fix: correct extract-voice API path from /tts to /voices prefix

Backend route is mounted under /voices, not /tts

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

@@ -2,7 +2,9 @@
 
 from __future__ import annotations
 
+import json
 import logging
+import subprocess
 import tempfile
 from pathlib import Path
 from typing import Any, Optional
@@ -430,6 +432,8 @@ def save_tts_job_to_library(
     storage_key = f"uploads/voice/tts/{job.id}.{audio_format}"
 
     tmp_path: Path | None = None
+    audio_duration: float | None = None
+    file_size = 0
     try:
         with tempfile.NamedTemporaryFile(suffix=f".{audio_format}", delete=False) as tmp:
             tmp_path = Path(tmp.name)
@@ -445,6 +449,23 @@ def save_tts_job_to_library(
             )
         file_size = tmp_path.stat().st_size
         storage_service.upload_file(tmp_path, storage_key, content_type=content_type)
+
+        # 从音频文件提取时长（ffprobe），作为 job.duration 的兜底
+        try:
+            proc = subprocess.run(
+                [
+                    "ffprobe", "-v", "quiet", "-print_format", "json",
+                    "-show_format", str(tmp_path),
+                ],
+                capture_output=True, text=True, timeout=10,
+            )
+            if proc.returncode == 0:
+                fmt = json.loads(proc.stdout).get("format", {})
+                dur = float(fmt.get("duration", 0))
+                if dur > 0:
+                    audio_duration = dur
+        except Exception:
+            logger.warning("ffprobe 提取时长失败: job_id=%s", job.id, exc_info=True)
     except HTTPException:
         raise
     except Exception as e:
@@ -482,7 +503,7 @@ def save_tts_job_to_library(
         mime_type=content_type,
         metadata=metadata_,
         file_size=file_size,
-        duration=job.duration or None,
+        duration=job.duration or audio_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,

apps/api/app/api/routes/voices.py

@@ -6,12 +6,26 @@
 from __future__ import annotations
 
 import logging
+import shutil
+import subprocess
+import tempfile
 import time
+from pathlib import Path
 from typing import Literal, Optional
+from uuid import uuid4
 
 from app.api.routes._helpers import get_user_plan
 from app.auth import AuthenticatedUser, get_current_user
-from app.dependencies import get_audio_url_signer, get_cosyvoice_service, get_db_session, get_user_repository
+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_project_repository,
+    get_user_repository,
+)
 from app.schemas.voice import (
     PresetVoiceItemResponse,
     PresetVoiceListResponse,
@@ -24,7 +38,7 @@ from app.schemas.voice_library import (
     UpdateVoiceLibraryRequest,
     VoiceLibraryItemResponse,
 )
-from fastapi import APIRouter, Depends, HTTPException, Query, Response, status
+from fastapi import APIRouter, Depends, File, Form, HTTPException, Query, Response, UploadFile, status
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.voice_clone_profile_repository import SQLAlchemyVoiceCloneProfileRepository
@@ -40,8 +54,12 @@ from packages.application.voice_library.use_cases import (
     QuotaExceededError,
     UpdateVoiceLibraryUseCase,
 )
+from packages.domain import Asset, AssetStatus
+from packages.domain.classification import AssetLibraryKind, ClassificationStatus
+from packages.domain.entities import AssetLibrary
 from packages.domain.preset_voices import PRESET_VOICES, get_preset_voice_by_id
 from packages.ports.user_repository import UserRepository
+from packages.shared.storage import SharedStorageService
 
