diff --git a/litellm/main.py b/litellm/main.py
index ddd37b47536..5356c6f3642 100644
--- a/litellm/main.py
+++ b/litellm/main.py
@@ -4797,7 +4797,9 @@ def embedding(  # noqa: PLR0915
             }
         )
 
-    litellm_params_dict = get_litellm_params(**kwargs)
+    litellm_params_dict = get_litellm_params(
+        custom_llm_provider=custom_llm_provider, **kwargs
+    )
 
     logging: LiteLLMLoggingObj = litellm_logging_obj  # type: ignore
     logging.update_environment_variables(
@@ -6488,7 +6490,9 @@ def transcription(
         **non_default_params,
     )
 
-    litellm_params_dict = get_litellm_params(**kwargs)
+    litellm_params_dict = get_litellm_params(
+        custom_llm_provider=custom_llm_provider, **kwargs
+    )
 
     litellm_logging_obj.update_environment_variables(
         model=model,
@@ -6717,7 +6721,9 @@ def speech(  # noqa: PLR0915
 
     if max_retries is None:
         max_retries = litellm.num_retries or openai.DEFAULT_MAX_RETRIES
-    litellm_params_dict = get_litellm_params(**kwargs)
+    litellm_params_dict = get_litellm_params(
+        custom_llm_provider=custom_llm_provider, **kwargs
+    )
 
     # Get provider-specific text-to-speech config and map parameters
     text_to_speech_provider_config = (
diff --git a/tests/test_litellm/litellm_core_utils/test_get_litellm_params.py b/tests/test_litellm/litellm_core_utils/test_get_litellm_params.py
index b39943b3e49..633a3ad2e84 100644
--- a/tests/test_litellm/litellm_core_utils/test_get_litellm_params.py
+++ b/tests/test_litellm/litellm_core_utils/test_get_litellm_params.py
@@ -95,6 +95,30 @@ class TestGetLitellmParamsBaseModel:
         assert result["base_model"] is None
 
 
+class TestGetLitellmParamsCustomLlmProvider:
+    """Verify custom_llm_provider handling — explicit value vs kwargs."""
+
+    def test_explicit_custom_llm_provider(self):
+        """Explicit custom_llm_provider should appear in result."""
+        result = get_litellm_params(custom_llm_provider="hosted_vllm")
+        assert result["custom_llm_provider"] == "hosted_vllm"
+
+    def test_custom_llm_provider_not_in_kwargs(self):
+        """When custom_llm_provider is not passed, it defaults to None.
+
+        This mirrors the bug in embedding()/transcription()/speech() where
+        custom_llm_provider was extracted as a named parameter and thus
+        removed from **kwargs before being passed to get_litellm_params().
+        """
+        result = get_litellm_params(**{"api_key": "test"})
+        assert result["custom_llm_provider"] is None
+
+    def test_explicit_custom_llm_provider_with_kwargs(self):
+        """Explicit custom_llm_provider should not be overridden by kwargs."""
+        result = get_litellm_params(custom_llm_provider="hosted_vllm", api_key="test")
+        assert result["custom_llm_provider"] == "hosted_vllm"
+
+
 class TestGetLitellmParamsExplicitFields:
     """Verify explicit parameters are always present in the result."""
 
@@ -125,4 +149,3 @@ class TestGetLitellmParamsExplicitFields:
     def test_no_log_from_explicit_param(self):
         result = get_litellm_params(no_log=True)
         assert result["no-log"] is True
-