alembic/versions/084_lipsync_jobs_style.py

@@ -0,0 +1,26 @@
+"""lipsync_jobs 新增 style 字段（TTS 语气风格）
+
+Revision ID: 084_lipsync_jobs_style
+Revises: 083_cover_title_config
+Create Date: 2026-09-21
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "084_lipsync_jobs_style"
+down_revision = "083_cover_title_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("style", sa.String(length=32), nullable=False, server_default=""),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("lipsync_jobs", "style")

apps/api/app/api/routes/lipsync.py

@@ -111,6 +111,8 @@ def create_lipsync_job(
             voice_id=body.voice_id,
             script_text=body.script_text,
             speed=body.speed,
+            style=body.style or "",
+            volume=body.volume if body.volume is not None else 50,
             emotion=body.emotion,
             enable_video_loop=body.enable_video_loop,
             project_id=body.project_id,
@@ -211,6 +213,8 @@ def preview_tts(
             voice_id=body.voice_id,
             script_text=body.script_text,
             speed=body.speed,
+            style=body.style or "",
+            volume=body.volume if body.volume is not None else 50,
             emotion=body.emotion,
         )
     except MediaKitError as exc:

apps/api/app/api/routes/tts.py

@@ -207,6 +207,9 @@ def synthesize(
     synthesis_meta = {
         "speed": request.speed,
         "emotion": request.emotion or "",
+        "style": request.style or "",
+        "volume": request.volume if request.volume is not None else 50,
+        "pitch": request.pitch if request.pitch is not None else 1.0,
         "language": request.language or "zh-CN",
     }
     if request.metadata_:
@@ -654,6 +657,9 @@ def preview_tts(
             text=request.text,
             voice_id=actual_voice_id,
             speed=request.speed,
+            style=request.style or "",
+            volume=request.volume if request.volume is not None else 50,
+            pitch=request.pitch,
             emotion=request.emotion,
             language=getattr(request, "language", "zh-CN"),
         )

apps/api/app/schemas/lipsync.py

@@ -29,6 +29,7 @@ class LipsyncJobResponse(BaseModel):
     voice_id: str = ""
     script_text: str = ""
     speed: float = 1.0
+    style: str = ""
     emotion: str = ""
     mediakit_task_id: str
     status: str
@@ -67,9 +68,14 @@ class CreateLipsyncJobRequest(BaseModel):
     voice_id: str = Field("", description="音色 ID（预置音色或克隆音色 profile UUID）")
     script_text: str = Field("", description="要合成的文案（直生模式必填，最长 5000 字符）")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
     emotion: str = Field(
         "",
-        description="情绪（英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted，或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶；空为默认自然）",
+        description="[deprecated] 旧情绪参数，内部映射为 style",
     )
 
     enable_video_loop: bool = Field(
@@ -123,10 +129,15 @@ class AiAvatarTtsPreviewRequest(BaseModel):
     voice_id: str = Field(..., min_length=1, max_length=128, description="音色 ID")
     script_text: str = Field(..., min_length=1, max_length=5000, description="要合成的文案")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
     emotion: str = Field(
         "neutral",
         max_length=32,
-        description="情绪（英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted，或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶；默认 neutral）",
+        description="[deprecated] 旧情绪参数，内部映射为 style；默认 neutral",
     )
 
 