 router = APIRouter()
 logger = logging.getLogger(__name__)
@@ -507,3 +525,243 @@ def delete_voice(
     if not deleted:
         raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Voice not found")
     return
+
+
+# ── 提取视频配音 ─────────────────────────────────────────────────────
+
+# 支持的视频格式
+EXTRACT_VIDEO_MIMES = frozenset({"video/mp4", "video/quicktime", "video/webm", "video/x-msvideo"})
+MAX_EXTRACT_SIZE = 500 * 1024 * 1024  # 500MB
+
+
+@router.post(
+    "/extract-voice",
+    status_code=status.HTTP_201_CREATED,
+)
+def extract_voice_from_video(
+    file: UploadFile = File(...),
+    project_id: str = Form(...),
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    project_repository=Depends(get_project_repository),
+    asset_library_repository=Depends(get_asset_library_repository),
+    asset_repository=Depends(get_asset_repository),
+    storage_service: SharedStorageService = Depends(get_storage_service),
+    sign_url=Depends(get_audio_url_signer),
+):
+    """从上传的视频中提取人声配音。
+
+    流程：
+    1. 接收视频文件（mp4/mov/webm）
+    2. ffmpeg 提取音频 + 降噪 + 编码为 mp3
+    3. 上传到 OSS，创建 Asset 记录到配音素材库
+    4. 返回素材信息（时长、文件大小、URL）
+    """
+    user_id = authenticated_user.user.id
+
+    # 校验文件类型
+    content_type = file.content_type or ""
+    if content_type and content_type not in EXTRACT_VIDEO_MIMES:
+        # 兜底：按扩展名判断
+        ext = (file.filename or "").rsplit(".", 1)[-1].lower()
+        ext_to_mime = {"mp4": "video/mp4", "mov": "video/quicktime", "webm": "video/webm", "avi": "video/x-msvideo"}
+        if ext not in ext_to_mime:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="仅支持 mp4/mov/webm/avi 格式的视频文件",
+            )
+        content_type = ext_to_mime[ext]
+
+    # 找到（或自动创建）用户 voice 素材库（复用 TTS 的逻辑）
+    library = _find_or_create_voice_library_for_extract(
+        user_id=user_id,
+        project_repository=project_repository,
+        asset_library_repository=asset_library_repository,
+    )
+
+    tmp_dir = None
+    try:
+        tmp_dir = Path(tempfile.mkdtemp(prefix="voice_extract_"))
+        video_path = tmp_dir / f"input_{uuid4().hex[:8]}_{file.filename or 'video.mp4'}"
+        audio_path = tmp_dir / f"output_{uuid4().hex[:8]}.mp3"
+
+        # 保存上传的视频到临时文件
+        with open(video_path, "wb") as f:
+            total = 0
+            while chunk := file.file.read(1024 * 1024):  # 1MB chunks
+                total += len(chunk)
+                if total > MAX_EXTRACT_SIZE:
+                    raise HTTPException(
+                        status_code=status.HTTP_413_REQUEST_ENTITY_TOO_LARGE,
+                        detail="视频文件过大，最大支持 500MB",
+                    )
+                f.write(chunk)
+
+        if video_path.stat().st_size == 0:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="视频文件为空")
+
+        # ffmpeg: 提取音频 + 降噪 + 编码 mp3
+        # 滤镜链：highpass(去低频噪声) → afftdn(FFT降噪) → lowpass(去高频噪声)
+        ffmpeg_cmd = [
+            "ffmpeg",
+            "-y",
+            "-i",
+            str(video_path),
+            "-vn",  # 不要视频
+            "-af",
+            "highpass=f=80,afftdn=nf=-25:tn=1,lowpass=f=8000",
+            "-acodec",
+            "libmp3lame",
+            "-ab",
+            "192k",
+            "-ar",
+            "44100",
+            "-ac",
+            "1",  # 单声道（人声足够）
+            str(audio_path),
+        ]
+
+        result = subprocess.run(
+            ffmpeg_cmd,
+            capture_output=True,
+            timeout=300,  # 5 分钟超时
+        )
+
+        if result.returncode != 0:
+            stderr_text = result.stderr.decode("utf-8", errors="replace")[-500:]
+            logger.error("ffmpeg 提取配音失败: %s", stderr_text)
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+                detail="视频音频提取失败，可能该视频没有音轨或格式不支持",
