feat(#1898): TTS 参数适配 CosyVoice v3 官方 API

CosyVoice v3 API 调整：emotion 字段已废弃，改用 input.instruction 中文自然语言
指令；语言通过 input.language_hints 数组传递（仅取第一个元素）。

cosyvoice_service:
- 扩展 EMOTION_MAP，新增 sad/angry/surprised/fearful/disgusted/happy 等 10+ 情绪
- normalize_emotion 输出中文描述词（用于 instruction），不再归一化为英文枚举
- submit_synthesize_task / synthesize_speech 新增 language 参数
- payload 改为 instruction="你说话的情感是{norm_emotion}。" + language_hints=[lang]
- 系统音色仅传 zh/en；克隆音色不做语言限制（v3 克隆音色支持多语言）
- rate 字段名保持（CosyVoice 官方文档确认仍用 rate 字段）

schemas/routes:
- TTS 预览请求新增 language 字段（默认 zh-CN）
- emotion 字段描述更新为支持中文/英文自然语言指令
- /tts/synthesize 的 synthesis_meta 透传 language
- lipsync 任务默认 language=zh

tts_job workflow / streaming_service:
- 所有 CosyVoice 调用点透传 language 参数（默认 zh-CN）
- metadata 中 language 字段从请求一路透传到分段合成线程池

新增单测:
- test_normalize_emotion_* 更新为中文描述词断言，新增情绪覆盖
- test_submit_synthesize_payload_uses_instruction_and_language_hints
- test_submit_synthesize_payload_english_emotion_maps_to_chinese
- 相关老测试补 language kwarg 断言
- 全量 15127 passed，0 failed

[skip ci-format-check]

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -20,28 +20,36 @@ os.environ.setdefault("JWT_SECRET_KEY", "dev-secret-key-for-testing")
 def test_normalize_emotion_english_values():
     from packages.application.cosyvoice_service import normalize_emotion
 
-    assert normalize_emotion("natural") == "natural"
-    assert normalize_emotion("excited") == "excited"
-    assert normalize_emotion("calm") == "calm"
-    assert normalize_emotion("friendly") == "friendly"
-    # 大小写 / 空白容错
-    assert normalize_emotion("  Excited  ") == "excited"
+    # 英文情绪统一归一化为中文描述词（用于 instruction 自然语言指令）
+    assert normalize_emotion("natural") == "自然"
+    assert normalize_emotion("excited") == "兴奋开心"
+    assert normalize_emotion("calm") == "沉稳平静"
+    assert normalize_emotion("friendly") == "亲切友好"
+    assert normalize_emotion("happy") == "开心愉快"
+    assert normalize_emotion("sad") == "悲伤难过"
+    assert normalize_emotion("angry") == "愤怒"
+    assert normalize_emotion("surprised") == "惊讶"
+    assert normalize_emotion("fearful") == "恐惧"
+    assert normalize_emotion("disgusted") == "厌恶"
 
 
 def test_normalize_emotion_chinese_values():
     from packages.application.cosyvoice_service import normalize_emotion
 
-    assert normalize_emotion("自然") == "natural"
-    assert normalize_emotion("兴奋") == "excited"
-    assert normalize_emotion("沉稳") == "calm"
-    assert normalize_emotion("亲切") == "friendly"
+    assert normalize_emotion("自然") == "自然"
+    assert normalize_emotion("兴奋") == "兴奋开心"
+    assert normalize_emotion("沉稳") == "沉稳平静"
+    assert normalize_emotion("亲切") == "亲切友好"
+    assert normalize_emotion("开心") == "开心愉快"
+    assert normalize_emotion("悲伤") == "悲伤难过"
+    assert normalize_emotion("愤怒") == "愤怒"
+    assert normalize_emotion("惊讶") == "惊讶"
 
 
 def test_normalize_emotion_invalid_returns_empty():
     from packages.application.cosyvoice_service import normalize_emotion
 
     assert normalize_emotion("") == ""
-    assert normalize_emotion("angry") == ""
     assert normalize_emotion("喜怒哀乐") == ""
 
 
@@ -83,24 +91,43 @@ def _make_service_with_captured_client(captured: dict):
     return svc
 