apps/api/app/schemas/tts.py

@@ -16,9 +16,15 @@ class TTSSynthesizeRequest(BaseModel):
     output_name: str = Field("", description="输出文件名")
     language: str = Field("zh-CN", description="语言")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
+    pitch: Optional[float] = Field(None, ge=0.5, le=2.0, description="音调（0.5-2.0），默认 1.0")
     emotion: str = Field(
         "",
-        description="情绪（中文/英文：自然/兴奋/沉稳/亲切/开心/悲伤/愤怒/惊讶/恐惧/厌恶 等；通过 instruction 自然语言指令控制）",
+        description="[deprecated] 旧情绪参数，内部映射为 style；新接入请使用 style",
     )
     voice_model: str = Field("", description="语音模型名称")
     voice_clone_profile_id: str = Field("", description="关联的音色克隆档案 ID")
@@ -113,9 +119,14 @@ class TTSPreviewRequest(BaseModel):
     text: str = Field(..., min_length=1, max_length=200, description="合成文本，限制 200 字")
     voice_id: str = Field(..., min_length=1, description="音色 ID")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
-    emotion: str = Field("", description="情绪（中文/英文：自然/兴奋/沉稳/亲切/开心/悲伤/愤怒/惊讶/恐惧/厌恶 等）")
+    style: Optional[str] = Field(
+        None,
+        description="语气风格（natural/sweet/excited/professional/news/livestream），可选；优先级高于 emotion",
+    )
+    volume: Optional[int] = Field(None, ge=0, le=100, description="音量（0-100），默认 50")
+    emotion: str = Field("", description="[deprecated] 旧情绪参数，内部映射为 style")
     language: str = Field("zh-CN", description="语言（zh-CN/en-US 等）")
-    pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（预留，当前未使用）")
+    pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（0.5-2.0），默认 1.0")
 
 
 class TTSPreviewResponse(BaseModel):

apps/api/app/services/lipsync_service.py

@@ -111,6 +111,8 @@ class LipsyncService:
         script_text: str,
         speed: float,
         emotion: str,
+        style: str = "",
+        volume: int = 50,
     ) -> str:
         """TTS 直生：调 CosyVoice 合成音频并转存 OSS，返回可公网访问的音频 URL.
 
@@ -124,7 +126,9 @@ class LipsyncService:
                 text=script_text,
                 voice_id=actual_voice_id,
                 speed=speed,
-                emotion=emotion,  # normalize 在 CosyVoiceService 内部完成
+                style=style,
+                volume=volume,
+                emotion=emotion,
                 language="zh",
             )
         except CosyVoiceError as exc:
@@ -424,6 +428,8 @@ class LipsyncService:
         voice_id: str = "",
         script_text: str = "",
         speed: float = 1.0,
+        style: str = "",
+        volume: int = 50,
         emotion: str = "",
         enable_video_loop: bool = True,
         project_id: str = "",
@@ -472,6 +478,7 @@ class LipsyncService:
             voice_id=voice_id or "",
             script_text=script_text or "",
             speed=speed,
+            style=style or "",
             emotion=emotion or "",
             # 音频直传（含预合成）直接进入 pending（后续同步改为 submitted）；TTS 模式进入 tts_processing
             status="tts_processing" if is_tts_mode else "pending",
@@ -493,6 +500,8 @@ class LipsyncService:
                         voice_id,
                         script_text,
                         speed,
+                        style or "",
+                        volume,
                         emotion or "",
                     )
                 )
@@ -527,6 +536,8 @@ class LipsyncService:
         voice_id: str,
         script_text: str,
         speed: float = 1.0,
+        style: str = "",
+        volume: int = 50,
         emotion: str = "neutral",
     ) -> dict:
         """同步做 TTS 合成 + 下载 + ffprobe + 句子时间戳计算.

apps/api/app/tasks/lipsync_tts.py

@@ -82,7 +82,9 @@ def tts_synthesize_and_submit(
     voice_id: str,
     script_text: str,
     speed: float,
-    emotion: str,
+    style: str = "",
+    volume: int = 50,
+    emotion: str = "",
 ):
     """异步执行 TTS 合成 + OSS 转存 + MediaKit 提交.
 
@@ -159,6 +161,8 @@ def tts_synthesize_and_submit(
                 text=script_text,
                 voice_id=voice_id,
                 speed=speed,
+                style=style,
+                volume=volume,
                 emotion=emotion,
                 language="zh",
             )

packages/adapters/sqlalchemy_impl/models.py

@@ -711,6 +711,7 @@ class LipsyncJobModel(Base):
     voice_id = Column(String(200), nullable=False, default="")
     script_text = Column(Text, nullable=False, default="")
     speed = Column(Float, nullable=False, default=1.0)
+    style = Column(String(32), nullable=False, default="")
     emotion = Column(String(20), nullable=False, default="")
 