+            )
+
+        if not audio_path.exists() or audio_path.stat().st_size == 0:
+            raise HTTPException(
+                status_code=status.HTTP_422_UNPROCESSABLE_ENTITY,
+                detail="音频提取结果为空",
+            )
+
+        # 获取音频时长
+        duration = _get_audio_duration(audio_path)
+        file_size = audio_path.stat().st_size
+
+        # 上传到 OSS
+        audio_ext = "mp3"
+        storage_key = f"uploads/voice/extracted/{uuid4().hex}.{audio_ext}"
+        storage_service.upload_file(audio_path, storage_key, content_type="audio/mpeg")
+
+        # 创建 Asset 记录
+        original_name = (file.filename or "video").rsplit(".", 1)[0]
+        asset_name = f"{original_name}-配音"
+
+        asset = Asset.create(
+            project_id=library.project_id,
+            library_id=library.id,
+            name=asset_name,
+            storage_key=storage_key,
+            mime_type="audio/mpeg",
+            metadata={
+                "source": "video_extract",
+                "original_video": file.filename or "unknown",
+            },
+            file_size=file_size,
+            duration=duration,
+            status=AssetStatus.READY,
+            classification_status=ClassificationStatus.PENDING,
+            uploaded_by_user_id=user_id,
+        )
+        asset = asset_repository.create(asset)
+
+        return {
+            "id": asset.id,
+            "name": asset.name,
+            "audio_url": sign_url(storage_key),
+            "duration": duration,
+            "file_size": file_size,
+            "status": "completed",
+            "source": "video_extract",
+        }
+
+    except HTTPException:
+        raise
+    except subprocess.TimeoutExpired:
+        raise HTTPException(
+            status_code=status.HTTP_504_GATEWAY_TIMEOUT,
+            detail="视频处理超时，请尝试较短的视频",
+        ) from None
+    except Exception as e:
+        logger.exception("提取视频配音失败: %s", e)
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="提取配音失败，请稍后重试",
+        ) from e
+    finally:
+        # 清理临时文件
+        if tmp_dir and Path(tmp_dir).exists():
+            shutil.rmtree(tmp_dir, ignore_errors=True)
+
+
+def _find_or_create_voice_library_for_extract(*, user_id, project_repository, asset_library_repository):
+    """为用户找到或创建 voice 素材库（与 TTS 保存逻辑一致）。"""
+    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",
+        )
+
+    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
+
+    # 自动创建
+    from sqlalchemy.exc import IntegrityError
+
+    project = projects[0]
+    library = AssetLibrary.create(
+        project_id=project.id,
+        name="配音素材库",
+        kind=AssetLibraryKind.VOICE,
+    )
+    try:
+        return asset_library_repository.create(library)
+    except IntegrityError:
+        session = getattr(asset_library_repository, "session", None)
+        if session is not None:
+            try:
+                session.rollback()
+            except Exception:
+                pass
+        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
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail="配音素材库创建失败",
+        ) from None
+
+
+def _get_audio_duration(audio_path: Path) -> float:
+    """用 ffprobe 获取音频时长（秒）。"""
+    try:
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "quiet",
+                "-show_entries",
+                "format=duration",
+                "-of",
+                "csv=p=0",
+                str(audio_path),
+            ],
+            capture_output=True,
+            timeout=10,
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            return float(result.stdout.strip())
+    except (ValueError, subprocess.TimeoutExpired):
+        pass
+    return 0.0

