feat: 운영 안정성과 세션 음성 경험 개선

apps/api/app/test_voice_service.py

@@ -12,9 +12,12 @@ from .services.voice import (
     TTS_ENDPOINT,
     TTS_MODEL,
     TTS_MODEL_FALLBACK,
+    HIGGS_TTS_ENDPOINT,
+    HIGGS_TTS_MODEL,
     VoicePreset,
     VoiceService,
     assess_end_of_turn,
+    build_higgs_prompt,
     build_tts_payload,
     resolve_voice,
     resolve_voice_from_map,
@@ -96,6 +99,16 @@ class VoicePresetResolutionTest(unittest.TestCase):
 
 
 class TTSPayloadTest(unittest.TestCase):
+    def test_higgs_prompt_keeps_first_word_before_emotion_tags(self) -> None:
+        voice = VoicePreset(preset="soft-young-fem", openai_voice="coral", rate=0.96)
+
+        prompt = build_higgs_prompt("그냥 학교 가도 아무 의미 없는 것 같아요.", voice)
+
+        self.assertTrue(prompt.startswith("그냥 "))
+        self.assertIn("<|emotion:helplessness|>", prompt)
+        self.assertIn("<|prosody:speed_slow|>", prompt)
+        self.assertNotIn("<|emotion:helplessness|> ", prompt)
+
     def test_payload_contains_openai_tts_fields_and_clamps_high_speed(self) -> None:
         voice = VoicePreset(
             preset="soft-young-fem",
@@ -219,6 +232,48 @@ class VoiceServiceStreamTest(unittest.IsolatedAsyncioTestCase):
         self.assertEqual(payload["instructions"], "Keep the tone grounded.")
         self.assertEqual([chunk.audio for chunk in chunks], [b"\x80\x80", b"\xff\x00"])
 
+    async def test_higgs_tts_uses_local_synthetic_voice_server_in_dev(self) -> None:
+        client = _CaptureTTSClient([b"RIFF", b"synthetic-wav"])
+        service = VoiceService(
+            api_key="",
+            environment="dev",
+            tts_provider="higgs",
+            higgs_base_url="http://127.0.0.1:9881",
+        )
+        service._higgs_client = client  # type: ignore[assignment]
+        voice = VoicePreset(preset="soft-young-fem", openai_voice="coral")
+
+        chunks = [
+            chunk
+            async for chunk in service.synthesize_stream(
+                "엄마한테 말하지 않는 거죠?",
+                voice,
+            )
+        ]
+
+        self.assertFalse(service.stt_available())
+        self.assertTrue(service.tts_available(voice))
+        self.assertEqual(service.tts_provider_for_voice(voice), "higgs")
+        self.assertEqual(service.tts_model_for_voice(voice), HIGGS_TTS_MODEL)
+        self.assertEqual(service.tts_media_type_for_voice(voice), "audio/wav")
+        self.assertEqual(len(client.calls), 1)
+        method, endpoint, payload = client.calls[0]
+        self.assertEqual((method, endpoint), ("POST", HIGGS_TTS_ENDPOINT))
+        self.assertIn("<|emotion:fear|>", str(payload["text"]))
+        self.assertEqual(payload["preset"], "soft-young-fem")
+        self.assertEqual(b"".join(chunk.audio for chunk in chunks), b"RIFFsynthetic-wav")
+
+    async def test_higgs_tts_is_fail_closed_outside_dev(self) -> None:
+        service = VoiceService(
+            api_key="",
+            environment="prod",
+            tts_provider="higgs",
+        )
+        voice = VoicePreset(preset="soft-young-fem", openai_voice="coral")
+
+        self.assertFalse(service.tts_available(voice))
+        self.assertEqual(service.tts_provider(), "disabled-non-dev")
+
     async def test_dev_p1_sample_tts_streams_local_mp3_without_openai_key(self) -> None:
         with TemporaryDirectory() as tmp:
             sample_dir = Path(tmp)