xiaoxia-saas/apps/api/app/services/lipsync_service.py

"""对口型 Service — #1796 MediaKit 对口型业务逻辑, #1809 参数调整, #1845 配音前置.

职责：
- 创建/查询对口型任务
- 三输入模式：
  1. TTS 直生（voice_id + script_text）→ 走 Celery 异步（降级路径）
  2. 直接音频（audio_url，前端未传 timings）→ 同步下载 + 算 timings + 提交 MediaKit
  3. 预合成音频（audio_url + sentence_timings，#1845 新主路径）→ 同步 ffprobe 校验时长 +
     写入前端传来的 timings → 直接提交 MediaKit（~2-3s）
- 调用 MediaKit 客户端提交异步任务
- 轮询更新任务状态（中间状态同步 DB，成片转存自家 OSS）
- 用户隔离（每个用户只能操作自己的任务）
"""

from __future__ import annotations

import io
import logging
import uuid
from datetime import datetime, timezone
from typing import Optional
from urllib.parse import urlparse

from app.services.mediakit_client import (
    STATUS_COMPLETED,
    STATUS_FAILED,
    STATUS_RUNNING,
    MediaKitClient,
    MediaKitError,
    get_mediakit_client,
)

# Celery 异步任务：TTS 合成 + MediaKit 提交（降级路径）
from app.tasks.lipsync_tts import tts_synthesize_and_submit
from sqlalchemy.orm import Session

from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
from packages.application.cosyvoice_service import CosyVoiceError, normalize_emotion
from packages.domain.sentence_timings import (
    compute_sentence_timings,
    probe_audio_duration,
)
from packages.shared.storage import get_shared_storage_service
from packages.shared.url_security import ALLOWED_AUDIO_MIME_TYPES, safe_download_bytes

logger = logging.getLogger(__name__)

# 传给 MediaKit GPU worker / 回给前端播放的 OSS 预签名有效期：7 天。
MEDIAKIT_URL_TTL_SECONDS = 7 * 24 * 3600


class LipsyncService:
    """对口型任务 Service."""

    def __init__(
        self,
        db: Session,
        client: Optional[MediaKitClient] = None,
        cosyvoice_service=None,
        voice_clone_repo=None,
    ):
        self.db = db
        self.client = client or get_mediakit_client()
        self._cosyvoice = cosyvoice_service
        self._voice_clone_repo = voice_clone_repo

    def _get_cosyvoice(self):
        """延迟获取 CosyVoiceService（与 tts 路由一致，含 OSS 预签名配置）."""
        if self._cosyvoice is None:
            from app.dependencies import get_cosyvoice_service

            self._cosyvoice = get_cosyvoice_service()
        return self._cosyvoice

    def _resolve_voice_id(self, voice_id: str, user_id: str) -> str:
        """将克隆音色 profile UUID 解析为 CosyVoice voice_id。

        与 /tts/synthesize 保持一致：命中 profile → 校验归属 → 返回其 voice_id；
        未命中（预置音色 ID 或克隆 CosyVoice voice_id）原样返回。
        """
        if not voice_id:
            return ""
        if self._voice_clone_repo is None:
            try:
                from app.dependencies import get_voice_clone_profile_repository

                self._voice_clone_repo = get_voice_clone_profile_repository(self.db)
            except Exception:
                return voice_id
        try:
            profile = self._voice_clone_repo.get(voice_id)
        except Exception:
            return voice_id
        if profile is None:
            return voice_id
        if getattr(profile, "user_id", "") != user_id:
            raise MediaKitError("无权访问该音色", code="VoiceForbidden")
        if not getattr(profile, "voice_id", ""):
            raise MediaKitError("音色克隆尚未完成，请稍后再试", code="VoiceNotReady")
        return profile.voice_id

    def _synthesize_and_persist_audio(
        self,
        *,
        user_id: str,
        job_id: str,
        voice_id: str,
        script_text: str,
        speed: float,
        emotion: str,
    ) -> 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.