apps/api/app/services/plan_generator_service.py

@@ -234,6 +234,11 @@ class PlanGeneratorService:
                 # 有缓存的素材片段起点从随机镜头段选取，无缓存走随机起点兜底
                 asset_scene_points = self._fetch_asset_scene_points(asset_ids)
 
+                # 正式生成也随机重排片段顺序（降重，默认开启无开关）
+                # smart_match 决定选哪些素材，shuffle 只改变分配到 clips 的顺序
+                asset_ids = list(asset_ids)  # 复制避免修改调用方原列表
+                random.shuffle(asset_ids)
+
         distribute_assets(
             clips,
             asset_ids,

apps/web/src/api/tts/index.ts

@@ -28,4 +28,5 @@ export {
   deleteTTSJob,
   getTtsVoices,
   previewTts,
+  extractVideoVoice,
 } from "./jobs"

apps/web/src/api/tts/jobs.ts

@@ -70,3 +70,56 @@ export const previewTts = async (data: TTSPreviewRequest): Promise<TTSPreviewRes
   const response = await apiClient.post<TTSPreviewResponse>("/tts/preview", data)
   return response.data
 }
+
+/**
+ * 从视频中提取配音（上传视频 → 后端提取人声 → 保存到配音素材库）
+ * 支持 mp4/mov/webm 格式
+ */
+export const extractVideoVoice = async (
+  file: File,
+  onProgress?: (percent: number) => void,
+): Promise<{ asset_id: string; duration: number }> => {
+  const formData = new FormData()
+  formData.append("file", file)
+
+  return new Promise((resolve, reject) => {
+    const xhr = new XMLHttpRequest()
+    xhr.open("POST", "/api/v1/voices/extract-voice")
+
+    // 携带认证 token（从 localStorage 获取，与 apiClient 拦截器一致）
+    const token = localStorage.getItem("access_token")
+    if (token) {
+      xhr.setRequestHeader("Authorization", `Bearer ${token}`)
+    }
+
+    xhr.timeout = 10 * 60 * 1000 // 10 分钟超时
+
+    xhr.upload.onprogress = (e) => {
+      if (e.lengthComputable && onProgress) {
+        onProgress(Math.round((e.loaded / e.total) * 100))
+      }
+    }
+
+    xhr.onload = () => {
+      if (xhr.status >= 200 && xhr.status < 300) {
+        try {
+          resolve(JSON.parse(xhr.responseText))
+        } catch {
+          reject(new Error("服务器返回数据解析失败"))
+        }
+      } else {
+        try {
+          const err = JSON.parse(xhr.responseText)
+          reject(new Error(err.detail || err.message || `提取失败: HTTP ${xhr.status}`))
+        } catch {
+          reject(new Error(`提取失败: HTTP ${xhr.status}`))
+        }
+      }
+    }
+
+    xhr.onerror = () => reject(new Error("网络错误，请检查网络连接"))
+    xhr.ontimeout = () => reject(new Error("上传超时（10分钟），请检查网络或尝试更小的文件"))
+
+    xhr.send(formData)
+  })
+}

apps/web/src/pages/voices/VoiceLibrary.tsx

@@ -16,7 +16,12 @@
  */
 import React, { useCallback, useEffect, useState } from "react"
 import { useSearchParams } from "react-router-dom"
-import { UploadOutlined, AudioOutlined, RobotOutlined } from "@ant-design/icons"
+import {
+  UploadOutlined,
+  AudioOutlined,
+  RobotOutlined,
+  VideoCameraOutlined,
+} from "@ant-design/icons"
 import { Button } from "@/components/ui"
 import PageHead from "@/components/layout/PageHead"
 import { type AssetItem } from "@/api/assets"
@@ -35,6 +40,8 @@ import { useTtsSynthesize } from "./hooks/useTtsSynthesize"
 import { useVoiceUpload } from "./hooks/useVoiceUpload"
 import { useMaterialDelete } from "./hooks/useMaterialDelete"
 import { useMaterialBatchDelete } from "./hooks/useMaterialBatchDelete"
+import { useVideoExtract } from "./hooks/useVideoExtract"
+import VideoExtractModal from "./components/VideoExtractModal"
 import "./voices.css"
 
 let toastIdSeq = 0
@@ -159,6 +166,18 @@ const VoiceLibrary: React.FC = () => {
     handleUploadClose,
   } = useVoiceUpload({ showToast })
 
+  // ── 提取视频配音 ──────────────────────────────────────
+  const {
+    extractOpen,
+    extractFile,
+    extractProgress,
+    isExtracting,
+    setExtractOpen,
+    handleFileSelect: handleExtractFileSelect,
+    handleExtract,
+    handleExtractClose,
+  } = useVideoExtract({ showToast })
+
   // ── URL 参数自动打开上传弹窗 ────────────────────────────
   const [searchParams, setSearchParams] = useSearchParams()
 
@@ -200,6 +219,14 @@ const VoiceLibrary: React.FC = () => {
       >
         上传音频
       </Button>
+      <Button
+        buttonType="primary"
+        buttonSize="sm"
+        icon={<VideoCameraOutlined />}
+        onClick={() => setExtractOpen(true)}
+      >
+        提取视频配音
+      </Button>
       <Button
         buttonType="ghost"
         buttonSize="sm"
@@ -300,7 +327,18 @@ const VoiceLibrary: React.FC = () => {
         />
       )}
 
-      {/* ── 弹窗集合 ──────────────────────────────────── */}
+      {/* ── 视频提取配音弹窗 ─────────────────────────────── */}
+      <VideoExtractModal
+        open={extractOpen}
+        file={extractFile}
+        progress={extractProgress}
+        isExtracting={isExtracting}
+        onClose={handleExtractClose}
+        onFileSelect={handleExtractFileSelect}
+        onExtract={handleExtract}
+      />
+
+      {/* ── 弹窗集合 ─────────────────────────────────── */}
       <VoiceModals
         cloneModalOpen={cloneModalOpen}
         onCloneClose={() => setCloneModalOpen(false)}

