refactor: Consolidate cache file loading logic and enhance cache saving for different data types

qwen3-tts-backend/core/cache_manager.py

@@ -45,6 +45,13 @@ class VoiceCacheManager:
     def get_audio_hash(self, audio_data: bytes) -> str:
         return hashlib.sha256(audio_data).hexdigest()
 
+    def _load_cache_file(self, cache_file: Path):
+        if cache_file.suffix == '.pt':
+            import torch
+            return torch.load(cache_file, weights_only=False)
+        with open(cache_file, 'rb') as f:
+            return np.load(f, allow_pickle=False)
+
     async def get_cache(self, user_id: int, ref_audio_hash: str, db: Session) -> Optional[Dict[str, Any]]:
         try:
             cache_entry = get_cache_entry(db, user_id, ref_audio_hash)
@@ -63,8 +70,7 @@ class VoiceCacheManager:
                 logger.warning(f"Cache path out of cache dir: {resolved_cache_file}")
                 return None
 
-            with open(cache_file, 'rb') as f:
-                cache_data = np.load(f, allow_pickle=False)
+            cache_data = self._load_cache_file(cache_file)
 
             logger.info(f"Cache hit: user={user_id}, hash={ref_audio_hash[:8]}..., access_count={cache_entry.access_count}")
             return {
@@ -94,8 +100,7 @@ class VoiceCacheManager:
                 logger.warning(f"Cache path out of cache dir: {resolved_cache_file}")
                 return None
 
-            with open(cache_file, 'rb') as f:
-                cache_data = np.load(f, allow_pickle=False)
+            cache_data = self._load_cache_file(cache_file)
 
             cache_entry.last_accessed = datetime.utcnow()
             cache_entry.access_count += 1
@@ -122,16 +127,19 @@ class VoiceCacheManager:
     ) -> str:
         async with self._lock:
             try:
-                cache_filename = f"{user_id}_{ref_audio_hash}.npy"
-                cache_path = self.cache_dir / cache_filename
-
                 if hasattr(cache_data, "detach"):
                     cache_data = cache_data.detach().cpu().numpy()
-                elif not isinstance(cache_data, np.ndarray):
-                    cache_data = np.asarray(cache_data, dtype=np.float32)
 
-                with open(cache_path, 'wb') as f:
-                    np.save(f, cache_data, allow_pickle=False)
+                if isinstance(cache_data, np.ndarray):
+                    cache_filename = f"{user_id}_{ref_audio_hash}.npy"
+                    cache_path = self.cache_dir / cache_filename
+                    with open(cache_path, 'wb') as f:
+                        np.save(f, cache_data, allow_pickle=False)
+                else:
+                    import torch
+                    cache_filename = f"{user_id}_{ref_audio_hash}.pt"
+                    cache_path = self.cache_dir / cache_filename
+                    torch.save(cache_data, cache_path)
 
                 cache_entry = create_cache_entry(
                     db=db,