        Raises:
            MediaKitError: 合成失败
        """
        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
        cosyvoice = self._get_cosyvoice()
        try:
            result = cosyvoice.submit_synthesize_task(
                text=script_text,
                voice_id=actual_voice_id,
                speed=speed,
                emotion=normalize_emotion(emotion),
            )
        except CosyVoiceError as exc:
            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
        except ValueError as exc:
            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc

        temp_url = result.get("audio_url", "")
        if not temp_url:
            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")

        # 转存到自家 OSS，避免临时 URL 过期导致 MediaKit 拉取失败
        try:
            audio_data = safe_download_bytes(
                temp_url,
                purpose="lipsync_tts_audio",
                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
                timeout=60.0,
            )
            storage = get_shared_storage_service()
            storage_key = f"lipsync-tts/{user_id}/{job_id}.mp3"
            permanent_url = storage.upload_file(io.BytesIO(audio_data), storage_key, content_type="audio/mpeg")
            logger.info("对口型 TTS 音频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
            return permanent_url
        except Exception as exc:
            logger.warning("TTS 音频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
            return temp_url

    def _submit_audio_direct(
        self,
        *,
        job: LipsyncJobModel,
        supplied_timings: Optional[list] = None,
        supplied_duration: Optional[float] = None,
    ) -> None:
        """音频直传模式（包含 #1845 预合成路径）：同步下载 → ffprobe → timings → 提交 MediaKit.

        直接在 HTTP 请求内完成，不走 Celery。job.status 成功后置为 submitted。
        失败时把 job 标成 failed 并 commit，然后抛 MediaKitError。

        Args:
            job: 已 commit 的 LipsyncJobModel（audio_url / video_url 已写入）
            supplied_timings: 前端传来的预合成 timings（可选，可信时直接用）
            supplied_duration: 前端传来的预合成时长（可选，用于优先避免重复探测）
        """
        # 1. 下载音频
        audio_data: bytes | None = None
        try:
            audio_data = safe_download_bytes(
                job.audio_url,
                purpose="lipsync_direct_audio",
                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
                timeout=60.0,
            )
            logger.info(
                "[lipsync] 直传音频下载完成: job_id=%s size=%d",
                job.id,
                len(audio_data) if audio_data else 0,
            )
        except Exception as exc:
            logger.warning("[lipsync] 直传音频下载失败，跳过 timings 计算: job_id=%s err=%s", job.id, exc)

        # 2. ffprobe 探测时长（优先用前端传入的预合成时长，但以 ffprobe 为准做兜底校验）
        audio_duration = 0.0
        if audio_data:
            audio_duration = probe_audio_duration(audio_data)
        if audio_duration <= 0 and supplied_duration and supplied_duration > 0:
            audio_duration = supplied_duration
            logger.info(
                "[lipsync] ffprobe 失败，使用前端传入的预合成时长: job_id=%s duration=%.2f", job.id, audio_duration
            )

        # 3. 句子时间戳：优先用前端预合成传入的 timings（后端预合成接口已经算过，可信）；
        #    否则若音频下载成功则重算；否则不设置（不阻塞主流程）
        timings: Optional[list] = None
        if supplied_timings:
            timings = supplied_timings
            logger.info("[lipsync] 使用前端预合成句子时间戳: job_id=%s sentences=%d", job.id, len(timings))
        elif audio_data and audio_duration > 0 and job.script_text:
            try:
                timings = compute_sentence_timings(audio_data, job.script_text, audio_duration)
                logger.info(
                    "[lipsync] 后端重算句子时间戳: job_id=%s sentences=%d duration=%.2f",
                    job.id,
                    len(timings) if timings else 0,
                    audio_duration,
                )
            except Exception as exc:
                logger.warning("[lipsync] 句子时间戳计算失败（不阻塞）: job_id=%s err=%s", job.id, exc)

        if timings:
            job.sentence_timings = timings

        # 4. 签名 URL 并提交 MediaKit
        video_url = self._sign_media_url(job.video_url)
        signed_audio_url = self._sign_media_url(job.audio_url)
        job.audio_url = signed_audio_url

