diff --git a/litellm/endpoints/speech/speech_to_completion_bridge/handler.py b/litellm/endpoints/speech/speech_to_completion_bridge/handler.py index 6c33621ec89..64d02a17d46 100644 --- a/litellm/endpoints/speech/speech_to_completion_bridge/handler.py +++ b/litellm/endpoints/speech/speech_to_completion_bridge/handler.py @@ -7,6 +7,8 @@ from typing import TYPE_CHECKING, Final from typing_extensions import TypedDict if TYPE_CHECKING: + import httpx + from litellm import LiteLLMLoggingObj from litellm.types.llms.openai import HttpxBinaryResponseContent @@ -85,6 +87,8 @@ class SpeechToCompletionBridgeHandler: headers: dict, logging_obj: "LiteLLMLoggingObj", custom_llm_provider: str, + api_base: str | None, + timeout: "float | httpx.Timeout | None", ) -> "HttpxBinaryResponseContent": received_args: Final = locals() from litellm import completion @@ -104,7 +108,7 @@ class SpeechToCompletionBridgeHandler: model=model, input=input, optional_params=optional_params, - litellm_params=litellm_params, + litellm_params={**litellm_params, "api_base": api_base, "timeout": timeout}, headers=headers, litellm_logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, diff --git a/litellm/main.py b/litellm/main.py index 6f77c017023..b0e756680ff 100644 --- a/litellm/main.py +++ b/litellm/main.py @@ -8751,6 +8751,8 @@ def speech( headers=headers or {}, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + api_base=api_base, + timeout=timeout, ) # Vertex AI Text-to-Speech (Google Cloud TTS) @@ -8802,6 +8804,8 @@ def speech( headers=headers or {}, logging_obj=logging_obj, custom_llm_provider=custom_llm_provider, + api_base=api_base, + timeout=timeout, ) elif custom_llm_provider == "runwayml": from litellm.llms.runwayml.text_to_speech.transformation import ( diff --git a/tests/unit/llms/vertex_ai/text_to_speech/test_transformation.py b/tests/unit/llms/vertex_ai/text_to_speech/test_transformation.py index b4ccb1cf081..7e6f5afe5f8 100644 --- a/tests/unit/llms/vertex_ai/text_to_speech/test_transformation.py +++ b/tests/unit/llms/vertex_ai/text_to_speech/test_transformation.py @@ -1098,3 +1098,49 @@ def test_lyria_malformed_item_position_does_not_change_the_mapped_error(payload: custom_llm_provider="vertex_ai", original_exception=exc_info.value, ) + + +_PCM_AUDIO: Final = b"\x00\x00\x01\x00" * 2400 +_GENERATE_CONTENT_RESPONSE: Final = { + "candidates": [ + { + "content": { + "role": "model", + "parts": [ + { + "inlineData": { + "mimeType": "audio/L16;codec=pcm;rate=24000", + "data": base64.b64encode(_PCM_AUDIO).decode(), + } + } + ], + }, + "finishReason": "STOP", + } + ], + "usageMetadata": {"promptTokenCount": 5, "candidatesTokenCount": 10, "totalTokenCount": 15}, +} + + +@pytest.mark.parametrize("custom_llm_provider", ("vertex_ai", "vertex_ai_beta")) +@pytest.mark.asyncio +async def test_aspeech_vertex_ai_gemini_tts_routes_to_configured_api_base( + respx_mock: respx.MockRouter, google_token_endpoint: responses.RequestsMock, custom_llm_provider: str +) -> None: + gateway_route: Final = respx_mock.post( + "https://vertex.gateway.internal/v1/projects/test-project/locations/global" + "/publishers/google/models/gemini-3.8-flash-tts:generateContent" + ).mock(return_value=httpx.Response(200, json=_GENERATE_CONTENT_RESPONSE)) + + response: Final = await