.")
-
-def find_best_verse(transcription: str) -> Dict[str, Any]:
-    transcription_norm = normalize_text(transcription)
-    if not transcription_norm:
-        return {"error": "Empty transcription"}
-    
-    candidates = []
-    
-    # Pre-compile regex for whole word check
-    pattern_str = r'(?:^|\s)' + re.escape(transcription_norm) + r'(?:\s|$)'
-    whole_word_regex = re.compile(pattern_str)
-
-    for verse in all_verses:
-        verse_norm = verse["verse_norm"]
-        
-        is_whole_word = False
-        containment = 0.0
-        ratio = 0.0
-        
-        # Fast substring check
-        if transcription_norm in verse_norm:
-            containment = 1.0
-            matcher = SequenceMatcher(None, transcription_norm, verse_norm)
-            ratio = matcher.ratio()
-            
-            # Check for whole word match
-            if whole_word_regex.search(verse_norm):
-                is_whole_word = True
-        else:
-            matcher = SequenceMatcher(None, transcription_norm, verse_norm)
-            match = matcher.find_longest_match(0, len(transcription_norm), 0, len(verse_norm))
-            containment = match.size / len(transcription_norm) if len(transcription_norm) > 0 else 0
-            ratio = matcher.ratio()
-        
-        candidates.append({
-            "verse": verse,
-            "containment": containment,
-            "ratio": ratio,
-            "is_whole_word": is_whole_word
-        })
-    
-    # Sort by whole_word (desc), containment (desc), ratio (desc)
-    candidates.sort(key=lambda x: (x["is_whole_word"], x["containment"], x["ratio"]), reverse=True)
-    
-    # If we have whole word matches, ignore partial matches
-    if candidates and candidates[0]["is_whole_word"]:
-        candidates = [c for c in candidates if c["is_whole_word"]]
-    
-    # Filter strong matches (>= 80% containment)
-    strong_matches = [c for c in candidates if c["containment"] >= 0.8]
-    
-    def format_match(candidate):
-        verse_data = candidate["verse"]
-        return {
-            "surah_number": verse_data["surah_number"],
-            "surah_name": verse_data["surah_name"],
-            "ayah_number": verse_data["ayah_number"],
-            "verse_text": verse_data["verse_text"],
-            "similarity_score": round(candidate["containment"], 4)
-        }
-
-    if not strong_matches:
-        # No strong matches found
-        if candidates:
-            top_match = candidates[0]
-            return {
-                "error": "No confident match found",
-                "best_similarity": round(top_match["containment"], 4),
-                "possible_match": format_match(top_match)
-            }
-        else:
-            return {"error": "No matches found"}
-
-    if len(strong_matches) > 1:
-        # Multiple strong matches -> return top 5
-        top_5 = strong_matches[:5]
-        return {
-            "matches": [format_match(m) for m in top_5]
-        }
-    else:
-        # Single dominant match
-        return format_match(strong_matches[0])
-
-@app.get("/")
-def root():
-    return {"message": "Bayan AI بيان... LIVE!"}
-
-@app.post("/recognize")
-async def recognize(file: UploadFile = File(...)):
-    # Allow both audio and video
-    is_video = file.content_type and file.content_type.startswith("video/")