     # MediaKit 任务状态

packages/application/cosyvoice_service.py

@@ -26,12 +26,43 @@ from packages.shared.config import get_shared_settings
 logger = logging.getLogger(__name__)
 
 
-# CosyVoice v3 情绪通过 input.instruction 中文自然语言指令控制（不再使用枚举 emotion 字段）。
-# 官方文档：instruction 格式严格为 "你说话的情感是<情感值>。"，结尾中文句号不可省略；
-# 情感值必须是 7 种英文枚举之一：neutral/happy/sad/angry/surprised/fearful/disgusted。
+# ── Style（语气风格）→ CosyVoice instruct 自然语言指令 ──
+# 前端 PR#2002 传 6 种 style：natural/sweet/excited/professional/news/livestream。
+# style 是新的统一参数；emotion 为 deprecated 兼容别名，内部映射为 style。
 # 参考：https://help.aliyun.com/zh/model-studio/cosyvoice-voice-list
-# 前端可传英文枚举或中文标签（中立/开心/难过/生气/惊讶/恐惧/厌恶），统一归一化为英文枚举。
-# 映射表 key（不区分大小写）: 英文枚举/旧英文/中文标签 → 7 种标准英文枚举
+STYLE_INSTRUCTION_MAP: dict[str, str] = {
+    "natural": "用自然、平和的语气说话。",
+    "sweet": "用温柔甜美、亲切柔和的语气说话。",
+    "excited": "用兴奋、激动的语气说话。",
+    "professional": "用专业、正式的语气说话。",
+    "news": "用新闻播报的语气说话。",
+    "livestream": "用直播解说的语气说话。",
+}
+
+# 有效 style 值集合（供 schema / 校验使用）
+VALID_STYLES: frozenset[str] = frozenset(STYLE_INSTRUCTION_MAP.keys())
+
+# ── 旧 emotion → 新 style 兼容映射（方案 B：统一 style，emotion deprecated）──
+_EMOTION_TO_STYLE: dict[str, str] = {
+    "neutral": "natural",
+    "happy": "excited",
+    "sad": "sweet",
+    "angry": "excited",
+    "surprised": "excited",
+    "fearful": "sweet",
+    "disgusted": "natural",
+}
+
+# 严格格式系统音色：style → emotion 回退（用于无法使用自由文本指令的音色）
+_STYLE_TO_EMOTION: dict[str, str] = {
+    "natural": "neutral",
+    "sweet": "sad",
+    "excited": "happy",
+    "professional": "neutral",
+    # news / livestream 无直接对应 emotion，特殊处理
+}
+
+# ── 旧 emotion 映射表（deprecated，保留以兼容历史数据）──
 EMOTION_MAP: dict[str, str] = {
     # ── 7 种标准英文枚举（CosyVoice v3 官方支持的情感值）──
     "neutral": "neutral",
@@ -122,6 +153,70 @@ def build_emotion_instruction(voice_id: str, emotion_enum: str) -> str:
     return ""
 
