fix(speech): forward api_key to the TTS bridge and isolate response hidden params

basedpyright-code-budget.json

@@ -99,7 +99,7 @@
     "limit": 0
   },
   "reportUnknownArgumentType": {
-    "limit": 44530
+    "limit": 44528
   },
   "reportUnknownLambdaType": {
     "limit": 109
@@ -138,9 +138,9 @@
     "limit": 139
   },
   "reportUnusedImport": {
-    "limit": 545
+    "limit": 544
   },
   "reportUnusedVariable": {
-    "limit": 146
+    "limit": 145
   }
 }

litellm/main.py

@@ -8013,7 +8013,7 @@ def speech(
 
     if max_retries is None:
         max_retries = litellm.num_retries or openai.DEFAULT_MAX_RETRIES
-    litellm_params_dict: Final = get_litellm_params(metadata=metadata, **kwargs)
+    litellm_params_dict: Final = get_litellm_params(metadata=metadata, api_key=api_key or dynamic_api_key, **kwargs)
 
     # Get provider-specific text-to-speech config and map parameters
     text_to_speech_provider_config = ProviderConfigManager.get_provider_text_to_speech_config(

litellm/types/llms/openai.py

@@ -107,10 +107,17 @@ EmbeddingInput = str | list[str]
 
 
 class HttpxBinaryResponseContent(_HttpxBinaryResponseContent):
-    _hidden_params: dict = {}
+    _hidden_params: dict
+
+    def __init__(self, response: httpx.Response) -> None:
+        super().__init__(response)
+        self._hidden_params = {}  # mutable-ok: mutable-dict contract shared with ModelResponse logging consumers
 
     def set_response_cost(self, response_cost: float | None) -> None:
-        self._hidden_params = {"response_cost": response_cost}
+        if response_cost is None:
+            self._hidden_params.pop("response_cost", None)
+            return
+        self._hidden_params["response_cost"] = response_cost
 
 
 class NotGiven:

ruff-strict-budget.json

@@ -108,7 +108,7 @@
     "limit": 3
   },
   "F401": {
-    "limit": 14
+    "limit": 13
   },
   "LOG015": {
     "limit": 5
@@ -171,7 +171,7 @@
     "limit": 175
   },
   "RUF012": {
-    "limit": 240
+    "limit": 239
   },
   "RUF015": {
     "limit": 8
@@ -183,7 +183,7 @@
     "limit": 4
   },
   "RUF059": {
-    "limit": 67
+    "limit": 66
   },
   "RUF100": {
     "limit": 0

tests/test_litellm/test_main.py

@@ -3042,6 +3042,8 @@ async def test_aspeech_gemini_bridge_keeps_proxy_metadata_for_spend_tracking(
     respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
+    monkeypatch.delenv("GEMINI_API_KEY", raising=False)
+    monkeypatch.delenv("GOOGLE_API_KEY", raising=False)
     recorder: Final = _SuccessEventRecorder()
     monkeypatch.setattr(litellm, "callbacks", [recorder])
     mock_route: Final = respx_mock.post(
@@ -3057,6 +3059,7 @@ async def test_aspeech_gemini_bridge_keeps_proxy_metadata_for_spend_tracking(
     )
 
     assert mock_route.called
+    assert mock_route.calls.last.request.headers["x-goog-api-key"] == "fake-gemini-key"
     speech_event: Final = await _wait_for_success_event(recorder, call_type="aspeech")
     assert speech_event.spend_metadata["user_api_key"] == "hashed-virtual-key"
     assert speech_event.spend_metadata["user_api_key_user_id"] == "user-1"

tests/test_litellm/types/llms/test_types_llms_openai.py

@@ -7,6 +7,7 @@ import pytest
 import json
 
 import litellm
+from litellm.types.llms.openai import HttpxBinaryResponseContent
 
 
 def test_generic_event():
@@ -522,3 +523,34 @@ class TestOpenAIFileObjectBatchGuardrailSerialization:
 
         page = FileListPage(object="list", data=[self._file_object()], has_more=False)
         assert "litellm_batch_guardrail" not in page.model_dump(mode="json")["data"][0]
+
+
+def _binary_content(payload: bytes) -> HttpxBinaryResponseContent:
+    import httpx
+
+    return HttpxBinaryResponseContent(httpx.Response(200, content=payload))
+
+
+def test_httpx_binary_response_content_hidden_params_are_per_instance():
+    first = _binary_content(b"first")
+    second = _binary_content(b"second")
+
+    first._hidden_params["response_cost"] = 0.5
+
+    assert second._hidden_params == {}
+
+
+def test_set_response_cost_none_leaves_hidden_params_empty():
+    binary_response = _binary_content(b"audio")
+
+    binary_response.set_response_cost(None)
+
+    assert "response_cost" not in binary_response._hidden_params
+
+    binary_response.set_response_cost(0.25)
+
+    assert binary_response._hidden_params["response_cost"] == 0.25
+
+    binary_response.set_response_cost(None)
+
+    assert "response_cost" not in binary_response._hidden_params

type-discipline-budget.json

@@ -3,7 +3,7 @@
     "limit": 22733
   },
   "LIT002": {
-    "limit": 26864
+    "limit": 26863
   },
   "LIT003": {
     "limit": 269