-    is_audio = file.content_type and file.content_type.startswith("audio/")
-    
-    if not is_audio and not is_video:
-        raise HTTPException(status_code=400, detail="File must be an audio or video file")
-    
-    # Save to temp file
-    contents = await file.read()
-    file_extension = os.path.splitext(file.filename)[1] or (".mp4" if is_video else ".wav")
-    
-    with tempfile.NamedTemporaryFile(delete=False, suffix=file_extension) as tmp:
-        tmp.write(contents)
-        input_path = tmp.name
-    
-    audio_path = input_path
-    temp_audio_path = None
-    
-    try:
-        if is_video:
-            # Check if ffmpeg is installed
-            if not shutil.which("ffmpeg"):
-                raise HTTPException(status_code=500, detail="ffmpeg not found on server")
-            
-            temp_audio_path = input_path + "_converted.wav"
-            # Extract audio quickly and silently
-            # -vn: no video, -acodec pcm_s16le: wav format, -ar 16000: whisper preferred sample rate
-            # -y: overwrite, -loglevel error: be silent
-            cmd = [
-                "ffmpeg", "-y", "-i", input_path,
-                "-vn", "-acodec", "pcm_s16le", "-ar", "16000", "-ac", "1",
-                "-loglevel", "error",
-                temp_audio_path
-            ]
-            subprocess.run(cmd, check=True)
-            audio_path = temp_audio_path
-            
-        transcription = pipe(audio_path)["text"]
-    except subprocess.CalledProcessError as e:
-        raise HTTPException(status_code=500, detail=f"Video conversion error: {str(e)}")
-    except Exception as e:
-        raise HTTPException(status_code=500, detail=f"Transcription error: {str(e)}")
-    finally:
-        # Clean up all temp files
-        if os.path.exists(input_path):
-            os.unlink(input_path)
-        if temp_audio_path and os.path.exists(temp_audio_path):
-            os.unlink(temp_audio_path)
-    
-    result = find_best_verse(transcription)
-    result["transcription"] = transcription
-    
+from fastapi import FastAPI, UploadFile, File, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import JSONResponse
+import torch
+from transformers import pipeline
+import json
+import os
+from difflib import SequenceMatcher
+from typing import Dict, Any, Optional
+import tempfile
+import subprocess
+import shutil
+
+app = FastAPI(
+    title="Bayan AI بيان",
+    description="",
+    version="1.0.0"
+)
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],  # Allow all origins for local development
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# CPU only on free tier
+device = -1
+
+# Load Whisper pipeline (model downloads/caches automatically on first run)
+pipe = pipeline(
+    "automatic-speech-recognition",
+    model="tarteel-ai/whisper-tiny-ar-quran",
+    device=device,
+)
+
+# Standard Surah names (1–114)
+surah_names = {
+    1: "Al-Fatiha (الفاتحة)",
+    2: "Al-Baqarah (البقرة)",
+    3: "Aal-E-Imran (آل عمران)",
+    4: "An-Nisa (النساء)",
+    5: "Al-Maidah (المائدة)",
+    6: "Al-An'am (الأنعام)",
+    7: "Al-A'raf (الأعراف)",
+    8: "Al-Anfal (الأنفال)",
+    9: "At-Tawbah (التوبة)",
+    10: "Yunus (يونس)",
+    11: "Hud (هود)",
+    12: "Yusuf (يوسف)",
+    13: "Ar-Ra'd (الرعد)",
+    14: "Ibrahim (إبراهيم)",