litellm.aspeech( + model=f"{custom_llm_provider}/gemini-3.8-flash-tts", + input="Say hello", + voice="Kore", + vertex_credentials=_AUTHORIZED_USER, + vertex_location="global", + api_base="https://vertex.gateway.internal", + ) + + assert gateway_route.call_count == 1 + assert gateway_route.calls.last.request.headers["authorization"] == "Bearer minted-google-token" + assert _PCM_AUDIO in response.content diff --git a/tests/unit/test_main.py b/tests/unit/test_main.py index c832da5dbfc..48a4c1b5f09 100644 --- a/tests/unit/test_main.py +++ b/tests/unit/test_main.py @@ -4777,6 +4777,67 @@ def test_speech_mistral_routes_to_configured_api_base(respx_mock: respx.MockRout assert response.content == audio_bytes +GEMINI_TTS_MODEL: Final = "gemini-3.8-flash-tts" +GEMINI_GATEWAY_BASE: Final = "https://gemini.gateway.internal/v1beta" +GEMINI_DEFAULT_GENERATE_CONTENT_URL: Final = ( + rf"https://generativelanguage\.googleapis\.com/[^/]+/models/{GEMINI_TTS_MODEL}:generateContent" +) + + +@pytest.mark.parametrize("endpoint_param", ["api_base", "base_url"]) +@pytest.mark.parametrize("sync_mode", [True, False]) +async def test_speech_gemini_tts_routes_to_configured_api_base( + respx_mock: respx.MockRouter, monkeypatch: pytest.MonkeyPatch, endpoint_param: str, sync_mode: bool +): + monkeypatch.setattr(litellm, "disable_aiohttp_transport", True) + gateway_route: Final = respx_mock.post(f"{GEMINI_GATEWAY_BASE}/models/{GEMINI_TTS_MODEL}:generateContent").mock( + return_value=httpx.Response(200, json=_gemini_tts_generate_content_response()) + ) + speech_kwargs: Final = { + "model": f"gemini/{GEMINI_TTS_MODEL}", + "input": "hello from litellm", + "voice": "Kore", + "api_key": "fake-gemini-key", + endpoint_param: GEMINI_GATEWAY_BASE, + } + + response: Final = litellm.speech(**speech_kwargs) if sync_mode else await litellm.aspeech(**speech_kwargs) + + assert gateway_route.call_count == 1 + assert gateway_route.calls.last.request.headers["x-goog-api-key"] == "fake-gemini-key" + assert b"pcm-audio-bytes" in response.content + + +def test_speech_gemini_tts_applies_the_caller_timeout(respx_mock: respx.MockRouter): + route: Final = respx_mock.post(url__regex=GEMINI_DEFAULT_GENERATE_CONTENT_URL).mock( + return_value=httpx.Response(200, json=_gemini_tts_generate_content_response()) + ) + + litellm.speech( + model=f"gemini/{GEMINI_TTS_MODEL}", input="hello", voice="Kore", api_key="fake-gemini-key", timeout=7.5 + ) + + assert route.calls.last.request.extensions["timeout"] == httpx.Timeout(7.5).as_dict() + + +def test_speech_gemini_tts_without_a_timeout_waits_as_long_as_the_same_chat_completion(respx_mock: respx.MockRouter): + route: Final = respx_mock.post(url__regex=GEMINI_DEFAULT_GENERATE_CONTENT_URL).mock( + return_value=httpx.Response(200, json=_gemini_tts_generate_content_response()) + ) + + litellm.speech(model=f"gemini/{GEMINI_TTS_MODEL}", input="hello", voice="Kore", api_key="fake-gemini-key") + litellm.completion( + model=f"gemini/{GEMINI_TTS_MODEL}", + messages=[{"role": "user", "content": "hello"}], + modalities=["audio"], + audio={"voice": "Kore", "format": "pcm16"}, + api_key="fake-gemini-key", + ) + + speech_call, chat_call = route.calls + assert speech_call.request.extensions["timeout"] == chat_call.request.extensions["timeout"] + + FOUNDRY_HOST: Final = "https://my-project.services.ai.azure.com"