feat(tts): 正式合成支持克隆音色、保存到配音库改写 assets 素材体系、克隆接口支持 asset_id

1. TTS synthesize() 复用 preview 的克隆音色解析逻辑：voice_id 传克隆 profile UUID 时
   校验归属(403)/克隆完成状态(400 音色克隆尚未完成)，job.voice_id 存解析后的
   CosyVoice voice_id，voice_clone_profile_id 记录 profile id。
2. save_tts_job_to_library 改写为新素材体系：TTS 音频持久化独立副本到
   uploads/voice/tts/，创建 file_type=audio/status=ready 的 asset 挂用户 voice
   素材库（无则自动创建），duration/metadata(source=tts_job) 完整记录；
   返回结构保持 id/name/audio_url/duration 兼容前端；移除旧 voice_library 表写入。
3. voice clone 创建接口支持 asset_id 字段：校验素材归属与音频类型，用 asset
   storage_key 生成 7 天有效签名 URL 供 worker 下载，不依赖前端签名；
   source_audio_url 方式保留兼容。

补集成测试：synthesize 克隆音色 4 用例、save-to-library asset 断言、
clone from asset 5 用例（成功/404/403/非音频400/两参同传400）。

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

@@ -7,13 +7,15 @@ from typing import Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
+from app.core.storage import get_storage_service
 from app.dependencies import (
+    get_asset_library_repository,
+    get_asset_repository,
     get_audio_url_signer,
     get_cosyvoice_service,
     get_db_session,
-    get_user_repository,
+    get_project_repository,
     get_voice_clone_profile_repository,
-    get_voice_library_repository,
 )
 from app.schemas.tts import (
     ListTTSJobResponse,
@@ -32,7 +34,6 @@ from sqlalchemy.orm import Session
 from packages.adapters.sqlalchemy_impl.tts_job_repository import (
     SQLAlchemyTTSJobRepository,
 )
-from packages.adapters.sqlalchemy_impl.voice_library_repository import SQLAlchemyVoiceLibraryRepository
 from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
 from packages.application.tts_job.streaming_service import TTSStreamingService
 from packages.application.tts_job.use_cases import (
@@ -44,13 +45,12 @@ from packages.application.tts_job.use_cases import (
     TTSJobNotFoundError,
 )
 from packages.application.tts_job.workflow import TTSWorkflowService
-from packages.application.voice_library.commands import CreateVoiceLibraryCommand
-from packages.application.voice_library.use_cases import (
-    CreateVoiceLibraryUseCase,
-    QuotaExceededError,
-)
+from packages.domain import Asset, AssetLibrary, AssetLibraryKind, AssetStatus, ClassificationStatus
 from packages.domain.voice_presets import list_voices
-from packages.ports.user_repository import UserRepository
+from packages.ports.asset_library_repository import AssetLibraryRepository
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -134,27 +134,47 @@ def synthesize(
     """
     user_id = authenticated_user.user.id
 
-    # 校验 voice_clone_profile_id 归属（防止越权使用他人克隆音色）
-    if request.voice_clone_profile_id:
-        profile = voice_clone_repo.get(request.voice_clone_profile_id)
-        if profile is None:
+    # 解析 voice_id：前端可能传克隆音色 profile UUID（而非 CosyVoice voice_id），
+    # 与 /tts/preview 保持一致：命中 profile → 校验归属 → 取 CosyVoice voice_id
+    actual_voice_id = request.voice_id
+    voice_clone_profile_id = request.voice_clone_profile_id
+    resolved_profile = None
+    if actual_voice_id:
+        resolved_profile = voice_clone_repo.get(actual_voice_id)
+        if resolved_profile is not None:
+            voice_clone_profile_id = actual_voice_id
+
+    # 显式传了 voice_clone_profile_id（且与 voice_id 不同）时再查一次归属
+    if voice_clone_profile_id and (resolved_profile is None or resolved_profile.id != voice_clone_profile_id):
+        resolved_profile = voice_clone_repo.get(voice_clone_profile_id)
+        if resolved_profile is None:
             raise HTTPException(
                 status_code=status.HTTP_404_NOT_FOUND,
                 detail="Voice clone profile not found",
             )
-        if profile.user_id != user_id:
+
+    if resolved_profile is not None:
+        if resolved_profile.user_id != user_id:
             raise HTTPException(
                 status_code=status.HTTP_403_FORBIDDEN,
-                detail="Access denied to voice clone profile",
+                detail="无权访问该音色",
             )
+        if not resolved_profile.voice_id:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="音色克隆尚未完成，请稍后再试",
+            )
+        # 命中克隆音色：job 存解析后的 CosyVoice voice_id
+        if voice_clone_profile_id == actual_voice_id:
+            actual_voice_id = resolved_profile.voice_id
 