+    15: "Al-Hijr (الحجر)",
+    16: "An-Nahl (النحل)",
+    17: "Al-Isra (الإسراء)",
+    18: "Al-Kahf (الكهف)",
+    19: "Maryam (مريم)",
+    20: "Ta-Ha (طه)",
+    21: "Al-Anbiya (الأنبياء)",
+    22: "Al-Hajj (الحج)",
+    23: "Al-Mu'minun (المؤمنون)",
+    24: "An-Nur (النور)",
+    25: "Al-Furqan (الفرقان)",
+    26: "Ash-Shu'ara (الشعراء)",
+    27: "An-Naml (النمل)",
+    28: "Al-Qasas (القصص)",
+    29: "Al-Ankabut (العنكبوت)",
+    30: "Ar-Rum (الروم)",
+    31: "Luqman (لقمان)",
+    32: "As-Sajdah (السجدة)",
+    33: "Al-Ahzab (الأحزاب)",
+    34: "Saba (سبأ)",
+    35: "Fatir (فاطر)",
+    36: "Ya-Sin (يس)",
+    37: "As-Saffat (الصافات)",
+    38: "Sad (ص)",
+    39: "Az-Zumar (الزمر)",
+    40: "Ghafir (غافر)",
+    41: "Fussilat (فصلت)",
+    42: "Ash-Shura (الشورى)",
+    43: "Az-Zukhruf (الزخرف)",
+    44: "Ad-Dukhkhan (الدخان)",
+    45: "Al-Jathiya (الجاثية)",
+    46: "Al-Ahqaf (الأحقاف)",
+    47: "Muhammad (محمد)",
+    48: "Al-Fath (الفتح)",
+    49: "Al-Hujurat (الحجرات)",
+    50: "Qaf (ق)",
+    51: "Adh-Dhariyat (الذاريات)",
+    52: "At-Tur (الطور)",
+    53: "An-Najm (النجم)",
+    54: "Al-Qamar (القمر)",
+    55: "Ar-Rahman (الرحمن)",
+    56: "Al-Waqi'ah (الواقعة)",
+    57: "Al-Hadid (الحديد)",
+    58: "Al-Mujadila (المجادلة)",
+    59: "Al-Hashr (الحشر)",
+    60: "Al-Mumtahina (الممتحنة)",
+    61: "As-Saff (الصف)",
+    62: "Al-Jumu'ah (الجمعة)",
+    63: "Al-Munafiqoon (المنافقون)",
+    64: "At-Taghabun (التغابن)",
+    65: "At-Talaq (الطلاق)",
+    66: "At-Tahrim (التحريم)",
+    67: "Al-Mulk (الملك)",
+    68: "Al-Qalam (القلم)",
+    69: "Al-Haqqah (الحاقة)",
+    70: "Al-Ma'arij (المعارج)",
+    71: "Nooh (نوح)",
+    72: "Al-Jinn (الجن)",
+    73: "Al-Muzzammil (المزمل)",
+    74: "Al-Muddathir (المدثر)",
+    75: "Al-Qiyamah (القيامة)",
+    76: "Al-Insan (الإنسان)",
+    77: "Al-Mursalat (المرسلات)",
+    78: "An-Naba (النبأ)",
+    79: "An-Nazi'at (النازعات)",
+    80: "Abasa (عبس)",
+    81: "At-Takwir (التكوير)",
+    82: "Al-Infitar (الإنفطار)",
+    83: "Al-Mutaffifin (المطففين)",
+    84: "Al-Inshiqaq (الإنشقاق)",
+    85: "Al-Buruj (البروج)",
+    86: "At-Tariq (الطارق)",
+    87: "Al-A'la (الأعلى)",
+    88: "Al-Ghashiyah (الغاشية)",
+    89: "Al-Fajr (الفجر)",
+    90: "Al-Balad (البلد)",
+    91: "Ash-Shams (الشمس)",
+    92: "Al-Lail (الليل)",
+    93: "Ad-Duha (الضحى)",
+    94: "Ash-Sharh (الشرح)",
+    95: "At-Tin (التين)",
+    96: "Al-Alaq (العلق)",
+    97: "Al-Qadr (القدر)",
+    98: "Al-Bayyina (البينة)",
+    99: "Az-Zalzalah (الزلزلة)",
+    100: "Al-Adiyat (العاديات)",
+    101: "Al-Qari'ah (القارعة)",
+    102: "At-Takathur (التكاثر)",
+    103: "Al-Asr (العصر)",
+    104: "Al-Humazah (الهمزة)",
+    105: "Al-Fil (الفيل)",
+    106: "Quraish (قريش)",
+    107: "Al-Ma'un (الماعون)",
+    108: "Al-Kawthar (الكوثر)",
+    109: "Al-Kafirun (الكافرون)",
+    110: "An-Nasr (النصر)",
+    111: "Al-Masad (المسد)",
+    112: "Al-Ikhlas (الإخلاص)",
+    113: "Al-Falaq (الفلق)",
+    114: "An-Nas (الناس)",
+}
+
+# Phrases to ignore (e.g., common introductions)
+PHRASES_TO_IGNORE = [
+    "بِسْمِ اللَّهِ الرَّحْمَنِ الرَّحِيمِ",
+    "أعوذ بالله من الشيطان الرجيم",
+    "صدق الله العظيم",
+]
+
+import re
+
+def normalize_text(text: str) -> str:
+    """Robust normalization for Arabic text."""
