Build and run agents you can see, understand and trust.
Python agentchatbotlarge-language-modelsllmllm-agentmcpmulti-agentmulti-modalreact-agentrealtime-agentvoice-agentvoice-assistant
feat(tts): support per-call options in DashScope TTS (#3002)
cxyyy665d2dd540a78e
diff --git a/src/agentscope/tts/_dashscope/_model.py b/src/agentscope/tts/_dashscope/_model.pyindex 758ea1975..cb08e8ec9 100644--- a/src/agentscope/tts/_dashscope/_model.py+++ b/src/agentscope/tts/_dashscope/_model.py@@ -127,14 +127,16 @@ class DashScopeTTSModel(TTSModelBase):import dashscope- response = dashscope.MultiModalConversation.call(- model=self.model,- api_key=self.credential.api_key.get_secret_value(),- text=text,- voice=self.parameters.voice,- stream=True,- **kwargs,- )+ request_kwargs: dict[str, Any] = {+ "model": self.model,+ "text": text,+ "voice": self.parameters.voice,+ }+ request_kwargs.update(kwargs)+ request_kwargs["api_key"] = self.credential.api_key.get_secret_value()+ request_kwargs["stream"] = True++ response = dashscope.MultiModalConversation.call(**request_kwargs)if self.stream:return self._parse_into_async_generator(response)diff --git a/tests/tts_dashscope_test.py b/tests/tts_dashscope_test.pyindex 39d27c2ac..6c80f9c92 100644--- a/tests/tts_dashscope_test.py+++ b/tests/tts_dashscope_test.py@@ -22,6 +22,8 @@ from typing import Any, AsyncGeneratorfrom unittest import IsolatedAsyncioTestCasefrom unittest.mock import MagicMock, Mock, patch+from utils import AnyString, AnyValue+from agentscope.app._service._tts_model import _resolve_tts_classfrom agentscope.credential import DashScopeCredentialfrom agentscope.tts import (@@ -40,6 +42,31 @@ _TTS_SAMPLE_RATE = 24000_TTS_CHANNELS = 1_TTS_SAMPLE_WIDTH = 2 # bytes (= 16 bit)_WAV_HEADER_LEN = 44+_EXPECTED_SINGLE_AUDIO_RESPONSE = {+ "content": {+ "type": "data",+ "id": AnyString(),+ "source": {+ "type": "base64",+ "data": AnyString(),+ "media_type": _MEDIA_TYPE,+ },+ "name": None,+ "created_at": AnyString(),+ "finished_at": None,+ },+ "id": AnyString(),+ "created_at": AnyString(),+ "type": "tts",+ "usage": {+ "input_tokens": 0,+ "output_tokens": 0,+ "time": AnyValue(),+ "type": "tts",+ },+ "metadata": None,+ "is_last": True,+}# ---------------------------------------------------------------------------@@ -105,6 +132,7 @@ class _DummyTTS(TTSModelBase):text: str | None = None,**kwargs: Any,) -> TTSResponse | AsyncGenerator[TTSResponse, None]:+ """Return an empty TTS response."""del text, kwargsreturn TTSResponse(content=None)@@ -121,9 +149,11 @@ class _RealtimeDummyTTS(_DummyTTS):self.close_calls = 0async def connect(self) -> None:+ """Record a mock connection."""self.connect_calls += 1async def close(self) -> None:+ """Record a mock disconnection."""self.close_calls += 1@@ -240,6 +270,100 @@ class TestDashScopeTTSModel(IsolatedAsyncioTestCase):async for chunk in result:self.assertFalse(chunk.is_last)+ async def test_per_call_voice_override(self) -> None:+ """A temporary voice reaches the SDK without changing the default."""+ for stream in (False, True):+ with self.subTest(stream=stream):+ model = self._make_model(stream=stream)+ for kwargs, expected in (+ ({"voice": "Serena"}, "Serena"),+ ({}, "Cherry"),+ ):+ self.mock_mmc.call.return_value = _make_api_generator(+ [b"AAAA"],+ )+ result = await model.synthesize("Hello", **kwargs)+ responses = (+ [chunk async for chunk in result]+ if stream+ else [result]+ )+ self.assertEqual(+ self.mock_mmc.call.call_args.kwargs,+ {+ "model": "qwen3-tts-flash",+ "api_key": "test",+ "text": "Hello",+ "voice": expected,+ "stream": True,+ },+ )+ self.assertEqual(+ model.parameters.model_dump(),+ {"voice": "Cherry"},+ )+ self.assertEqual(+ [+ {+ **response,+ "content": response.content.model_dump(),+ "usage": dict(response.usage),+ }+ for response in responses+ ],+ [_EXPECTED_SINGLE_AUDIO_RESPONSE],+ )++ async def test_request_options_keep_credentials_and_transport(+ self,+ ) -> None:+ """Request options preserve configured credentials and transport."""+ for stream in (False, True):+ with self.subTest(stream=stream):+ model = self._make_model(stream=stream)+ self.mock_mmc.call.return_value = _make_api_generator(+ [b"AAAA"],+ )+ result = await model.synthesize(+ "Hello",+ model="qwen-tts",+ voice="Serena",+ language_type="English",+ api_key="ignored",+ stream=False,+ )+ responses = (+ [chunk async for chunk in result] if stream else [result]+ )+ self.assertEqual(+ self.mock_mmc.call.call_args.kwargs,+ {+ "model": "qwen-tts",+ "api_key": "test",+ "text": "Hello",+ "voice": "Serena",+ "language_type": "English",+ "stream": True,+ },+ )+ self.assertEqual(+ model.parameters.model_dump(),+ {"voice": "Cherry"},+ )+ self.assertEqual(model.model, "qwen3-tts-flash")+ self.assertEqual(model.stream, stream)+ self.assertEqual(+ [+ {+ **response,+ "content": response.content.model_dump(),+ "usage": dict(response.usage),+ }+ for response in responses+ ],+ [_EXPECTED_SINGLE_AUDIO_RESPONSE],+ )+async def test_aggregates_chunks(self) -> None:"""All API chunks are aggregated into one self-contained WAV."""self.mock_mmc.call.return_value = _make_api_generator(