 
+def resolve_style(style: str = "", emotion: str = "") -> str:
+    """统一解析 style 参数（方案 B）.
+
+    - style 有值且合法：直接使用（style 优先级最高）。
+    - style 为空但 emotion 有值：将旧 emotion 归一化后映射为 style。
+    - 两者皆空：返回空串（调用方不传 instruction）。
+
+    Args:
+        style: 新的语气风格（natural/sweet/excited/professional/news/livestream）
+        emotion: 旧的情绪参数（deprecated，内部映射为 style）
+
+    Returns:
+        解析后的 style 字符串；无需 instruct 时返回空串
+    """
+    s = (style or "").strip().lower()
+    if s:
+        if s in VALID_STYLES:
+            return s
+        logger.warning("未知的 style 值 %r，忽略 style 参数", style)
+    # 回退：emotion → style
+    norm = normalize_emotion(emotion)
+    if not norm:
+        return ""
+    mapped = _EMOTION_TO_STYLE.get(norm)
+    if not mapped:
+        logger.warning("emotion %r 无法映射到 style，跳过 instruct", norm)
+    return mapped or ""
+
+
+def build_style_instruction(voice_id: str, style: str) -> str:
+    """根据 voice 类型构造 style instruction.
+
+    - natural：返回空串（不额外加 instruct，使用 CosyVoice 默认自然语气）。
+    - 克隆/设计音色：使用中文自然语言指令（DashScope 允许任意自然语言）。
+    - 系统音色中支持 emotion instruct 的白名单音色：
+      若 style 可映射到 emotion，用严格格式 "你说话的情感是<emotion>。"；
+      news/livestream 尝试直接用中文 instruct（部分音色支持自由文本）。
+    - 其他系统音色：返回空串。
+
+    Args:
+        voice_id: CosyVoice voice 参数
+        style: 已通过 resolve_style() 解析的 style 值
+
+    Returns:
+        拼接好的 instruction 字符串；无需 instruct 时返回空串
+    """
+    if not style or style == "natural":
+        return ""
+    # 克隆音色：直接使用中文自然语言指令
+    if _is_cloned_voice(voice_id):
+        return STYLE_INSTRUCTION_MAP.get(style, "")
+    # 系统音色白名单：优先映射到严格 emotion 格式
+    if voice_id in _SYSTEM_VOICES_WITH_EMOTION_INSTRUCT:
+        emotion_val = _STYLE_TO_EMOTION.get(style)
+        if emotion_val:
+            return f"你说话的情感是{emotion_val}。"
+        # news/livestream 无 emotion 对应，尝试自由中文 instruct
+        desc = STYLE_INSTRUCTION_MAP.get(style, "")
+        if desc:
+            logger.info("音色 %s 使用自由文本 style instruct: %s", voice_id, desc)
+            return desc
+    return ""
+
+
 def normalize_emotion(emotion: str) -> str:
     """将前端情绪值归一化为 CosyVoice v3 官方英文枚举，用于拼入 instruction.
 
@@ -573,6 +668,8 @@ class CosyVoiceService:
         volume: int = 50,
         emotion: str = "",
         language: str = "zh",
+        style: str = "",
+        pitch: float = 1.0,
     ) -> dict:
         """提交语音合成任务（同步非流式，直接返回结果）.
 
@@ -586,9 +683,10 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
-            emotion: 情绪，英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted
-                     或前端中文标签（中立/开心/难过/生气/惊讶/恐惧/厌恶），兼容旧值
-                     natural/excited/calm/friendly；空串不传，未知值默认 neutral
+            style: 语气风格（natural/sweet/excited/professional/news/livestream），
+                   新的统一参数；优先级高于 emotion
+            pitch: 音调（0.5-2.0），1.0 为默认值
+            emotion: 【deprecated】旧情绪参数，内部通过 resolve_style() 映射为 style
             language: 语言代码（zh/en 等，默认 zh；系统音色仅 zh/en 传 language_hints）
 
         Returns:
@@ -617,11 +715,20 @@ class CosyVoiceService:
             "rate": speed,
             "volume": volume,
         }
-        # 情绪 → instruction（按 voice 类型选择格式）
-        norm_emotion = normalize_emotion(emotion)
-        emotion_instruction = build_emotion_instruction(voice_id, norm_emotion)
-        if emotion_instruction:
-            input_payload["instruction"] = emotion_instruction
+        # pitch: CosyVoice API 支持 [0.5, 2.0]，非默认值时才传
+        if pitch and pitch != 1.0:
+            input_payload["pitch"] = pitch
+        # style 优先：显式传 style 时走 build_style_instruction()；
+        # 无 style 时回退到旧的 emotion → build_emotion_instruction() 逻辑（向后兼容）
+        s = (style or "").strip().lower()
+        instruction = ""
+        if s:
+            instruction = build_style_instruction(voice_id, s)
+        if not instruction:
+            norm_emotion = normalize_emotion(emotion)
+            instruction = build_emotion_instruction(voice_id, norm_emotion)
+        if instruction:
+            input_payload["instruction"] = instruction
         # 语言 → language_hints 数组（仅取第一个元素生效）；
         # 系统音色（非克隆/非 voice_id 中包含下划线以外的短 ID）仅传 zh/en，其他语言不传避免报错
         norm_lang = normalize_language(language)
@@ -682,6 +789,8 @@ class CosyVoiceService:
         emotion: str = "",
         language: str = "zh",
         timeout: float = 120.0,
+        style: str = "",
+        pitch: float = 1.0,
     ) -> SynthesizeResult:
         """语音合成（同步非流式）.
 