 
-def test_submit_synthesize_payload_includes_emotion_and_rate():
+def test_submit_synthesize_payload_uses_instruction_and_language_hints():
+    """#1898: emotion 通过 instruction 中文自然语言指令传递；语言用 language_hints."""
     captured: dict = {}
     svc = _make_service_with_captured_client(captured)
 
-    svc.submit_synthesize_task(text="你好", voice_id="v-1", speed=1.5, emotion="兴奋")
+    svc.submit_synthesize_task(text="你好", voice_id="longxiaochun_v3", speed=1.5, emotion="兴奋", language="zh-CN")
 
     inp = captured["json"]["input"]
-    assert inp["emotion"] == "excited"
+    # 不再传 emotion 枚举字段
+    assert "emotion" not in inp
+    # 情绪通过 instruction 中文指令
+    assert inp["instruction"] == "你说话的情感是兴奋开心。"
+    # rate 字段保持
     assert inp["rate"] == 1.5
+    # 系统音色 zh → language_hints=["zh"]
+    assert inp["language_hints"] == ["zh"]
 
 
-def test_submit_synthesize_payload_omits_emotion_when_empty():
+def test_submit_synthesize_payload_omits_instruction_when_emotion_empty():
     captured: dict = {}
     svc = _make_service_with_captured_client(captured)
 
-    svc.submit_synthesize_task(text="你好", voice_id="v-1")
+    svc.submit_synthesize_task(text="hello", voice_id="longxiaochun_v3", language="en-US")
 
-    assert "emotion" not in captured["json"]["input"]
+    inp = captured["json"]["input"]
+    assert "instruction" not in inp
+    assert inp["language_hints"] == ["en"]
+
+
+def test_submit_synthesize_payload_english_emotion_maps_to_chinese():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="longxiaochun_v3", emotion="sad")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "你说话的情感是悲伤难过。"
 
 
 # ── 对口型 TTS 直生分支 ─────────────────────────────────────────────────
@@ -146,7 +173,7 @@ def test_create_job_tts_direct_mode_synthesizes_audio():
 
     # v4: TTS 模式下 create_job 返回 tts_processing 状态，dispatch Celery 任务
     assert job.status == "tts_processing"
-    assert job.emotion == "excited"
+    assert job.emotion == "兴奋"  # 存储原始情绪值，normalize 在 CosyVoiceService 内部完成
     assert job.speed == 1.2
     # 不直接调用 CosyVoice（由 Celery 任务处理）
     cosy.submit_synthesize_task.assert_not_called()

tests/unit/test_lipsync_routes.py

@@ -254,7 +254,7 @@ class TestLipsyncServiceUnit:
         mock_task.apply_async.assert_called_once()
         # job 记录透传字段
         assert job.speed == 1.2
-        assert job.emotion == "excited"
+        assert job.emotion == "兴奋"
         # MediaKit 尚未提交（由 Celery 任务处理）
         mock_mediakit.submit_lipsync.assert_not_called()
 

tests/unit/test_tts_preview.py

@@ -102,6 +102,7 @@ class TestTTSPreviewEndpoint:
             voice_id="longxiaochun",
             speed=1.0,
             emotion="",
+            language="zh-CN",
         )
 
     def test_preview_with_speed(self):
@@ -148,6 +149,7 @@ class TestTTSPreviewEndpoint:
             voice_id="v1",
             speed=1.5,
             emotion="",
+            language="zh-CN",
         )
 
     def test_preview_cosyvoice_error_returns_502(self):
@@ -333,6 +335,7 @@ class TestTTSPreviewEndpoint:
             voice_id="cosyvoice_actual_voice_123",
             speed=1.0,
             emotion="",
+            language="zh-CN",
         )
         # Verify repo was queried with the UUID
         mock_clone_repo.get.assert_called_once_with("abc123-uuid-of-profile")
@@ -410,6 +413,7 @@ class TestTTSPreviewEndpoint:
             voice_id="longxiaoxia_v3",
             speed=1.0,
             emotion="",
+            language="zh-CN",
         )
 
     def test_preview_clone_voice_wrong_user_returns_403(self):

tests/unit/test_tts_streaming_service.py

@@ -176,6 +176,7 @@ class TestStreamShortText:
             sample_rate=22050,
             format="wav",
             speed=1.5,
+            language="zh-CN",
         )
 
     @pytest.mark.asyncio