diff --git a/vibevoice/processor/vibevoice_processor.py b/vibevoice/processor/vibevoice_processor.py
index 66d0a9d..2995b3b 100644
--- a/vibevoice/processor/vibevoice_processor.py
+++ b/vibevoice/processor/vibevoice_processor.py
@@ -56,23 +56,38 @@ class VibeVoiceProcessor:
         """
         import os
         import json
+        from transformers.utils import cached_file
         from .vibevoice_tokenizer_processor import VibeVoiceTokenizerProcessor
         from vibevoice.modular.modular_vibevoice_text_tokenizer import (
             VibeVoiceTextTokenizer, 
             VibeVoiceTextTokenizerFast
         )
         
-        # Load processor configuration
+        # Try to load from local path first, then from HF hub
         config_path = os.path.join(pretrained_model_name_or_path, "preprocessor_config.json")
+        config = None
+        
         if os.path.exists(config_path):
+            # Local path exists
             with open(config_path, 'r') as f:
                 config = json.load(f)
         else:
-            logger.warning(f"No preprocessor_config.json found at {pretrained_model_name_or_path}, using defaults")
-            config = {
-                "speech_tok_compress_ratio": 3200,
-                "db_normalize": True,
-            }
+            # Try to load from HF hub
+            try:
+                config_file = cached_file(
+                    pretrained_model_name_or_path,
+                    "preprocessor_config.json",
+                    **kwargs
+                )
+                with open(config_file, 'r') as f:
+                    config = json.load(f)
+            except Exception as e:
+                logger.warning(f"Could not load preprocessor_config.json from {pretrained_model_name_or_path}: {e}")
+                logger.warning("Using default configuration")
+                config = {
+                    "speech_tok_compress_ratio": 3200,
+                    "db_normalize": True,
+                }
         
         # Extract main processor parameters
         speech_tok_compress_ratio = config.get("speech_tok_compress_ratio", 3200)