     use_case = CreateTTSJobUseCase(repository)
     job = use_case.execute(
         user_id=user_id,
         input_text=request.text,
-        voice_id=request.voice_id,
+        voice_id=actual_voice_id,
         voice_model=request.voice_model,
-        voice_clone_profile_id=request.voice_clone_profile_id,
+        voice_clone_profile_id=voice_clone_profile_id,
         metadata=request.metadata_,
     )
 
@@ -284,6 +304,41 @@ def delete_tts_job(
     return
 
 
+def _find_or_create_voice_library(
+    *,
+    user_id: str,
+    project_repository: ProjectRepository,
+    asset_library_repository: AssetLibraryRepository,
+) -> AssetLibrary:
+    """在用户可访问的项目中找到（或自动创建）voice 素材库。
+
+    与前端配音素材页逻辑一致：素材库挂在项目下，配音素材读取
+    getAssetsByKind("voice") → 用户所有可访问项目中的 voice 库。
+    优先使用已有 voice 库；没有则在第一个可访问项目中自动创建。
+    """
+    projects = project_repository.find_accessible_projects(user_id)
+    if not projects:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="没有可用的项目，请先创建项目后再保存配音素材",
+        )
+
+    for project in projects:
+        for lib in asset_library_repository.find_by_project(project.id):
+            kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
+            if kind == AssetLibraryKind.VOICE.value:
+                return lib
+
+    # 所有项目都没有 voice 库 → 在第一个可访问项目中自动创建默认配音素材库
+    project = projects[0]
+    library = AssetLibrary.create(
+        project_id=project.id,
+        name="配音素材库",
+        kind=AssetLibraryKind.VOICE,
+    )
+    return asset_library_repository.create(library)
+
+
 @router.post(
     "/jobs/{job_id}/save-to-library",
     response_model=SaveToLibraryResponse,
@@ -294,13 +349,17 @@ def save_tts_job_to_library(
     request: SaveToLibraryRequest = SaveToLibraryRequest(),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     tts_repository: SQLAlchemyTTSJobRepository = Depends(_get_repository),
-    voice_library_repository: SQLAlchemyVoiceLibraryRepository = Depends(get_voice_library_repository),
-    user_repository: UserRepository = Depends(get_user_repository),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    asset_library_repository: AssetLibraryRepository = Depends(get_asset_library_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
     sign_url=Depends(get_audio_url_signer),
 ) -> SaveToLibraryResponse:
-    """将已完成的 TTS 合成结果保存到配音库。
+    """将已完成的 TTS 合成结果保存到配音素材库（assets 表新素材体系）。
 
-    自动携带音色名、时长、语速等元信息。
+    流程：把 TTS 输出音频转存到用户素材 OSS 路径 → 创建 file_type=audio、
+    status=ready 的 asset（挂用户 voice 素材库）→ 返回前端可用结构。
+    配额策略与素材上传一致（上传/ingest 链路无额外配额拦截）。
     """
     user_id = authenticated_user.user.id
 
@@ -318,64 +377,107 @@ def save_tts_job_to_library(
             detail="TTS job is not completed yet",
         )
 
-    # 构建配音素材名称
+    if not job.output_audio_url and not job.output_audio_key:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="TTS job 缺少输出音频，无法保存",
+        )
+
+    # 素材名称
     name = request.name or f"TTS-{job.id[:8]}"
 
