feat(character): add scenario voice trigger tags

config/character_config.py

@@ -1,7 +1,7 @@
 import yaml
 
 class CharacterConfig:
-    def __init__(self, name, color, sprite_prefix, gpt_model_path=None, sovits_model_path=None, refer_audio_path=None, prompt_text=None, prompt_lang=None, sprites=[], emotion_tags="", sprite_scale=1.0, character_setting="", speech_speed=1.0, speech_volume=1.0, pronunciation_map=None):
+    def __init__(self, name, color, sprite_prefix, gpt_model_path=None, sovits_model_path=None, refer_audio_path=None, prompt_text=None, prompt_lang=None, sprites=[], emotion_tags="", sprite_scale=1.0, character_setting="", speech_speed=1.0, speech_volume=1.0, pronunciation_map=None, scenarios=None):
         # 角色基本信息
         self.name = name
         self.color = color
@@ -10,6 +10,7 @@ def __init__(self, name, color, sprite_prefix, gpt_model_path=None, sovits_model
         self.character_setting = character_setting
         self.sprite_scale = sprite_scale
         self.emotion_tags = emotion_tags
+        self.scenarios = scenarios or []
         self.speech_speed = speech_speed
         self.speech_volume = speech_volume
         self.pronunciation_map = pronunciation_map or {}
@@ -65,6 +66,7 @@ def parse_dic(char_data):
                 speech_speed=char_data.get("speech_speed", 1.0),
                 speech_volume=char_data.get("speech_volume", 1.0),
                 pronunciation_map=char_data.get("pronunciation_map", None),
+                scenarios=char_data.get("scenarios", []),
             )
         return character
 

config/character_manager.py

@@ -400,9 +400,14 @@ def delete_all_sprites(self, character_name: str) -> Tuple[str, List[str], str]:
         # 清空角色属性
         character.sprites = []
         character.emotion_tags = ""
-
+        # 场景语音也放在同一个 VOICE_DIR 下，目录已删，YAML 引用必须同步清空
+        if hasattr(character, 'scenarios'):
+            for _sc in (character.scenarios or []):
+                _sc.voice_path = None
+            character.scenarios = []
+
         self._config_manager.save_characters_config()
-        
+
         return f"已删除 {character_name} 的所有立绘！", [], ""
 
 
@@ -646,6 +651,124 @@ def save_sprite_scale(self, name: str, scale: float) -> str:
         return f"找不到角色: {name}"
 
 
+    # ---------- 情景模块 ----------
+
+
+    def get_character_scenarios(self, character_name: str) -> List[dict]:
+        """获取角色的情景配置列表"""
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return []
+        if hasattr(character, 'model_dump'):
+            return character.model_dump().get('scenarios', [])
+        return getattr(character, 'scenarios', []) or []
+
+
+    def save_character_scenarios(self, character_name: str, scenarios: List[dict]) -> str:
+        """保存角色的完整情景配置（替换模式）"""
+        if not character_name:
+            return "请先选择角色！"
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return f"找不到角色: {character_name}"
+        from config.schema import CharacterScenario
+        validated = [CharacterScenario.model_validate(s) for s in scenarios]
+        character.scenarios = validated
+        self._config_manager.save_characters_config()
+        return "情景配置已保存"
+
+
+    def upload_scenario_voice(
+        self, character_name: str, scenario_index: int,
+        voice_file: str, voice_text: str, voice_type: str = "",
+    ) -> Tuple[str, Optional[str]]:
+        """为指定情景上传语音文件"""
+        if not character_name:
+            return "请先选择角色！", None
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return f"找不到角色: {character_name}", None
+        if not character.scenarios or scenario_index < 0 or scenario_index >= len(character.scenarios):
+            return "情景不存在！", None
+        scenario = character.scenarios[scenario_index]
+        if not voice_file:
+            return "请选择语音文件！", None
+
+        voice_dir = os.path.join(VOICE_DIR, character.sprite_prefix)
+        Path(voice_dir).mkdir(parents=True, exist_ok=True)
+        file_ext = Path(voice_file).suffix
+        # 用场景名的 hash 做文件名，避免删标签后索引复用覆盖旧音频
+        import re as _re
+        _safe = _re.sub(r'[^\w]', '_', scenario.name or f"tag{scenario_index}")[:40]
+        voice_filename = f"scenario_{_safe}{file_ext}"
+        voice_path = os.path.join(voice_dir, voice_filename)
+        shutil.copyfile(voice_file, voice_path)
+
+        scenario.voice_path = voice_path
+        scenario.voice_text = voice_text
+        if voice_type:
+            scenario.voice_type = voice_type
+
+        self._config_manager.save_characters_config()
+        return f"语音已上传到情景 {scenario_index+1}", voice_path
+
+
+    def delete_scenario_voice(
+        self, character_name: str, scenario_index: int,
+    ) -> str:
+        """删除指定情景的语音文件和引用"""
+        if not character_name:
+            return "请先选择角色！"
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return f"找不到角色: {character_name}"
+        if not character.scenarios or scenario_index < 0 or scenario_index >= len(character.scenarios):
+            return "情景不存在！"
+        scenario = character.scenarios[scenario_index]
+        # 先持久化清空引用，再删文件；防止中途崩溃导致 YAML 残留失效路径
+        old_voice_path = scenario.voice_path
+        scenario.voice_path = None
+        scenario.voice_text = None
+        scenario.voice_type = None
+        self._config_manager.save_characters_config()
+        if old_voice_path:
+            try:
+                os.remove(old_voice_path)
+            except OSError:
+                pass
+        return "语音已删除"
+
+
+    def save_scenario_voice_text(
+        self, character_name: str, scenario_index: int, voice_text: str,
+    ) -> str:
+        """单独保存情景的语音文本"""
+        if not character_name:
+            return "请先选择角色！"
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return f"找不到角色: {character_name}"
+        scenario = character.scenarios[scenario_index]
+        scenario.voice_text = voice_text
+        self._config_manager.save_characters_config()
+        return "语音文本已保存"
+
+
+    def save_scenario_voice_type(
+        self, character_name: str, scenario_index: int, voice_type: str,
+    ) -> str:
+        """单独保存情景的语音类型（preset/reference）"""
+        if not character_name:
+            return "请先选择角色！"
+        character = self._config_manager.get_character_by_name(character_name)
+        if not character:
+            return f"找不到角色: {character_name}"
+        scenario = character.scenarios[scenario_index]
+        scenario.voice_type = voice_type if voice_type else None
+        self._config_manager.save_characters_config()
+        return "语音类型已保存"
+
+
     def load_characters_from_file(self) -> Tuple[str, List[List[str]]]:
         """
         重新加载人物设定文件。