+    text = re.sub(r"[إأآاٱ]", "ا", text)
+    text = re.sub(r"ى", "ي", text)
+    text = re.sub(r"ؤ", "ء", text)
+    text = re.sub(r"ئ", "ء", text)
+    text = re.sub(r"g", "ة", text)
+    text = re.sub(r"ة", "ه", text)
+    text = re.sub(r"[\u064B-\u065F\u0670]", "", text) # Tashkeel
+    text = re.sub(r"[\u06D6-\u06ED]", "", text) 
+    text = re.sub(r"ء", "", text) # Remove Hamza to handle varying forms
+    return " ".join(text.strip().split())
+
+# Pre-load all verses at startup
+all_verses = []
+
+surahs_dir = "surahs_json_files"
+if not os.path.isdir(surahs_dir):
+    raise FileNotFoundError("Missing 'surahs_json_files/' folder.")
+
+for filename in sorted(os.listdir(surahs_dir)):
+    if filename.endswith(".json"):
+        try:
+            surah_number = int(filename.split("_")[0])
+        except:
+            continue
+        surah_name = surah_names.get(surah_number, f"Surah {surah_number}")
+        file_path = os.path.join(surahs_dir, filename)
+        
+        with open(file_path, "r", encoding="utf-8") as f:
+            data = json.load(f)
+        
+        verses = [ayah["text"] for ayah in data.get("ayahs", []) if "text" in ayah]
+        
+        for ayah_number, verse_text in enumerate(verses, start=1):
+            verse_norm = normalize_text(verse_text)
+            all_verses.append({
+                "surah_number": surah_number,
+                "surah_name": surah_name,
+                "ayah_number": ayah_number,
+                "verse_text": verse_text,
+                "verse_norm": verse_norm
+            })
+
+print(f"Loaded {len(all_verses)} verses from {len(os.listdir(surahs_dir))} surahs.")
+
+def find_best_verse(transcription: str) -> Dict[str, Any]:
+    transcription_norm = normalize_text(transcription)
+    
+    # Remove phrases to ignore (Bismillah, A'udhu billah)
+    for phrase in PHRASES_TO_IGNORE:
+        phrase_norm = normalize_text(phrase)
+        if phrase_norm in transcription_norm:
+            # Replace and clean up extra spaces
+            transcription_norm = transcription_norm.replace(phrase_norm, "").strip()
+            transcription_norm = " ".join(transcription_norm.split())
+
+    if not transcription_norm:
+        return {"error": "Empty transcription"}
+    
+    candidates = []
+    
+    # Pre-compile regex for whole word check
+    pattern_str = r'(?:^|\s)' + re.escape(transcription_norm) + r'(?:\s|$)'
+    whole_word_regex = re.compile(pattern_str)
+
+    for verse in all_verses:
+        verse_norm = verse["verse_norm"]
+        
+        is_whole_word = False
+        containment = 0.0
+        ratio = 0.0
+        
+        # Fast substring check
+        if transcription_norm in verse_norm:
+            containment = 1.0
+            matcher = SequenceMatcher(None, transcription_norm, verse_norm)
+            ratio = matcher.ratio()
+            
+            # Check for whole word match
+            if whole_word_regex.search(verse_norm):
+                is_whole_word = True
+        else:
+            matcher = SequenceMatcher(None, transcription_norm, verse_norm)
+            match = matcher.find_longest_match(0, len(transcription_norm), 0, len(verse_norm))
+            containment = match.size / len(transcription_norm) if len(transcription_norm) > 0 else 0
+            ratio = matcher.ratio()
+        
+        candidates.append({
+            "verse": verse,
+            "containment": containment,
+            "ratio": ratio,
+            "is_whole_word": is_whole_word
+        })
+    
+    # Sort by whole_word (desc), containment (desc), ratio (desc)
+    candidates.sort(key=lambda x: (x["is_whole_word"], x["containment"], x["ratio"]), reverse=True)
+    
