feat: design model support for qwen-tts backend. remove unnecessary port split for qwen models

1 files changed, 100 insertions, 2 deletions

diff --git a/app/tests/test_tts.py b/app/tests/test_tts.py
index 2b2ac1c..49eb6f6 100644
--- a/app/tests/test_tts.py
+++ b/app/tests/test_tts.py
@@ -31,6 +31,8 @@ from converter.clients import (
     TTS_LANGUAGES,
     VOICE_MODE_CLONE,
     VOICE_MODE_CUSTOM,
+    VOICE_MODE_DESIGN,
+    VOICE_MODES,
     AudioCppTTSClient,
     FasterTTSClient,
     QwenTTSClient,
@@ -494,6 +496,80 @@ class QwenTTSClientGenerateTests(unittest.TestCase):
         mock_generate.assert_not_called()
 
 
+class QwenTTSClientVoiceDesignTests(unittest.TestCase):
+    """Qwen VoiceDesign mode: instructions and the /run_voice_design call."""
+
+    def setUp(self):
+        self._tmp = tempfile.TemporaryDirectory()
+
+    def tearDown(self):
+        self._tmp.cleanup()
+
+    def _client(self, instructions=None):
+        client = QwenTTSClient.__new__(QwenTTSClient)
+        client.chunks_dir = Path(self._tmp.name)
+        client.voice_mode = VOICE_MODE_DESIGN
+        client.language = config.LANGUAGE
+        client.instructions = (instructions if instructions is not None
+                               else config.INSTRUCT).strip()
+        client.api_info = {"named_endpoints": {"/run_voice_design": {
+            "parameters": [
+                {"parameter_name": "text"},
+                {"parameter_name": "lang_disp"},
+                {"parameter_name": "design"},
+            ]}}}
+        return client
+
+    def _fake_output(self) -> str:
+        out = Path(self._tmp.name) / "server_out.wav"
+        out.write_bytes(b"\x01\x00")
+        return str(out)
+
+    def test_voice_mode_design_is_valid(self):
+        self.assertIn(VOICE_MODE_DESIGN, VOICE_MODES)
+
+    def test_generate_payload_and_return(self):
+        client = self._client(instructions="A warm narrator")
+        fake = MagicMock(return_value=(self._fake_output(),))
+        with patch.object(client, "_generate_voice_design", fake):
+            result = client._generate_sub_request(
+                "Hello there.", self._tmp.name, 1, 1, 1)
+        fake.assert_called_once_with("Hello there.")
+        self.assertEqual(Path(result).name, "part_01.wav")
+
+    def test_payload_uses_design_field_language_and_instruction(self):
+        client = self._client(instructions="A warm narrator")
+        captured = {}
+
+        def fake_predict(**payload):
+            captured.update(payload)
+            return (self._fake_output(),)
+
+        client.client = MagicMock()
+        client.client.predict.side_effect = fake_predict
+        result = client._generate_voice_design("Hi.")
+        self.assertEqual(captured["text"], "Hi.")
+        self.assertEqual(captured["lang_disp"], config.LANGUAGE)
+        self.assertEqual(captured["design"], "A warm narrator")
+        self.assertNotIn("seed", captured)  # not accepted by this endpoint
+        self.assertEqual(result, (self._fake_output(),))
+
+    def test_payload_defaults_instructions_to_config(self):
+        client = self._client(instructions=None)
+        self.assertEqual(client.instructions,
+                         (config.INSTRUCT or "").strip())
+
+    def test_unknown_api_falls_back_to_the_requested_name(self):
+        client = self._client()
+        client.api_info = {"named_endpoints": {}}
+        client.client = MagicMock()
+        client.client.predict.return_value = (self._fake_output(),)
+        client._generate_voice_design("Hi.")
+        _, kwargs = client.client.predict.call_args
+        self.assertEqual(kwargs["api_name"], "/run_voice_design")
+
+
+
 class AudioCppTTSClientHealthTests(unittest.TestCase):
     """Connection behavior of the audio.cpp client."""
 
@@ -1682,6 +1758,28 @@ class BackendWiringTests(unittest.TestCase):
                 AudiobookConverter(voice_mode=VOICE_MODE_CLONE,
                                    backend=BACKEND_QWEN)
 
+    def test_qwen_design_mode_without_instructions_rejected(self):
+        # A VoiceDesign run needs a description; an empty instructions
+        # value (not even the config default) is refused up front.
+        with patch("converter.converter.QwenTTSClient"):
+            with self.assertRaises(ValueError):
+                AudiobookConverter(voice_mode=VOICE_MODE_DESIGN,
+                                   backend=BACKEND_QWEN, instructions="   ")
+
+    def test_qwen_design_mode_threads_instructions_to_the_client(self):
+        with patch("converter.converter.QwenTTSClient") as mock_qwen:
+            AudiobookConverter(voice_mode=VOICE_MODE_DESIGN,
+                               backend=BACKEND_QWEN,
+                               instructions="A warm adult female narrator")
+        _, kwargs = mock_qwen.call_args
+        self.assertEqual(kwargs["instructions"],
+                         "A warm adult female narrator")
+
+    def test_qwen_design_narrator_tag_uses_designed(self):
+        self.assertEqual(AudiobookConverter.compute_narrator_tag(
+            BACKEND_QWEN, None, VOICE_MODE_DESIGN, None,
+            "A warm adult female narrator"), "designed")
+
     def test_api_url_override_reaches_each_client(self):
         # A remote conversion threads api_url through to the selected client.
         with patch("converter.converter.AudioCppTTSClient") as mock_audiocpp:
@@ -1709,8 +1807,8 @@ class BackendWiringTests(unittest.TestCase):
             chunks_dir=converter_mod.CHUNKS_FOLDER,
             voice_mode=VOICE_MODE_CUSTOM, voice_clone_ref_audio=None,
             voice_clone_ref_text=None, skip_transcription=False,
-            language=config.LANGUAGE, api_url="http://10.0.0.5:7860",
-            quiet=False)
+            language=config.LANGUAGE, instructions=None,
+            api_url="http://10.0.0.5:7860", quiet=False)
 
     def test_audiocpp_clone_mode_does_not_require_reference(self):
         # Cloning is server-side for the audiocpp backend, so the