        try:
            result = self.client.submit_lipsync(
                video_url=video_url,
                audio_url=signed_audio_url,
                enable_video_loop=job.enable_video_loop,
                client_token=job.id,
            )
            job.mediakit_task_id = result["task_id"]
            job.status = "submitted"
            job.submitted_at = datetime.now(timezone.utc)
            self.db.commit()
            logger.info(
                "[lipsync] 直传音频已提交 MediaKit: job_id=%s task_id=%s",
                job.id,
                result["task_id"],
            )
        except MediaKitError as exc:
            job.status = "failed"
            job.error_message = str(exc)
            job.error_code = exc.code
            logger.error("[lipsync] 直传音频提交 MediaKit 失败: job_id=%s err=%s", job.id, exc)
            self.db.commit()
            raise

    # ── 创建任务 ──────────────────────────────────────────────────────────

    def create_job(
        self,
        *,
        user_id: str,
        video_url: str,
        audio_url: str = "",
        audio_duration: Optional[float] = None,
        sentence_timings: Optional[list] = None,
        voice_id: str = "",
        script_text: str = "",
        speed: float = 1.0,
        emotion: str = "",
        enable_video_loop: bool = True,
        project_id: str = "",
    ) -> LipsyncJobModel:
        """创建对口型任务.

        三种输入模式：
        - TTS 直生：voice_id + script_text（audio_url 留空）
          → 创建 DB 记录（状态 tts_processing），dispatch Celery 异步任务（降级路径）。
          API 响应 <1s。
        - 直接音频：audio_url 非空 + 无 sentence_timings
          → 同步下载音频 + 重算 timings + 提交 MediaKit（几秒完成）。
        - 预合成音频（#1845 新主路径）：audio_url 非空 + 传 sentence_timings
          → 同步 ffprobe 校验时长 + 写入 timings + 提交 MediaKit（~2-3s）。

        Raises:
            MediaKitError: 参数校验失败或 MediaKit 提交失败
        """
        # 0. 输入校验
        is_pre_synth = bool(audio_url) and bool(sentence_timings)
        bool(audio_url) and not is_pre_synth
        is_tts_mode = not bool(audio_url)

        if is_tts_mode:
            if not (voice_id and script_text):
                raise MediaKitError(
                    "必须提供 audio_url 或 voice_id+script_text",
                    code="InvalidInput",
                )
            # TTS 模式：在 HTTP 请求中同步校验音色归属，快速失败
            self._resolve_voice_id(voice_id, user_id)
        elif is_pre_synth:
            # 预合成模式：script_text 可空（因为 timings 已自带句子文本），但仍建议传
            if not isinstance(sentence_timings, list) or len(sentence_timings) == 0:
                raise MediaKitError("预合成模式 sentence_timings 不能为空", code="InvalidInput")

        # 1. 创建数据库记录
        job_id = str(uuid.uuid4())
        job = LipsyncJobModel(
            id=job_id,
            user_id=user_id,
            project_id=project_id,
            video_url=video_url,
            audio_url=audio_url,
            enable_video_loop=enable_video_loop,
            voice_id=voice_id or "",
            script_text=script_text or "",
            speed=speed,
            emotion=normalize_emotion(emotion) if is_tts_mode else (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",
        )
        self.db.add(job)
        self.db.flush()

        # ⚠️ 必须先 commit 再发 Celery 任务 / 后续同步操作，避免事务竞态
        self.db.commit()
        self.db.refresh(job)

        if is_tts_mode:
            # 2a. TTS 模式：dispatch Celery 异步任务处理 TTS 合成 + MediaKit 提交（降级路径）
            try:
                tts_synthesize_and_submit.apply_async(
                    args=(
                        job_id,
                        user_id,
                        voice_id,
                        script_text,
                        speed,
                        normalize_emotion(emotion),
                    )
                )
            except Exception as exc:
                logger.exception(
                    "Celery 任务提交失败，TTS 任务已创建但未触发执行: job_id=%s err=%s",
                    job_id,
                    exc,
                )
                job.status = "failed"
                job.error_message = f"Celery 任务投递失败: {exc}"
                job.error_code = "AsyncDispatchFailed"
                job.updated_at = datetime.now(timezone.utc)
                self.db.commit()
        else:
            # 2b/2c. 直接音频 / 预合成音频：同步路径
            self._submit_audio_direct(
                job=job,
                supplied_timings=sentence_timings,
                supplied_duration=audio_duration,
            )
            self.db.refresh(job)

