xiaoxia-saas/apps/api/app/api/routes/lipsync.py

"""对口型 API 路由 — #1796 MediaKit 对口型, #1809 参数调整, #1845 配音前置.

接口：
  POST   /api/v1/lipsync/jobs          提交对口型任务（支持 TTS/直传/预合成 三种模式）
  GET    /api/v1/lipsync/jobs          任务列表
  GET    /api/v1/lipsync/jobs/{id}     任务详情
  POST   /api/v1/lipsync/jobs/{id}/refresh  刷新任务状态
  POST   /api/v1/lipsync/jobs/{id}/cancel   取消任务
  POST   /api/v1/lipsync/tts-preview   #1845 步骤1 TTS 预合成（同步 HTTP，~2-3s）
"""

from __future__ import annotations

import logging
from datetime import UTC

from app.auth import AuthenticatedUser, get_current_user
from app.dependencies import (
    get_db_session,
    get_voice_clone_profile_repository,
)
from app.schemas.lipsync import (
    AiAvatarTtsPreviewRequest,
    AiAvatarTtsPreviewResponse,
    CreateLipsyncJobRequest,
    LipsyncJobResponse,
)
from app.services.lipsync_service import LipsyncService
from app.services.mediakit_client import MediaKitError
from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query
from sqlalchemy.orm import Session

logger = logging.getLogger(__name__)

router = APIRouter()


def _get_service(
    db: Session = Depends(get_db_session),
    voice_clone_repo=Depends(get_voice_clone_profile_repository),
) -> LipsyncService:
    # voice_clone_repo 用于克隆音色 profile 解析
    return LipsyncService(
        db,
        voice_clone_repo=voice_clone_repo,
    )


# ── POST /jobs — 提交对口型任务 ───────────────────────────────────────────


@router.post("/jobs", response_model=LipsyncJobResponse, status_code=201)
def create_lipsync_job(
    body: CreateLipsyncJobRequest,
    current_user: AuthenticatedUser = Depends(get_current_user),
    db: Session = Depends(get_db_session),
    svc: LipsyncService = Depends(_get_service),
):
    """提交对口型任务.

    三种模式：
    - TTS 直生（旧版/降级）：传 {video_url, voice_id, script_text, speed?, emotion?}，
      后端 dispatch Celery 异步任务。
    - 直接音频：传 {video_url, audio_url}，后端同步下载+算timings+提交MediaKit。
    - 预合成音频（#1845 新主路径）：传 {video_url, audio_url, audio_duration, sentence_timings}，
      后端同步ffprobe+写入timings+直接提交MediaKit（~2-3s）。
    """
    user_id = current_user.user.id

    try:
        job = svc.create_job(
            user_id=user_id,
            video_url=body.video_url,
            audio_url=body.audio_url,
            audio_duration=body.audio_duration,
            sentence_timings=body.sentence_timings,
            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,
        )
    except ValueError as exc:
        raise HTTPException(status_code=400, detail=str(exc)) from exc
    except MediaKitError as exc:
        status_code = 502
        if exc.code in ("VoiceForbidden",):
            status_code = 403
        elif exc.code in ("InvalidInput", "TTSInvalidParam", "VoiceNotReady"):
            status_code = 400
        raise HTTPException(
            status_code=status_code,
            detail={
                "code": exc.code,
                "message": str(exc),
                "request_id": getattr(exc, "request_id", ""),
            },
        ) from exc
    except Exception as exc:
        logger.error("创建对口型任务异常: %s", exc, exc_info=True)
        raise HTTPException(
            status_code=400,
            detail=f"创建对口型任务失败: {exc}",
        ) from exc

    return job


# ── POST /tts-preview — #1845 步骤1 TTS 预合成 ──────────────────────────


@router.post("/tts-preview", response_model=AiAvatarTtsPreviewResponse)
def preview_tts(
    body: AiAvatarTtsPreviewRequest,
    current_user: AuthenticatedUser = Depends(get_current_user),
    db: Session = Depends(get_db_session),
    svc: LipsyncService = Depends(_get_service),
):
    """步骤1「生成配音」同步 TTS 预合成.

