Update app.py

app.py

@@ -1,7 +1,7 @@
 """
 VoiceAura Translation API
 Models:
-  1. SLPG/English_to_Urdu_Unsupervised_MT            (en    → ur)
+  1. SLPG/English_to_Urdu_Unsupervised_MT              (en    → ur)
   2. SLPG/Punjabi_Shahmukhi_to_Gurmukhi_Transliteration (pa-s → pa-g)
   3. SLPG/Punjabi_Gurmukhi_to_Shahmukhi_Transliteration (pa-g → pa-s)
 """
@@ -27,7 +27,7 @@ app.add_middleware(
     allow_headers=["*"],
 )
 
-# ── Model configs ────────────────────────────────────────
+# ── Model configs ─────────────────────────────────────────
 MODELS_CONFIG = {
     "en-ur": {
         "files": {
@@ -64,7 +64,7 @@ MODELS_CONFIG = {
     },
 }
 
-# ── Helpers ──────────────────────────────────────────────
+# ── Helpers ───────────────────────────────────────────────
 def download_file(url: str, path: str):
     if os.path.exists(path):
         print(f"[✓] Exists: {path}")
@@ -79,8 +79,16 @@ def download_file(url: str, path: str):
     print(f"[✓] Done: {path}")
 
 def detokenize(text: str) -> str:
-    """Remove fairseq BPE tokens (▁ symbols)"""
-    return text.replace("▁", "").strip()
+    """
+    Fairseq BPE output fix:
+    '▁ساڈا ▁گھر ▁راولپنڈی ▁وچ ▁ہے' → 'ساڈا گھر راولپنڈی وچ ہے'
+    ▁ = word boundary marker — replace with space, then clean up
+    """
+    # ▁ at start of word = space before it
+    result = text.replace(" ▁", " ").replace("▁", "")
+    # Clean extra spaces
+    result = " ".join(result.split())
+    return result.strip()
 
 def load_model(pair: str):
     cfg = MODELS_CONFIG[pair]
@@ -103,13 +111,13 @@ def load_model(pair: str):
     print(f"[✓] Model ready: {pair}")
     return model
 
-# ── Startup ──────────────────────────────────────────────
+# ── Startup ───────────────────────────────────────────────
 @app.on_event("startup")
 async def startup():
     for pair in MODELS_CONFIG:
         load_model(pair)
 
-# ── API ──────────────────────────────────────────────────
+# ── API ───────────────────────────────────────────────────
 class Req(BaseModel):
     text: str
     from_lang: str = "en"
@@ -133,13 +141,13 @@ def translate(req: Req):
     try:
         cfg    = MODELS_CONFIG[pair]
         model  = load_model(pair)
-        result = model.translate(req.text.strip())
+        raw    = model.translate(req.text.strip())
 
-        # Detokenize if needed (Punjabi models)
-        if cfg["detokenize"]:
-            result = detokenize(result)
+        print(f"[DEBUG] raw output [{pair}]: {raw}")  # debug log
+
+        result = detokenize(raw) if cfg["detokenize"] else raw
+        return {"success": True, "translation": result, "pair": pair, "raw": raw}
 
-        return {"success": True, "translation": result, "pair": pair}
     except Exception as e:
         print(f"Error [{pair}]: {e}")
         return {"success": False, "translation": str(e)}