apps/web/src/pages/voices/components/MaterialVoiceTab.tsx

@@ -136,6 +136,8 @@ export const MaterialVoiceTab: React.FC<MaterialVoiceTabProps> = ({
             const material = mapAssetToMaterial(asset)
             // duration 优先取顶层（后端从 metadata 提取），兜底 metadata
             const cardDuration = asset.duration || material.duration || 0
+            // AI 生成素材标识（metadata.source === "tts_job"）
+            const isAiMaterial = (asset.metadata as Record<string, unknown>)?.source === "tts_job"
             const isPlaying = playingId === asset.id
             const isSelected = selectedIds.has(asset.id)
             // 播放中以 audio 真实时长为准，未播放显示卡片时长
@@ -184,6 +186,7 @@ export const MaterialVoiceTab: React.FC<MaterialVoiceTabProps> = ({
                 <div className="xx-voice-info vmat-info">
                   <div className="xx-voice-name" title={asset.name}>
                     {asset.name}
+                    {isAiMaterial && <span className="vmat-ai-badge">AI</span>}
                   </div>
                   <div className="xx-voice-subtitle">
                     {asset.file_size ? `${formatFileSize(asset.file_size)}` : "--"}

apps/web/src/pages/voices/components/VideoExtractModal.tsx

@@ -0,0 +1,207 @@
+import React, { useRef } from "react"
+import { Modal } from "antd"
+import { InboxOutlined, CloseOutlined } from "@ant-design/icons"
+
+interface VideoExtractModalProps {
+  open: boolean
+  file: File | null
+  progress: number | null
+  isExtracting: boolean
+  onClose: () => void
+  onFileSelect: (file: File | null) => void
+  onExtract: () => void
+}
+
+const ACCEPT_TYPES = ".mp4,.mov,.webm"
+
+const VideoExtractModal: React.FC<VideoExtractModalProps> = ({
+  open,
+  file,
+  progress,
+  isExtracting,
+  onClose,
+  onFileSelect,
+  onExtract,
+}) => {
+  const inputRef = useRef<HTMLInputElement>(null)
+
+  return (
+    <Modal
+      title={<span style={{ fontSize: 16, fontWeight: 600 }}>提取视频配音</span>}
+      open={open}
+      onCancel={() => {
+        if (isExtracting) return
+        onClose()
+      }}
+      footer={null}
+      width={480}
+      maskClosable={!isExtracting}
+    >
+      {!file ? (
+        <div
+          className="vmat-upload-dropzone"
+          onClick={() => inputRef.current?.click()}
+          style={{
+            border: "2px dashed #d9d9d9",
+            borderRadius: 8,
+            padding: "40px 20px",
+            textAlign: "center",
+            cursor: "pointer",
+            transition: "border-color 0.3s",
+          }}
+          onMouseEnter={(e) => (e.currentTarget.style.borderColor = "#7c3aed")}
+          onMouseLeave={(e) => (e.currentTarget.style.borderColor = "#d9d9d9")}
+        >
+          <InboxOutlined style={{ fontSize: 32, color: "#7c3aed", marginBottom: 12 }} />
+          <p style={{ margin: "0 0 8px", fontSize: 14, color: "#333" }}>点击选择视频文件</p>
+          <span style={{ fontSize: 12, color: "#999" }}>支持 MP4、MOV、WebM 格式</span>
+          <input
+            ref={inputRef}
+            type="file"
+            accept={ACCEPT_TYPES}
+            style={{ display: "none" }}
+            onChange={(e) => {
+              const f = e.target.files?.[0]
+              if (f) onFileSelect(f)
+            }}
+          />
+        </div>
+      ) : (
+        <div>
+          <div
+            style={{
+              display: "flex",
+              alignItems: "center",
+              justifyContent: "space-between",
+              padding: "12px 16px",
+              background: "#fafafa",
+              borderRadius: 8,
+              marginBottom: 16,
+            }}
+          >
+            <span
+              style={{
+                flex: 1,
+                overflow: "hidden",
+                textOverflow: "ellipsis",
+                whiteSpace: "nowrap",
+                fontSize: 14,
+                fontWeight: 500,
+              }}
+              title={file.name}
+            >
+              {file.name}
+            </span>
+            <span style={{ fontSize: 12, color: "#999", marginLeft: 8, flexShrink: 0 }}>
+              {(file.size / (1024 * 1024)).toFixed(1)} MB
+            </span>
+            {!isExtracting && (
+              <button
+                type="button"
+                onClick={() => {
+                  if (inputRef.current) inputRef.current.value = ""
+                  onFileSelect(null)
+                }}
+                style={{
+                  border: "none",
+                  background: "none",
+                  cursor: "pointer",
+                  color: "#999",
+                  marginLeft: 8,
+                  fontSize: 14,
+                }}
+                aria-label="移除文件"
+              >
+                <CloseOutlined />
+              </button>
+            )}
+          </div>
+
+          {progress !== null && (
+            <div style={{ marginBottom: 12 }}>
+              <div
+                style={{
+                  height: 6,
+                  background: "#f0f0f0",
+                  borderRadius: 3,
+                  overflow: "hidden",
+                }}
+              >
+                <div
+                  style={{
+                    height: "100%",
+                    width: `${progress}%`,
+                    background: "linear-gradient(90deg, #7c3aed, #a78bfa)",
+                    borderRadius: 3,
+                    transition: "width 0.3s",
+                  }}
+                />
+              </div>
+              <div
+                style={{
+                  textAlign: "right",
+                  fontSize: 12,
+                  color: "#999",
+                  marginTop: 4,
+                }}
+              >
+                {progress}%
+              </div>
+            </div>
+          )}
+
+          {isExtracting && (
+            <p style={{ textAlign: "center", fontSize: 13, color: "#7c3aed", margin: "12px 0 0" }}>
+              {progress === 100 ? "正在提取人声，请稍候..." : "正在上传视频..."}
+            </p>
+          )}
+        </div>
+      )}
+
+      <div
+        style={{
+          display: "flex",
+          justifyContent: "flex-end",
+          gap: 8,
+          marginTop: 24,
+        }}
+      >
+        <button
+          type="button"
+          onClick={onClose}
+          disabled={isExtracting}
+          style={{
+            padding: "6px 16px",
+            borderRadius: 6,
+            border: "1px solid #d9d9d9",
+            background: "#fff",
+            cursor: isExtracting ? "not-allowed" : "pointer",
+            fontSize: 14,
+            opacity: isExtracting ? 0.5 : 1,
+          }}
+        >
+          取消
+        </button>
+        <button
+          type="button"
+          onClick={onExtract}
+          disabled={!file || isExtracting}
+          style={{
+            padding: "6px 16px",
+            borderRadius: 6,
+            border: "none",
+            background: !file || isExtracting ? "#d9d9d9" : "#7c3aed",
+            color: "#fff",
+            cursor: !file || isExtracting ? "not-allowed" : "pointer",
+            fontSize: 14,
+            fontWeight: 500,
+          }}
+        >
+          {isExtracting ? "提取中..." : "开始提取"}
+        </button>
+      </div>
+    </Modal>
+  )
+}
+
+export default VideoExtractModal