@@ -695,6 +804,8 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
+            style: 语气风格（natural/sweet/excited/professional/news/livestream）
+            pitch: 音调（0.5-2.0），1.0 为默认值
             timeout: 超时时间（秒），保留参数兼容
 
         Returns:
@@ -714,6 +825,8 @@ class CosyVoiceService:
             volume=volume,
             emotion=emotion,
             language=language,
+            style=style,
+            pitch=pitch,
         )
 
         return SynthesizeResult(

packages/application/tts_job/workflow.py

@@ -146,6 +146,9 @@ class TTSWorkflowService:
             _meta = dict(job.metadata)
             _speed = float(_meta.get("speed", 1.0) or 1.0)
             _emotion = str(_meta.get("emotion", "") or "")
+            _style = str(_meta.get("style", "") or "")
+            _volume = int(_meta.get("volume", 50) or 50)
+            _pitch = float(_meta.get("pitch", 1.0) or 1.0)
             _language = str(_meta.get("language", "zh-CN") or "zh-CN")
             submit_result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
@@ -153,6 +156,9 @@ class TTSWorkflowService:
                 sample_rate=job.sample_rate,
                 format=job.format,
                 speed=_speed,
+                style=_style,
+                volume=_volume,
+                pitch=_pitch,
                 emotion=_emotion,
                 language=_language,
             )
@@ -290,6 +296,8 @@ class TTSWorkflowService:
             job_metadata = job.metadata or {}
             speed = float(job_metadata.get("speed", 1.0))
             volume = int(job_metadata.get("volume", 50))
+            style = str(job_metadata.get("style", "") or "")
+            pitch = float(job_metadata.get("pitch", 1.0))
             emotion = str(job_metadata.get("emotion", "") or "")
             language = str(job_metadata.get("language", "zh-CN") or "zh-CN")
 
@@ -299,7 +307,9 @@ class TTSWorkflowService:
                 sample_rate=job.sample_rate,
                 format=job.format,
                 speed=speed,
+                style=style,
                 volume=volume,
+                pitch=pitch,
                 emotion=emotion,
                 language=language,
             )
@@ -415,6 +425,9 @@ class TTSWorkflowService:
         _seg_meta = job.metadata or {}
         _seg_speed = float(_seg_meta.get("speed", 1.0) or 1.0)
         _seg_emotion = str(_seg_meta.get("emotion", "") or "")
+        _seg_style = str(_seg_meta.get("style", "") or "")
+        _seg_volume = int(_seg_meta.get("volume", 50) or 50)
+        _seg_pitch = float(_seg_meta.get("pitch", 1.0) or 1.0)
 
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
             future_to_idx = {}
@@ -426,6 +439,9 @@ class TTSWorkflowService:
                     sample_rate=job.sample_rate,
                     format=job.format,
                     speed=_seg_speed,
+                    style=_seg_style,
+                    volume=_seg_volume,
+                    pitch=_seg_pitch,
                     emotion=_seg_emotion,
                     language=_seg_meta.get("language", "zh-CN") or "zh-CN",
                 )
@@ -517,6 +533,8 @@ class TTSWorkflowService:
         job_metadata = job.metadata or {}
         speed = float(job_metadata.get("speed", 1.0))
         volume = int(job_metadata.get("volume", 50))
+        style = str(job_metadata.get("style", "") or "")
+        pitch = float(job_metadata.get("pitch", 1.0))
         emotion = str(job_metadata.get("emotion", "") or "")
 
         # 分段文本（用于缺失段重新合成）
@@ -551,7 +569,9 @@ class TTSWorkflowService:
                         sample_rate=job.sample_rate,
                         format=job.format,
                         speed=speed,
+                        style=style,
                         volume=volume,
+                        pitch=pitch,
                         emotion=emotion,
                         language=job.metadata.get("language", "zh-CN") if hasattr(job, "metadata") else "zh-CN",
                     )

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -6,6 +6,7 @@ CI 增量映射:
   ai_avatar_cover_service 智能选帧
 """
 
+import importlib
 import os
 from unittest.mock import MagicMock, patch
 
@@ -166,6 +167,116 @@ def test_submit_synthesize_payload_cloned_voice_english_emotion():
     assert inp["instruction"] == "Speak in a sad tone."
 
