From eb25b1794610c90a9bc62adba4c11815fe83e805 Mon Sep 17 00:00:00 2001 From: Julien Parenti Date: Fri, 6 Nov 2026 02:41:12 -0800 Subject: [PATCH] fix(vertex_ai): honor vertex_location for the Cloud Text-to-Speech endpoint (#50255) Backport of #50255 to lts/1.108.x. Cherry-picked from 6fc9555f83b8130a0ffd1e18f38b165eebbebe75 (main), adapted by hand: the line's text-to-speech module has no Gemini TTS code, so the Gemini parts of the pick are dropped and chirp takes its host from the deployment's or call's own vertex_location only. --- .../text_to_speech/transformation.py | 45 ++++- .../text_to_speech/test_transformation.py | 173 +++++++++++++++++- 2 files changed, 214 insertions(+), 4 deletions(-) diff --git a/litellm/llms/vertex_ai/text_to_speech/transformation.py b/litellm/llms/vertex_ai/text_to_speech/transformation.py index 2fae2a9de1c..eaaa600b870 100644 --- a/litellm/llms/vertex_ai/text_to_speech/transformation.py +++ b/litellm/llms/vertex_ai/text_to_speech/transformation.py @@ -49,6 +49,43 @@ _LyriaVoice: TypeAlias = str | dict | None _JSON_OBJECT: Final = TypeAdapter(Mapping[str, object], config=ConfigDict(hide_input_in_errors=True)) _JSON_ARRAY: Final = TypeAdapter(tuple[object, ...], config=ConfigDict(hide_input_in_errors=True)) _STR: Final = TypeAdapter(str, config=ConfigDict(hide_input_in_errors=True)) +_CLOUD_TTS_LOCATIONS: Final = MappingProxyType( + { + **dict.fromkeys( + ( + "us", + "us-central1", + "us-east1", + "us-east4", + "us-east5", + "us-south1", + "us-west1", + "us-west2", + "us-west3", + "us-west4", + "us-west5", + ), + "us", + ), + **dict.fromkeys( + ( + "eu", + "europe-central2", + "europe-north1", + "europe-north2", + "europe-southwest1", + "europe-west1", + "europe-west3", + "europe-west4", + "europe-west8", + "europe-west9", + "europe-west10", + "europe-west12", + ), + "eu", + ), + } +) class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase): @@ -173,7 +210,7 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase): # Resolve Vertex AI credentials using VertexBase helpers vertex_credentials: Final = self.safe_get_vertex_ai_credentials(litellm_params_dict) vertex_project: Final = self.safe_get_vertex_ai_project(litellm_params_dict) - vertex_location: Final = self.safe_get_vertex_ai_location(litellm_params_dict) + vertex_location: Final = self.explicit_vertex_ai_location(litellm_params_dict) # Convert voice to string if it's a dict (extract name) # Actual voice mapping happens in map_openai_params @@ -314,7 +351,11 @@ class VertexAITextToSpeechConfig(BaseTextToSpeechConfig, VertexBase): if api_base: return api_base - return self.TTS_API_URL + location: Final = _JSON_OBJECT.validate_python(litellm_params).get("vertex_location") + cloud_tts_location: Final = _CLOUD_TTS_LOCATIONS.get(location) if isinstance(location, str) else None + if cloud_tts_location is None: + return self.TTS_API_URL + return f"https://{cloud_tts_location}-texttospeech.googleapis.com/v1/text:synthesize" def _validate_vertex_input( self, 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 dc6f35fcfe5..b4ccb1cf081 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 @@ -48,6 +48,73 @@ class TestVertexAITextToSpeechConfig: assert url == custom_url + @pytest.mark.parametrize( + ("vertex_location", "expected_host"), + [ + ("eu", "eu-texttospeech.googleapis.com"), + ("europe-west4", "eu-texttospeech.googleapis.com"), + ("europe-west1", "eu-texttospeech.googleapis.com"), + ("europe-west3", "eu-texttospeech.googleapis.com"), + ("europe-west8", "eu-texttospeech.googleapis.com"), + ("europe-west9", "eu-texttospeech.googleapis.com"), + ("europe-west12", "eu-texttospeech.googleapis.com"), + ("europe-west10", "eu-texttospeech.googleapis.com"), + ("europe-central2", "eu-texttospeech.googleapis.com"), + ("europe-north1", "eu-texttospeech.googleapis.com"), + ("europe-north2", "eu-texttospeech.googleapis.com"), + ("europe-southwest1", "eu-texttospeech.googleapis.com"), + ("us", "us-texttospeech.googleapis.com"), + ("us-central1", "us-texttospeech.googleapis.com"), + ("us-east1", "us-texttospeech.googleapis.com"), + ("us-east4", "us-texttospeech.googleapis.com"), + ("us-east5", "us-texttospeech.googleapis.com"), + ("us-south1", "us-texttospeech.googleapis.com"), + ("us-west1", "us-texttospeech.googleapis.com"), + ("us-west2", "us-texttospeech.googleapis.com"), + ("us-west3", "us-texttospeech.googleapis.com"), + ("us-west4", "us-texttospeech.googleapis.com"), + ("us-west5", "us-texttospeech.googleapis.com"), + ], + ) + def test_get_complete_url_uses_regional_endpoint_for_vertex_location( + self, vertex_location: str, expected_host: str + ) -> None: + config: Final = VertexAITextToSpeechConfig() + + url: Final = config.get_complete_url( + model="vertex_ai/chirp", + api_base=None, + litellm_params={"vertex_location": vertex_location}, + ) + + assert url == f"https://{expected_host}/v1/text:synthesize" + + @pytest.mark.parametrize( + "vertex_location", ["global", "europe-west2", "asia-northeast1", "northamerica-northeast1", None] + ) + def test_get_complete_url_keeps_global_endpoint_without_a_regional_match(self, vertex_location: str | None) -> None: + config: Final = VertexAITextToSpeechConfig() + + url: Final = config.get_complete_url( + model="vertex_ai/chirp", + api_base=None, + litellm_params={"vertex_location": vertex_location}, + ) + + assert url == "https://texttospeech.googleapis.com/v1/text:synthesize" + + def test_get_complete_url_api_base_wins_over_vertex_location(self) -> None: + config: Final = VertexAITextToSpeechConfig() + + custom_url: Final = "https://custom-tts-endpoint.example.com/v1/synthesize" + url: Final = config.get_complete_url( + model="vertex_ai/chirp", + api_base=custom_url, + litellm_params={"vertex_location": "eu"}, + ) + + assert url == custom_url + @patch.object(VertexAITextToSpeechConfig, "_ensure_access_token") @patch.object(VertexAITextToSpeechConfig, "_get_token_and_url") def test_transform_text_to_speech_request_body(self, mock_get_token, mock_ensure_token): @@ -616,8 +683,7 @@ def test_litellm_speech_vertex_ai_chirp(mock_get_token, mock_ensure_token, mock_ mock_post.assert_called_once() call_kwargs = mock_post.call_args.kwargs - # Verify the URL is the Google Cloud TTS API - assert call_kwargs["url"] == "https://texttospeech.googleapis.com/v1/text:synthesize" + assert call_kwargs["url"] == "https://us-texttospeech.googleapis.com/v1/text:synthesize" # Verify request body structure assert "json" in call_kwargs @@ -764,6 +830,109 @@ async def test_aspeech_vertex_ai_sends_ssml_input( } +@pytest.mark.asyncio +async def test_aspeech_vertex_ai_chirp_keeps_the_global_host_when_only_vertexai_location_is_set( + respx_mock: respx.MockRouter, google_token_endpoint: responses.RequestsMock, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setenv("VERTEXAI_LOCATION", "europe-west4") + route: Final = respx_mock.post(_SYNTHESIZE_URL).mock( + return_value=httpx.Response(200, json={"audioContent": base64.b64encode(b"global-audio").decode()}) + ) + + response: Final = await litellm.aspeech( + model="vertex_ai/chirp", + input="Bonjour", + voice="fr-FR-Chirp3-HD-Aoede", + vertex_credentials=_AUTHORIZED_USER, + ) + + assert route.call_count == 1 + assert response.content == b"global-audio" + + +@pytest.mark.asyncio +async def test_aspeech_vertex_ai_chirp_keeps_the_global_host_with_other_location_defaults( + respx_mock: respx.MockRouter, google_token_endpoint: responses.RequestsMock, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setenv("VERTEX_LOCATION", "europe-west4") + monkeypatch.setattr(litellm, "vertex_location", "us-central1") + route: Final = respx_mock.post(_SYNTHESIZE_URL).mock( + return_value=httpx.Response(200, json={"audioContent": base64.b64encode(b"global-audio").decode()}) + ) + + response: Final = await litellm.aspeech( + model="vertex_ai/chirp", + input="Bonjour", + voice="fr-FR-Chirp3-HD-Aoede", + vertex_credentials=_AUTHORIZED_USER, + ) + + assert route.call_count == 1 + assert response.content == b"global-audio" + + +@pytest.mark.asyncio +@pytest.mark.parametrize( + ("location_param", "location", "expected_host"), + [ + ("vertex_location", "eu", "eu-texttospeech.googleapis.com"), + ("vertex_location", "us-central1", "us-texttospeech.googleapis.com"), + ("vertex_ai_location", "europe-west4", "eu-texttospeech.googleapis.com"), + ], +) +async def test_aspeech_vertex_ai_chirp_calls_the_regional_host_for_an_explicit_location( + respx_mock: respx.MockRouter, + google_token_endpoint: responses.RequestsMock, + location_param: str, + location: str, + expected_host: str, +) -> None: + route: Final = respx_mock.post(f"https://{expected_host}/v1/text:synthesize").mock( + return_value=httpx.Response(200, json={"audioContent": base64.b64encode(b"regional-audio").decode()}) + ) + + response: Final = await litellm.aspeech( + model="vertex_ai/chirp", + input="Bonjour", + voice="fr-FR-Chirp3-HD-Aoede", + vertex_credentials=_AUTHORIZED_USER, + **{location_param: location}, + ) + + assert route.call_count == 1 + assert response.content == b"regional-audio" + + +@pytest.mark.asyncio +async def test_aspeech_vertex_ai_lyria_predict_still_uses_vertexai_location( + respx_mock: respx.MockRouter, google_token_endpoint: responses.RequestsMock, monkeypatch: pytest.MonkeyPatch +) -> None: + monkeypatch.setenv("VERTEXAI_LOCATION", "us-east5") + route: Final = respx_mock.post( + "https://us-east5-aiplatform.googleapis.com/v1/projects/test-project/locations/us-east5" + "/publishers/google/models/lyria-002:predict" + ).mock( + return_value=httpx.Response( + 200, + json={ + "predictions": [ + {"audioContent": base64.b64encode(b"lyria-audio").decode(), "mimeType": "audio/wav"}, + ] + }, + ) + ) + + response: Final = await litellm.aspeech( + model="vertex_ai/lyria-002", + input="calm piano", + voice="unused", + vertex_credentials=_AUTHORIZED_USER, + ) + + assert route.call_count == 1 + assert response.content == b"lyria-audio" + + _WAV_AUDIO_B64: Final = "UklGRiQAAABXQVZFZm10IA==" _WAV_AUDIO: Final = b"RIFF$\x00\x00\x00WAVEfmt " _UNLABELED_AUDIO_B64: Final = "bHlyaWEtMy1hdWRpbw=="