        return job

    # ── TTS 预合成（#1845 步骤1「生成配音」同步接口使用） ──────────────────

    def preview_tts(
        self,
        *,
        user_id: str,
        voice_id: str,
        script_text: str,
        speed: float = 1.0,
        emotion: str = "natural",
    ) -> dict:
        """同步做 TTS 合成 + 下载 + ffprobe + 句子时间戳计算.

        不创建 LipsyncJob、不转存 OSS，直接返回 CosyVoice 临时 URL（~24h 有效期）。
        耗时约 2-3 秒，由前端在步骤1点「生成配音」时同步等待。

        Returns:
            {"audio_url": str, "duration": float, "sentence_timings": list[dict]}

        Raises:
            MediaKitError: TTS 合成失败 / 下载失败 / ffprobe 失败
        """
        # 1. 音色解析（校验克隆音色归属）
        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
        cosyvoice = self._get_cosyvoice()

        # 2. TTS 合成（同步，~2-3s）
        try:
            result = cosyvoice.submit_synthesize_task(
                text=script_text,
                voice_id=actual_voice_id,
                speed=speed,
                emotion=normalize_emotion(emotion),
            )
        except CosyVoiceError as exc:
            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
        except ValueError as exc:
            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc

        temp_url = result.get("audio_url", "")
        if not temp_url:
            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")

        # 3. 下载音频到内存（用于 ffprobe + 静音检测）
        try:
            audio_data = safe_download_bytes(
                temp_url,
                purpose="tts_preview_audio",
                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
                timeout=60.0,
            )
        except Exception as exc:
            logger.warning("[tts-preview] TTS 音频下载失败，仍返回 audio_url: user_id=%s err=%s", user_id, exc)
            return {
                "audio_url": temp_url,
                "duration": 0.0,
                "sentence_timings": [],
            }

        # 4. ffprobe 时长
        duration = probe_audio_duration(audio_data)
        if duration <= 0:
            logger.warning("[tts-preview] ffprobe 未返回有效时长，timings 留空: user_id=%s", user_id)
            return {
                "audio_url": temp_url,
                "duration": 0.0,
                "sentence_timings": [],
            }

        # 5. 句子时间戳
        timings = compute_sentence_timings(audio_data, script_text, duration)

        logger.info(
            "[tts-preview] TTS 预合成完成: user_id=%s duration=%.2f sentences=%d",
            user_id,
            duration,
            len(timings),
        )
        return {
            "audio_url": temp_url,
            "duration": round(duration, 2),
            "sentence_timings": timings,
        }

    # ── 查询任务 ──────────────────────────────────────────────────────────

    def get_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
        """获取任务详情（用户隔离）."""
        return (
            self.db.query(LipsyncJobModel)
            .filter(LipsyncJobModel.id == job_id, LipsyncJobModel.user_id == user_id)
            .first()
        )

    def list_jobs(
        self,
        *,
        user_id: str,
        project_id: str = "",
        status: str = "",
        offset: int = 0,
        limit: int = 20,
    ) -> tuple[list[LipsyncJobModel], int]:
        """获取任务列表（分页 + 用户隔离）."""
        query = self.db.query(LipsyncJobModel).filter(LipsyncJobModel.user_id == user_id)
        if project_id:
            query = query.filter(LipsyncJobModel.project_id == project_id)
        if status:
            query = query.filter(LipsyncJobModel.status == status)

        total = query.count()
        items = query.order_by(LipsyncJobModel.created_at.desc()).offset(offset).limit(limit).all()
        return items, total