apps/web/src/pages/voices/hooks/useVideoExtract.ts

@@ -0,0 +1,74 @@
+import { useState, useCallback } from "react"
+import { useQueryClient } from "@tanstack/react-query"
+import { extractVideoVoice } from "@/api/tts"
+
+/**
+ * 视频提取配音 Hook
+ * 封装视频上传弹窗状态、提取进度、提取 mutation 逻辑
+ */
+interface UseVideoExtractProps {
+  showToast: (message: string, type: "success" | "error") => void
+}
+
+export function useVideoExtract({ showToast }: UseVideoExtractProps) {
+  const queryClient = useQueryClient()
+
+  const [extractOpen, setExtractOpen] = useState(false)
+  const [extractFile, setExtractFile] = useState<File | null>(null)
+  const [extractProgress, setExtractProgress] = useState<number | null>(null)
+  const [isExtracting, setIsExtracting] = useState(false)
+
+  const handleExtractClose = useCallback(() => {
+    setExtractOpen(false)
+    setExtractFile(null)
+    setExtractProgress(null)
+    setIsExtracting(false)
+  }, [])
+
+  const handleExtract = useCallback(async () => {
+    if (!extractFile) return
+    setIsExtracting(true)
+    setExtractProgress(0)
+    try {
+      await extractVideoVoice(extractFile, (p) => setExtractProgress(p))
+      // 刷新素材列表
+      queryClient.invalidateQueries({ queryKey: ["assets", "voice"] })
+      queryClient.invalidateQueries({ queryKey: ["voice-materials"] })
+      showToast("视频配音提取成功", "success")
+      handleExtractClose()
+    } catch (err: unknown) {
+      const msg = err instanceof Error ? err.message : "提取失败，请重试"
+      showToast(msg, "error")
+    } finally {
+      setIsExtracting(false)
+      setExtractProgress(null)
+    }
+  }, [extractFile, queryClient, showToast, handleExtractClose])
+
+  const handleFileSelect = useCallback(
+    (file: File | null) => {
+      if (!file) {
+        setExtractFile(null)
+        return
+      }
+      const validTypes = ["video/mp4", "video/quicktime", "video/webm"]
+      if (!validTypes.includes(file.type)) {
+        showToast("仅支持 MP4、MOV、WebM 格式的视频文件", "error")
+        return
+      }
+      setExtractFile(file)
+    },
+    [showToast],
+  )
+
+  return {
+    extractOpen,
+    setExtractOpen,
+    extractFile,
+    extractProgress,
+    isExtracting,
+    handleFileSelect,
+    handleExtract,
+    handleExtractClose,
+  }
+}

