[Bugfix][Qwen3TTS] Load speaker_id/voices from model configuration (#1079)

Signed-off-by: pablo <juanz9312@gmail.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: WeiQing Chen <40507679+david6666666@users.noreply.github.com>

examples/online_serving/qwen3_tts/README.md

@@ -75,6 +75,9 @@ curl -X POST http://localhost:8000/v1/audio/speech \
         "voice": "Vivian",
         "instructions": "Speak with great enthusiasm"
     }' --output excited.wav
+
+# List available voices in CustomVoice models
+curl http://localhost:8000/v1/audio/voices
 ```
 
 ## API Reference

tests/entrypoints/openai_api/test_serving_speech.py

@@ -201,6 +201,13 @@ def test_app():
     app = FastAPI()
     app.add_api_route("/v1/audio/speech", speech_server.create_speech, methods=["POST"], response_model=None)
 
+    # Add list_voices endpoint
+    async def list_voices():
+        speakers = sorted(speech_server.supported_speakers) if speech_server.supported_speakers else []
+        return {"voices": speakers}
+
+    app.add_api_route("/v1/audio/voices", list_voices, methods=["GET"])
+
     return app
 
 
@@ -268,6 +275,11 @@ class TestSpeechAPI:
         assert isinstance(audio_obj, CreateAudio)
         assert audio_obj.speed == 2.5
 
+    def test_list_voices_endpoint(self, client):
+        response = client.get("/v1/audio/voices")
+        assert response.status_code == 200
+        assert "voices" in response.json()
+
 
 class TestTTSMethods:
     """Unit tests for TTS validation and parameter building."""
@@ -311,9 +323,9 @@ class TestTTSMethods:
         req = OpenAICreateSpeechRequest(input="Hello", language="InvalidLang")
         assert "Invalid language" in speech_server._validate_tts_request(req)
 
-        # Invalid speaker
+        # When no speakers loaded, any voice is accepted (unconstrained)
         req = OpenAICreateSpeechRequest(input="Hello", voice="Invalid")
-        assert "Invalid speaker" in speech_server._validate_tts_request(req)
+        assert speech_server._validate_tts_request(req) is None
 
         # Valid request
         req = OpenAICreateSpeechRequest(input="Hello", voice="Vivian")
@@ -342,3 +354,26 @@ class TestTTSMethods:
         assert params["speaker"] == ["Ryan"]
         assert params["language"] == ["English"]
         assert params["task_type"] == ["CustomVoice"]
+
+    def test_load_supported_speakers(self):
+        """Test _load_supported_speakers."""
+        mock_engine_client = MagicMock()
+        mock_engine_client.errored = False
+        mock_engine_client.stage_list = None
+
+        # Mock talker_config with mixed-case speaker names
+        mock_talker_config = MagicMock()
+        mock_talker_config.spk_id = {"Ryan": 0, "Vivian": 1, "Aiden": 2}
+        mock_engine_client.model_config.hf_config.talker_config = mock_talker_config
+
+        mock_models = MagicMock()
+        mock_models.is_base_model.return_value = True
+
+        server = OmniOpenAIServingSpeech(
+            engine_client=mock_engine_client,
+            models=mock_models,
+            request_logger=MagicMock(),
+        )
+
+        # Verify speakers are normalized to lowercase
+        assert server.supported_speakers == {"ryan", "vivian", "aiden"}

vllm_omni/entrypoints/openai/api_server.py

@@ -762,6 +762,25 @@ async def create_speech(request: OpenAICreateSpeechRequest, raw_request: Request
         raise HTTPException(status_code=HTTPStatus.INTERNAL_SERVER_ERROR.value, detail=str(e)) from e
 
 
+@router.get(
+    "/v1/audio/voices",
+    responses={
+        HTTPStatus.OK.value: {"model": dict},
+        HTTPStatus.BAD_REQUEST.value: {"model": ErrorResponse},
+        HTTPStatus.NOT_FOUND.value: {"model": ErrorResponse},
+        HTTPStatus.INTERNAL_SERVER_ERROR.value: {"model": ErrorResponse},
+    },
+)
+async def list_voices(raw_request: Request):
+    """List available TTS voices/speakers from the loaded model."""
+    handler = Omnispeech(raw_request)
+    if handler is None:
+        return base(raw_request).create_error_response(message="The model does not support Speech API")
+
+    speakers = sorted(handler.supported_speakers) if handler.supported_speakers else []
+    return JSONResponse(content={"voices": speakers})
+
+
 # Health and Model endpoints for diffusion mode
 
 

vllm_omni/entrypoints/openai/serving_speech.py

@@ -19,17 +19,6 @@ logger = init_logger(__name__)
 
 # TTS Configuration (currently supports Qwen3-TTS)
 _TTS_MODEL_STAGES: set[str] = {"qwen3_tts"}
-_TTS_SPEAKERS: set[str] = {
-    "Vivian",
-    "Serena",
-    "Uncle_Fu",
-    "Dylan",
-    "Eric",
-    "Ryan",
-    "Aiden",
-    "Ono_Anna",
-    "Sohee",
-}
 _TTS_LANGUAGES: set[str] = {
     "Auto",
     "Chinese",
@@ -49,6 +38,30 @@ _TTS_MAX_NEW_TOKENS_MAX = 4096
 
 
 class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        # Load supported speakers
+        self.supported_speakers = self._load_supported_speakers()
+        logger.info(f"Loaded {len(self.supported_speakers)} supported speakers: {sorted(self.supported_speakers)}")
+
+    def _load_supported_speakers(self) -> set[str]:
+        """Load supported speakers (case-insensitive) from the model configuration."""
+        try:
+            talker_config = self.engine_client.model_config.hf_config.talker_config
+
+            # Check for speakers in either spk_id or speaker_id
+            for attr_name in ["spk_id", "speaker_id"]:
+                speakers_dict = getattr(talker_config, attr_name, None)
+                if speakers_dict and isinstance(speakers_dict, dict):
+                    # Normalize to lowercase for case-insensitive matching
+                    return {speaker.lower() for speaker in speakers_dict.keys()}
+
+            logger.warning("No speakers found in talker_config (checked spk_id and speaker_id)")
+        except Exception as e:
+            logger.warning(f"Could not load speakers from model config: {e}")
+
+        return set()
+
     def _is_tts_model(self) -> bool:
         """Check if the current model is a supported TTS model."""
         stage_list = getattr(self.engine_client, "stage_list", None)
@@ -63,6 +76,10 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
         """Validate TTS request parameters. Returns error message or None."""
         task_type = request.task_type or "CustomVoice"
 
+        # Normalize voice to lowercase for case-insensitive matching
+        if request.voice is not None:
+            request.voice = request.voice.lower()
+
         # Validate input is not empty
         if not request.input or not request.input.strip():
             return "Input text cannot be empty"
@@ -73,8 +90,8 @@ class OmniOpenAIServingSpeech(OpenAIServing, AudioMixin):
 
         # Validate speaker for CustomVoice task
         if task_type == "CustomVoice" and request.voice is not None:
-            if request.voice not in _TTS_SPEAKERS:
-                return f"Invalid speaker '{request.voice}'. Supported: {', '.join(sorted(_TTS_SPEAKERS))}"
+            if self.supported_speakers and request.voice not in self.supported_speakers:
+                return f"Invalid speaker '{request.voice}'. Supported: {', '.join(sorted(self.supported_speakers))}"
 
         # Validate Base task requirements
         if task_type == "Base":