-    # 构建元信息
-    metadata_ = {
+    # 找到（或自动创建）用户 voice 素材库
+    library = _find_or_create_voice_library(
+        user_id=user_id,
+        project_repository=project_repository,
+        asset_library_repository=asset_library_repository,
+    )
+
+    # 转存音频到素材 OSS 路径（tts-outputs/ 下的产物归 TTS 任务所有，
+    # 素材独立持有副本，删除 TTS 任务不影响配音库素材）
+    import tempfile
+    from pathlib import Path
+
+    audio_format = (job.format or "mp3").strip() or "mp3"
+    content_type_map = {
+        "mp3": "audio/mpeg",
+        "wav": "audio/wav",
+        "pcm": "audio/pcm",
+        "opus": "audio/opus",
+    }
+    content_type = content_type_map.get(audio_format, "audio/mpeg")
+    storage_key = f"uploads/voice/tts/{job.id}.{audio_format}"
+
+    tmp_path: Path | None = None
+    try:
+        with tempfile.NamedTemporaryFile(suffix=f".{audio_format}", delete=False) as tmp:
+            tmp_path = Path(tmp.name)
+        # 优先用 OSS storage_key（走 oss2 SDK，私有 bucket 也可下载）；
+        # 兜底用 output_audio_url（旧任务可能没有 key）
+        download_source = job.output_audio_key or job.output_audio_url
+        downloaded = storage_service.download_asset(download_source, tmp_path)
+        if not downloaded or not tmp_path.exists() or tmp_path.stat().st_size == 0:
+            raise HTTPException(
+                status_code=status.HTTP_502_BAD_GATEWAY,
+                detail="TTS 音频下载失败，无法保存到配音库",
+            )
+        file_size = tmp_path.stat().st_size
+        storage_service.upload_file(tmp_path, storage_key, content_type=content_type)
+    except HTTPException:
+        raise
+    except Exception as e:
+        logger.error("TTS 音频转存素材失败: job_id=%s, error=%s", job.id, e, exc_info=True)
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            detail="TTS 音频转存失败，无法保存到配音库",
+        ) from e
+    finally:
+        if tmp_path and tmp_path.exists():
+            try:
+                tmp_path.unlink()
+            except OSError:
+                pass
+
+    # 构建素材元信息
+    metadata_: dict[str, object] = {
         "source": "tts_job",
         "tts_job_id": job.id,
         "format": job.format,
         "sample_rate": job.sample_rate,
+        "voice_id": job.voice_id,
+        "voice_name": job.voice_model or "",
     }
     if job.metadata:
-        # 保留原始 job 的有用元信息
         for key in ("speed", "language"):
             if key in job.metadata:
                 metadata_[key] = job.metadata[key]
 
-    # 获取用户套餐（用于配额检查）
-    user = user_repository.find_by_id(user_id)
-    plan_name = getattr(user, "subscription_plan", "free") if user else "free"
-
-    # 构建命令并执行
-    command = CreateVoiceLibraryCommand(
-        user_id=user_id,
+    asset = Asset.create(
+        project_id=library.project_id,
+        library_id=library.id,
         name=name,
-        text=job.input_text,
-        voice_provider="cosyvoice",
-        voice_id=job.voice_id,
-        voice_name=job.voice_model or "",
-        audio_url=job.output_audio_url,
-        duration=job.duration,
-        file_size=job.file_size,
-        status="completed",
-        project_id=job.project_id or "",
-        tags=[],
-        metadata_=metadata_,
+        storage_key=storage_key,
+        mime_type=content_type,
+        metadata=metadata_,
+        file_size=file_size,
+        duration=job.duration or None,
+        status=AssetStatus.READY,
+        classification_status=ClassificationStatus.PENDING,  # 音频不参与内容分类，保持 pending 与 ingest 链路一致
+        uploaded_by_user_id=user_id,
     )
-
-    use_case = CreateVoiceLibraryUseCase(voice_library_repository)
-    try:
-        item = use_case.execute(command, plan_name=plan_name or "free")
-    except QuotaExceededError as exc:
-        raise HTTPException(
-            status_code=status.HTTP_429_TOO_MANY_REQUESTS,
-            detail=f"配音库配额已满（{exc.used}/{exc.limit}），请升级套餐",
-        ) from exc
+    asset = asset_repository.create(asset)
 
     return SaveToLibraryResponse(
-        id=item.id,
-        name=item.name,
-        audio_url=sign_url(item.audio_url) if item.audio_url else "",
-        duration=item.duration,
-        voice_id=item.voice_id,
-        voice_name=item.voice_name,
-        status=item.status,
+        id=asset.id,
+        name=asset.name,
+        audio_url=sign_url(storage_key),
+        duration=asset.duration or 0.0,
+        voice_id=job.voice_id,
+        voice_name=job.voice_model or "",
+        status="completed",
     )
 