config/schema.py

@@ -39,18 +39,28 @@ class Sprite(BaseModel):
     voice_path: Optional[FilePath] = Field(None, description="对应的语音文件的路径 (可选)")
     voice_text: Optional[str] = Field(None, description="语音对应的文本内容 (可选, 存在于某些条目中)")
 
+
+class CharacterScenario(BaseModel):
+    """角色的单个情景（情绪或动作），由 AI 自主识别"""
+    name: str = Field(..., description="情景名称（英文，如 happy / kiss / angry）")
+    voice_path: Optional[str] = Field(None, description="语音文件路径（可选）")
+    voice_type: Optional[str] = Field(None, description="语音类型: preset 或 reference")
+    voice_text: Optional[str] = Field(None, description="语音对应文本内容")
+
+
 class Character(BaseModel):
     """单个角色配置的实体模型"""
     # 角色基本信息
     name: str = Field(..., description="角色名称")
     color: str = Field(..., description="角色对话框或名字的颜色")
     sprite_prefix: str = Field(..., description="立绘文件名的通用前缀")
-    
+
     # 列表中可能包含 Sprite 模型，也可能只是原始字典
     sprites: List[Union[Sprite, dict]] = Field(default_factory=list, description="角色的立绘和对应语音的列表")
     character_setting: DefaultIfNone[str] = Field(default="", description="角色背景、性格和语言习惯的详细描述")
     sprite_scale: DefaultIfNone[float] = Field(default=1.0, description="立绘的缩放比例 (默认值 1.0)")
     emotion_tags: DefaultIfNone[str] = Field(default="", description="情绪标签和对应的立绘编号描述")
+    scenarios: List[CharacterScenario] = Field(default_factory=list, description="情景配置列表（由 AI 自主识别情绪或动作）")
 
     # gpt-sovits 相关的配置
     gpt_model_path: Optional[str] = Field('', description="角色 GPT 模型的路径 (可选)")

core/handlers/tts_message_handler.py

@@ -23,6 +23,7 @@
 from sdk.messages import LLMDialogMessage, TTSOutputMessage
 from core.runtime.app_runtime import get_app_runtime, tts_emit_to_ui_queue
 from i18n import tr as tr_i18n
+import re
 
 _config = ConfigManager()
 
@@ -31,6 +32,46 @@ def get_character_by_name(name: str):
     return _config.get_character_by_name(name)
 