    # ── 更新任务状态（轮询） ──────────────────────────────────────────────

    def refresh_job_status(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
        """从 MediaKit 拉取最新状态并更新本地记录."""
        job = self.get_job(job_id, user_id)
        if job is None:
            return None

        # 终态不需要再轮询
        if job.status in (STATUS_COMPLETED, "failed"):
            return job

        # 未提交的任务不轮询
        if not job.mediakit_task_id:
            return job

        try:
            status_data = self.client.get_task_status(job.mediakit_task_id)
        except MediaKitError as exc:
            logger.error("轮询对口型任务状态失败 [%s]: %s", job_id, exc)
            return job

        mk_status = status_data.get("status", STATUS_RUNNING)
        logger.info("MediaKit 对口型状态 [%s]: %s", job_id, mk_status)

        if mk_status == STATUS_COMPLETED:
            result = status_data.get("result", {})
            job.status = STATUS_COMPLETED
            temp_url = result.get("video_url", "")
            job.output_video_url = temp_url
            job.output_duration = result.get("duration", 0.0)
            job.completed_at = datetime.now(timezone.utc)
            job.updated_at = datetime.now(timezone.utc)
            self.db.commit()
            # 异步转存自家 OSS
            try:
                from app.tasks.lipsync_tts import persist_output_video_task

                persist_output_video_task.apply_async(args=(job_id, user_id, temp_url))
            except Exception as exc:
                logger.warning(
                    "提交输出视频异步转存任务失败，保留临时 URL: job_id=%s err=%s",
                    job_id,
                    exc,
                )
            self.db.refresh(job)
            return job
        elif mk_status == STATUS_FAILED:
            error = status_data.get("error", {})
            job.status = "failed"
            job.error_message = error.get("message", "任务执行失败")
            job.error_code = error.get("code", "TaskFailed")
            job.completed_at = datetime.now(timezone.utc)
        else:
            if isinstance(mk_status, str) and mk_status:
                job.status = mk_status
        job.updated_at = datetime.now(timezone.utc)
        self.db.commit()
        self.db.refresh(job)
        return job

    def _persist_output_video(self, temp_url: str, job_id: str, user_id: str) -> str:
        """将 MediaKit 输出的临时视频 URL 转存到自家 OSS. 失败时回退返回原始临时 URL."""
        if not temp_url:
            return ""
        try:
            import httpx

            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
                resp = client.get(temp_url)
                resp.raise_for_status()
                data = resp.content
            storage = get_shared_storage_service()
            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
            logger.info("对口型输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
            return self._sign_media_url(permanent_url) or temp_url
        except Exception as exc:
            logger.warning("对口型输出视频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
            return temp_url

    def _sign_media_url(self, url: str) -> str:
        """对自家 OSS 私有桶 URL 重签长有效期预签名."""
        if not url:
            return url
        try:
            storage = get_shared_storage_service()
            public_base = getattr(storage, "public_url", "")
            if not isinstance(public_base, str) or not public_base:
                return url
            own_host = urlparse(public_base).netloc.lower()
            host = urlparse(url).netloc.lower()
            if not own_host or host != own_host:
                return url  # 外部临时链接原样透传
            signed = storage.get_download_url(url, expires_seconds=MEDIAKIT_URL_TTL_SECONDS)
            return signed or url
        except Exception as exc:
            logger.warning("对口型 URL 重签失败，原样返回: url_prefix=%s err=%s", url[:80], exc)
            return url

    # ── 取消任务 ──────────────────────────────────────────────────────────

    def cancel_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
        """取消任务（仅 pending/tts_processing/submitted 状态可取消）."""
        job = self.get_job(job_id, user_id)
        if job is None:
            return None

        if job.status in ("pending", "tts_processing", "submitted"):
            job.status = "cancelled"
            job.updated_at = datetime.now(timezone.utc)
            self.db.commit()
            self.db.refresh(job)

        return job