+    # If we have whole word matches, ignore partial matches
+    if candidates and candidates[0]["is_whole_word"]:
+        candidates = [c for c in candidates if c["is_whole_word"]]
+    
+    # Filter strong matches (>= 80% containment)
+    strong_matches = [c for c in candidates if c["containment"] >= 0.8]
+    
+    def format_match(candidate):
+        verse_data = candidate["verse"]
+        return {
+            "surah_number": verse_data["surah_number"],
+            "surah_name": verse_data["surah_name"],
+            "ayah_number": verse_data["ayah_number"],
+            "verse_text": verse_data["verse_text"],
+            "similarity_score": round(candidate["containment"], 4)
+        }
+
+    if not strong_matches:
+        # No strong matches found
+        if candidates:
+            top_match = candidates[0]
+            return {
+                "error": "No confident match found",
+                "best_similarity": round(top_match["containment"], 4),
+                "possible_match": format_match(top_match)
+            }
+        else:
+            return {"error": "No matches found"}
+
+    if len(strong_matches) > 1:
+        # Multiple strong matches -> return top 5
+        top_5 = strong_matches[:5]
+        return {
+            "matches": [format_match(m) for m in top_5]
+        }
+    else:
+        # Single dominant match
+        return format_match(strong_matches[0])
+
+@app.get("/")
+def root():
+    return {"message": "Bayan AI بيان... LIVE!"}
+
+@app.post("/recognize")
+async def recognize(file: UploadFile = File(...)):
+    # Allow both audio and video
+    is_video = file.content_type and file.content_type.startswith("video/")
+    is_audio = file.content_type and file.content_type.startswith("audio/")
+    
+    if not is_audio and not is_video:
+        raise HTTPException(status_code=400, detail="File must be an audio or video file")
+    
+    # Save to temp file
+    contents = await file.read()
+    file_extension = os.path.splitext(file.filename)[1] or (".mp4" if is_video else ".wav")
+    
+    with tempfile.NamedTemporaryFile(delete=False, suffix=file_extension) as tmp:
+        tmp.write(contents)
+        input_path = tmp.name
+    
+    audio_path = input_path
+    temp_audio_path = None
+    
+    try:
+        if is_video:
+            # Check if ffmpeg is installed
+            if not shutil.which("ffmpeg"):
+                raise HTTPException(status_code=500, detail="ffmpeg not found on server")
+            
+            temp_audio_path = input_path + "_converted.wav"
+            # Extract audio quickly and silently
+            # -vn: no video, -acodec pcm_s16le: wav format, -ar 16000: whisper preferred sample rate
+            # -y: overwrite, -loglevel error: be silent
+            cmd = [
+                "ffmpeg", "-y", "-i", input_path,
+                "-vn", "-acodec", "pcm_s16le", "-ar", "16000", "-ac", "1",
+                "-loglevel", "error",
+                temp_audio_path
+            ]
+            subprocess.run(cmd, check=True)
+            audio_path = temp_audio_path
+            
+        transcription = pipe(audio_path)["text"]
+    except subprocess.CalledProcessError as e:
+        raise HTTPException(status_code=500, detail=f"Video conversion error: {str(e)}")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Transcription error: {str(e)}")
+    finally:
+        # Clean up all temp files
+        if os.path.exists(input_path):
+            os.unlink(input_path)
+        if temp_audio_path and os.path.exists(temp_audio_path):
+            os.unlink(temp_audio_path)
+    
+    result = find_best_verse(transcription)
+    result["transcription"] = transcription
+    
     return JSONResponse(content=result)