diff --git a/src/transformers/processing_utils.py b/src/transformers/processing_utils.py
index bb1344a43dcf..31360883dbf3 100644
--- a/src/transformers/processing_utils.py
+++ b/src/transformers/processing_utils.py
@@ -17,6 +17,7 @@
 
 import bisect
 import copy
+import functools
 import inspect
 import json
 import os
@@ -571,6 +572,15 @@ def __getitem__(self, key):
         raise AttributeError(f"{self.__class__.__name__} has no attribute {key}")
 
 
+@functools.lru_cache(maxsize=8)
+def _merge_typed_dict(preprocessor_typed_dict: type, modality_typed_dict: type) -> type:
+    return TypedDict(
+        "merged_typed_dict",
+        {**preprocessor_typed_dict.__annotations__, **modality_typed_dict.__annotations__},
+        total=False,
+    )
+
+
 class ProcessorMixin(PushToHubMixin):
     """
     This is a mixin used to provide saving/loading functionality for all processor classes.
@@ -1369,11 +1379,7 @@ class MyProcessingKwargs(ProcessingKwargs, CommonKwargs, TextKwargs, ImagesKwarg
                 if preprocessor is None or getattr(preprocessor, "valid_kwargs", None) is None:
                     continue
                 preprocessor_typed_dict_obj = getattr(preprocessor, "valid_kwargs")
-                typed_dict_obj = TypedDict(
-                    "merged_typed_dict",
-                    {**preprocessor_typed_dict_obj.__annotations__, **typed_dict_obj.__annotations__},
-                    total=False,
-                )
+                typed_dict_obj = _merge_typed_dict(preprocessor_typed_dict_obj, typed_dict_obj)
             validate_typed_dict(typed_dict_obj, output_kwargs[key])
         return output_kwargs