 
+# ── style（语气风格，#2002）────────────────────────────────────────────
+
+
+def test_style_natural_omits_instruction():
+    """style=natural 不加 instruct，使用 CosyVoice 默认自然语气。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="natural")
+
+    inp = captured["json"]["input"]
+    assert "instruction" not in inp
+
+
+def test_style_sweet_cloned_voice_uses_chinese_instruction():
+    """克隆音色 + style=sweet → 中文自然语言指令。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="sweet")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "用温柔甜美、亲切柔和的语气说话。"
+
+
+@pytest.mark.parametrize(
+    ("style", "expected_fragment"),
+    [
+        ("excited", "兴奋"),
+        ("professional", "专业"),
+        ("news", "新闻"),
+        ("livestream", "直播"),
+    ],
+)
+def test_style_values_cloned_voice(style, expected_fragment):
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style=style)
+
+    inp = captured["json"]["input"]
+    assert expected_fragment in inp["instruction"]
+
+
+def test_style_system_voice_uses_emotion_mapping():
+    """系统白名单音色 + style=sweet → 严格中文 emotion 格式（映射到 sad）。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="longanyang", style="sweet")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "你说话的情感是sad。"
+
+
+def test_style_takes_priority_over_emotion():
+    """同时传 style 和 emotion 时以 style 为准。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="myclone_voice", style="excited", emotion="sad")
+
+    inp = captured["json"]["input"]
+    assert "兴奋" in inp["instruction"]
+
+
+def test_unknown_style_ignored_falls_back_to_emotion():
+    """未知 style 值被忽略，回退到 emotion 逻辑。"""
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", style="nonexistent", emotion="sad")
+
+    inp = captured["json"]["input"]
+    assert inp["instruction"] == "Speak in a sad tone."
+
+
+def test_pitch_passed_only_when_non_default():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", pitch=1.5)
+
+    inp = captured["json"]["input"]
+    assert inp["pitch"] == 1.5
+
+
+def test_pitch_omitted_at_default():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="hi", voice_id="myclone_voice", pitch=1.0)
+
+    inp = captured["json"]["input"]
+    assert "pitch" not in inp
+
+
+def test_resolve_style_maps_emotion_to_style():
+    """旧 emotion 值通过 resolve_style 映射为 style。"""
+    mod = importlib.import_module("packages.application.cosyvoice_service")
+
+    assert mod.resolve_style(emotion="happy") == "excited"
+    assert mod.resolve_style(emotion="sad") == "sweet"
+    assert mod.resolve_style(emotion="neutral") == "natural"
+    assert mod.resolve_style(style="news") == "news"
+    # style 优先
+    assert mod.resolve_style(style="news", emotion="happy") == "news"
+    assert mod.resolve_style() == ""
+
+
 # ── 对口型 TTS 直生分支 ─────────────────────────────────────────────────
 
 

tests/unit/test_tts_preview.py

@@ -101,6 +101,9 @@ class TestTTSPreviewEndpoint:
             text="你好世界",
             voice_id="longxiaochun",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -148,6 +151,9 @@ class TestTTSPreviewEndpoint:
             text="测试",
             voice_id="v1",
             speed=1.5,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -334,6 +340,9 @@ class TestTTSPreviewEndpoint:
             text="克隆音色测试",
             voice_id="cosyvoice_actual_voice_123",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )
@@ -412,6 +421,9 @@ class TestTTSPreviewEndpoint:
             text="预设音色测试",
             voice_id="longxiaoxia_v3",
             speed=1.0,
+            style="",
+            volume=50,
+            pitch=1.0,
             emotion="",
             language="zh-CN",
         )