 
+def _read_scenarios_from_yaml(name_s: str) -> list:
+    """直接从 YAML 读取角色的 scenarios，避免跨进程缓存。"""
+    try:
+        _chars_path = Path("data/config/characters.yaml")
+        if not _chars_path.is_file():
+            return []
+        with open(_chars_path, "r", encoding="utf-8") as _fh:
+            _data = yaml.safe_load(_fh) or []
+        for _c in _data:
+            if _c.get("name") == name_s:
+                _scenarios = _c.get("scenarios") or []
+                # 归一化 voice_type 为小写，防御旧数据大小写不一致
+                for _s in _scenarios:
+                    if isinstance(_s, dict) and _s.get("voice_type"):
+                        _s["voice_type"] = str(_s["voice_type"]).strip().lower()
+                return _scenarios
+    except Exception:
+        pass
+    return []
+
+
+def _match_voice_tag(speech: str, scenarios: list):
+    """检测台词中的 [tagname] 标签，匹配角色情景配置。
+    返回匹配到的 scenario dict，未匹配返回 None。
+    同时返回去除标签后的纯净台词。
+    """
+    if not speech or not scenarios:
+        return None, speech
+    tags = re.findall(r'\[(\w+)\]', speech)
+    if not tags:
+        return None, speech
+    for tag in tags:
+        for s in scenarios:
+            s_name = s.name if hasattr(s, 'name') else s.get('name', '')
+            if s_name.lower() == tag.lower():
+                clean = re.sub(r'\s*\[\w+\]\s*', ' ', speech).strip()
+                return s, clean
+    return None, speech
+
+
 def _post_tts_busy(text: str) -> None:
     try:
         get_app_runtime().ui_update_manager.post_busy_bar(text, 0.0)
@@ -135,10 +176,42 @@ def handle(self, msg: LLMDialogMessage) -> None:
         asset_id = msg.asset_id
         text_processor = rt.text_processor
         speech_text = speech
+
+        # 语音触发标签：[tagname] 匹配角色情景，触发对应语音
+        # 必须在翻译处理之前匹配，因为标签只存在于原始 speech 中
+        _tag_ref_audio = None
+        _tag_prompt = None
+        _scenarios = _read_scenarios_from_yaml(name_s)
+        _matched, _clean_speech = _match_voice_tag(speech_text, _scenarios)
+        if _matched:
+            _m_name = _matched.name if hasattr(_matched, 'name') else _matched.get('name', '?')
+            _m_vt = _matched.voice_type if hasattr(_matched, 'voice_type') else _matched.get('voice_type')
+            _m_vp = _matched.voice_path if hasattr(_matched, 'voice_path') else _matched.get('voice_path')
+            _m_vtext = _matched.voice_text if hasattr(_matched, 'voice_text') else _matched.get('voice_text')
+            print(f"[VoiceTag] 匹配标签 [{_m_name}] type={_m_vt} vp={_m_vp}")
+            if _m_vt == "preset" and _m_vp:
+                print(f"[VoiceTag] 预设语音：跳过 TTS，直接播放 {_m_vp}")
+                audio_path = Path(_m_vp).resolve().as_posix()
+                _hide_tts_busy()
+                tts_emit_to_ui_queue(
+                    name_s, _m_vtext or _clean_speech, str(asset_id), audio_path,
+                    is_system_message=False, effect=msg.effect,
+                )
+                return
+            elif _m_vt == "reference" and _m_vp:
+                print(f"[VoiceTag] 参考语音：用 {_m_vp} 作为 TTS 参考音色")
+                speech_text = _clean_speech
+                _tag_ref_audio = _m_vp
+                _tag_prompt = _m_vtext or ""
+
         if translate:
             text_processor = None
             speech_text = rt.text_processor.remove_parentheses(translate)
             speech_text = rt.text_processor.replace_names(speech_text)
+
+        # 剥离语音标签 [xxx]，不显示在台词中
+        speech_text = re.sub(r'\s*\[\w+\]\s*', ' ', speech_text).strip()
+
         audio_path = ""
         if rt.tts_manager:
             _post_tts_busy(tr_i18n("desktop.tts_busy_synthesizing", name=name_s))
@@ -161,11 +234,23 @@ def handle(self, msg: LLMDialogMessage) -> None:
                 prompt_text = character_config.prompt_text
                 try:
                     sprite_data = character_config.sprites[sprite_id]
-                    if sprite_data.get("voice_text", None):
-                        ref_audio_path = Path(sprite_data.get("voice_path")).resolve().as_posix()
-                        prompt_text = sprite_data.get("voice_text")
+                    if isinstance(sprite_data, dict):
+                        _vt = sprite_data.get("voice_text")
+                    else:
+                        _vt = getattr(sprite_data, "voice_text", None)
+                    if _vt:
+                        if isinstance(sprite_data, dict):
+                            ref_audio_path = Path(sprite_data.get("voice_path") or "").resolve().as_posix()
+                        else:
+                            ref_audio_path = Path(getattr(sprite_data, "voice_path", "") or "").resolve().as_posix()
+                        prompt_text = _vt
                 except Exception:
                     print("没有立绘")
+                # 语音触发标签匹配到 reference 时，用标签配置覆盖参考音频
+                if _tag_ref_audio:
+                    ref_audio_path = Path(_tag_ref_audio).resolve().as_posix()
+                if _tag_prompt:
+                    prompt_text = _tag_prompt
                 if text_processor:
                     speech_text = text_processor.remove_parentheses(speech_text)
 

core/handlers/ui_message_handler.py

@@ -173,7 +173,8 @@ def handle(self, out: TTSOutputMessage) -> None:
         ui.hide_busy_bar()
         ch = _play()
         character_name = out.name
-        speech = out.text or ""
+        import re
+        speech = re.sub(r'\s*\[\w+\]\s*', '', out.text or "").strip()
         sprite_id = out.asset_id
         audio_path = out.audio_path
         if audio_path:

frontend/pnpm-workspace.yaml

@@ -0,0 +1,2 @@
+allowBuilds:
+  esbuild: true