    同步执行 TTS 合成 → 下载音频 → ffprobe 时长 → 句子时间戳计算，
    不创建 LipsyncJob、不转存 OSS，直接返回 CosyVoice 临时 URL（~24h 有效）。
    耗时约 2-3 秒。
    """
    user_id = current_user.user.id

    try:
        result = svc.preview_tts(
            user_id=user_id,
            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:
        status_code = 400
        if exc.code in ("VoiceForbidden",):
            status_code = 403
        elif exc.code in ("TTSNoAudio",):
            status_code = 502
        raise HTTPException(
            status_code=status_code,
            detail={
                "code": exc.code,
                "message": str(exc),
            },
        ) from exc
    except Exception as exc:
        logger.error("TTS 预合成异常: %s", exc, exc_info=True)
        raise HTTPException(
            status_code=400,
            detail=f"TTS 合成失败: {exc}",
        ) from exc

    return result


# ── GET /jobs — 任务列表 ─────────────────────────────────────────────────


@router.get("/jobs", response_model=dict)
def list_lipsync_jobs(
    project_id: str = Query("", description="项目 ID 过滤"),
    status: str = Query("", description="状态过滤"),
    offset: int = Query(0, ge=0),
    limit: int = Query(20, ge=1, le=100),
    current_user: AuthenticatedUser = Depends(get_current_user),
    svc: LipsyncService = Depends(_get_service),
):
    """获取对口型任务列表."""
    items, total = svc.list_jobs(
        user_id=current_user.user.id,
        project_id=project_id,
        status=status,
        offset=offset,
        limit=limit,
    )
    return {
        "items": [LipsyncJobResponse.model_validate(j) for j in items],
        "total": total,
        "offset": offset,
        "limit": limit,
    }


# ── GET /jobs/{job_id} — 任务详情 ────────────────────────────────────────


@router.get("/jobs/{job_id}", response_model=LipsyncJobResponse)
def get_lipsync_job(
    job_id: str,
    background: BackgroundTasks,
    current_user: AuthenticatedUser = Depends(get_current_user),
    svc: LipsyncService = Depends(_get_service),
):
    """获取对口型任务详情."""
    job = svc.get_job(job_id, current_user.user.id)
    if job is None:
        raise HTTPException(status_code=404, detail="任务不存在")

    if job.status not in ("completed", "failed"):
        # 三层防御 ①：如果距上次更新超过 30 秒，同步刷新一次（避免 background task
        # 静默失败导致前端永远看到 running）；否则挂后台异步刷新（避免阻塞轮询）。
        from datetime import datetime as _dt

        _now = _dt.now(UTC)
        _upd = job.updated_at
        # DB 返回的 DateTime 列可能是 naive（取决于方言/驱动）：代码写入统一用
        # datetime.now(UTC)，经 SQLAlchemy 存入 TIMESTAMP WITHOUT TIMEZONE 后再
        # 读回就是 UTC wall clock 的 naive datetime，直接补 UTC tz 即可；避免
        # TypeError: can't subtract offset-naive and offset-aware datetimes。
        if _upd is not None and _upd.tzinfo is None:
            _upd = _upd.replace(tzinfo=UTC)
        _stale = _upd is None or (_now - _upd).total_seconds() > 30
        if _stale:
            try:
                refreshed = svc.refresh_job_status(job_id, current_user.user.id)
                if refreshed is not None:
                    job = refreshed
            except Exception as exc:  # noqa: BLE001
                logger.error("同步刷新对口型状态失败 job_id=%s err=%s", job_id, exc, exc_info=True)
                background.add_task(svc.refresh_job_status, job_id, current_user.user.id)
        else:
            background.add_task(svc.refresh_job_status, job_id, current_user.user.id)

    return job


# ── POST /jobs/{job_id}/refresh — 刷新状态 ───────────────────────────────


@router.post("/jobs/{job_id}/refresh", response_model=LipsyncJobResponse)
def refresh_lipsync_job(
    job_id: str,
    current_user: AuthenticatedUser = Depends(get_current_user),
    svc: LipsyncService = Depends(_get_service),
):
    """从 MediaKit 拉取最新状态并更新."""
    job = svc.refresh_job_status(job_id, current_user.user.id)
    if job is None:
        raise HTTPException(status_code=404, detail="任务不存在")
    return job


# ── POST /jobs/{job_id}/cancel — 取消任务 ────────────────────────────────


@router.post("/jobs/{job_id}/cancel", response_model=LipsyncJobResponse)
def cancel_lipsync_job(
    job_id: str,
    current_user: AuthenticatedUser = Depends(get_current_user),
    svc: LipsyncService = Depends(_get_service),
):
    """取消对口型任务（仅 pending/tts_processing/submitted/processing 状态可取消）."""
    job = svc.cancel_job(job_id, current_user.user.id)
    if job is None:
        raise HTTPException(status_code=404, detail="任务不存在")
    if job.status != "cancelled":
        raise HTTPException(
            status_code=400,
            detail=f"任务状态 {job.status} 不可取消，仅 pending/tts_processing/submitted/processing 可取消",
        )
    return job