 
-
 @router.post("/preview", response_model=TTSPreviewResponse)
 def preview_tts(
     request: TTSPreviewRequest,

apps/api/app/api/routes/voice_clones.py

@@ -7,7 +7,13 @@ from typing import Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
-from app.dependencies import get_cosyvoice_service, get_voice_clone_profile_repository
+from app.core.storage import get_storage_service
+from app.dependencies import (
+    get_asset_repository,
+    get_cosyvoice_service,
+    get_project_repository,
+    get_voice_clone_profile_repository,
+)
 from app.schemas.voice_clone import (
     CreateVoiceCloneRequest,
     ListVoiceCloneResponse,
@@ -32,6 +38,9 @@ from packages.application.voice_clone.use_cases import (
 from packages.application.voice_clone.workflow import (
     VoiceCloneWorkflowService,
 )
+from packages.ports.asset_repository import AssetRepository
+from packages.ports.project_repository import ProjectRepository
+from packages.shared.storage import SharedStorageService
 
 logger = logging.getLogger(__name__)
 
@@ -83,23 +92,69 @@ def create_voice_clone(
     request: CreateVoiceCloneRequest,
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     workflow: VoiceCloneWorkflowService = Depends(_get_workflow_service),
+    asset_repository: AssetRepository = Depends(get_asset_repository),
+    project_repository: ProjectRepository = Depends(get_project_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
 ) -> VoiceCloneProfileResponse:
     """创建音色克隆任务。
 
     创建 VoiceCloneProfile → 提交 CosyVoice 克隆任务 → 触发 Celery 异步轮询。
-    如果有 source_audio_url，状态会变为 processing；否则保持 pending。
+    参考音频两种来源（二选一）：
+    - source_audio_url：前端直传后的音频 URL（兼容旧流程）
+    - asset_id：配音素材库中的音频素材，服务端用其 OSS storage_key 生成
+      预签名下载 URL（不依赖前端签名，避免签名过期导致克隆失败）
+    如果有参考音频，状态会变为 processing；否则保持 pending。
     """
     user_id = authenticated_user.user.id
+
+    source_audio_url = request.source_audio_url
+    clone_metadata = dict(request.metadata_ or {})
+
+    if request.asset_id:
+        if source_audio_url:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="asset_id 与 source_audio_url 只能传一个",
+            )
+        asset = asset_repository.find_by_id(request.asset_id)
+        if asset is None:
+            raise HTTPException(
+                status_code=status.HTTP_404_NOT_FOUND,
+                detail="素材不存在",
+            )
+        # 归属校验：素材挂在项目素材库下，用户必须能访问该项目
+        project = project_repository.find_by_id(asset.project_id)
+        if project is None or not project.can_access(user_id):
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="无权使用该素材",
+            )
+        # 类型校验：仅支持音频素材
+        file_type = asset.file_type if hasattr(asset, "file_type") else ""
+        if file_type != "audio":
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="仅支持音频素材进行音色克隆",
+            )
+        if not asset.storage_key:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="该素材缺少音频文件，无法用于克隆",
+            )
+        # 用 OSS storage_key 生成服务端预签名 URL（7 天有效，覆盖克隆重试周期）
+        source_audio_url = storage_service.get_download_url(asset.storage_key, expires_seconds=7 * 24 * 3600)
+        clone_metadata["source_asset_id"] = asset.id
+
     profile = workflow.start_clone(
         user_id=user_id,
         name=request.name,
         description=request.description,
-        source_audio_url=request.source_audio_url,
+        source_audio_url=source_audio_url,
         voice_model=request.voice_model,
         language=request.language,
         gender=request.gender,
         max_retries=request.max_retries,
-        metadata=request.metadata_,
+        metadata=clone_metadata,
     )
 
     # 如果 profile 处于 processing 且有 task_id，触发 Celery 异步轮询

apps/api/app/schemas/voice_clone.py

@@ -13,7 +13,8 @@ class CreateVoiceCloneRequest(BaseModel):
 
     name: str = Field(..., min_length=1, max_length=100, description="音色名称")
     description: str = Field("", description="音色描述")
-    source_audio_url: str = Field("", description="参考音频 URL")
+    source_audio_url: str = Field("", description="参考音频 URL（与 asset_id 二选一）")
+    asset_id: str = Field("", description="参考音频素材 ID（配音素材库中的音频 asset，与 source_audio_url 二选一）")
     voice_model: str = Field("", description="语音模型名称")
     language: str = Field("zh-CN", description="语言")
     gender: str = Field("unknown", description="性别")