apps/web/src/pages/voices/voices.css

@@ -193,6 +193,24 @@
   overflow: hidden;
   text-overflow: ellipsis;
   flex: 1;
+  display: flex;
+  align-items: center;
+}
+
+/* AI 配音标识 */
+.vmat-ai-badge {
+  display: inline-block;
+  margin-left: 6px;
+  padding: 1px 6px;
+  font-size: 11px;
+  font-weight: 600;
+  color: #7c3aed;
+  background: #f3f0ff;
+  border: 1px solid #ddd6fe;
+  border-radius: 4px;
+  line-height: 16px;
+  vertical-align: middle;
+  flex-shrink: 0;
 }
 
 .xx-voice-star {

apps/worker/worker_app/tasks/ingest.py

@@ -630,7 +630,7 @@ def ingest_asset(job_id: str) -> dict:
                 name=filename,
                 storage_key=job.storage_key,
                 mime_type=mime_type,
-                metadata={"ingest_error": error_reason},
+                metadata={"source": "upload", "ingest_error": error_reason},
                 file_size=int(metadata.get("size_bytes", 0)),
                 duration=float(metadata.get("duration", 0)),
                 width=int(metadata.get("width", 0)),
@@ -666,6 +666,7 @@ def ingest_asset(job_id: str) -> dict:
         if existing_asset is None:
             # 兜底：如果 API 端没有预先创建 Asset（旧版本兼容），则创建新记录
             logger.info("No pre-created asset found for storage_key=%s, creating new", job.storage_key)
+            metadata["source"] = "upload"
             asset = Asset.create(
                 project_id=job.project_id,
                 library_id=job.library_id,
@@ -687,6 +688,7 @@ def ingest_asset(job_id: str) -> dict:
             # 更新已有的 Asset 记录，补充元数据并将状态改为 READY
             asset = existing_asset
             asset.mime_type = mime_type
+            metadata["source"] = "upload"
             asset.metadata = metadata
             asset.file_size = int(metadata.get("size_bytes", 0))
             asset.duration = float(metadata.get("duration", 0))

tests/unit/test_plan_generator.py

@@ -1007,3 +1007,162 @@ class TestAssetDurationsAlwaysFetched:
             call_kwargs = mock_distribute.call_args
             asset_durations = call_kwargs.kwargs.get("asset_durations", call_kwargs[1].get("asset_durations"))
             assert asset_durations is None
+
+
+# ---------------------------------------------------------------------------
+# 测试：正式生成片段随机重排（Issue #1663）
+# ---------------------------------------------------------------------------
+
+
+class TestFormalGenerationShuffle:
+    """验证正式生成时片段顺序随机化。
+
+    Issue #1663: 正式生成时 smart_match 排序后对 asset_ids 做 random.shuffle，
+    使得同一批素材每次生成的视频片段顺序不同，有利于查重降重。
+    """
+
+    def _make_service_with_asset_repo(self):
+        """创建带 mock asset_repo 的 PlanGeneratorService（复用 TestAssetDurationsAlwaysFetched 模式）"""
+        from apps.api.app.services.plan_generator_service import PlanGeneratorService
+
+        plan_repo = StubEditPlanRepository()
+        clip_repo = StubEditPlanClipRepository()
+
+        asset_repo = MagicMock()
+
+        def fake_get(asset_id):
+            mock_asset = MagicMock()
+            mock_asset.duration = 30.0
+            mock_asset.quality_score = None
+            mock_asset.created_at = None
+            mock_asset.metadata = {}
+            return mock_asset
+
+        asset_repo.get = MagicMock(side_effect=fake_get)
+
+        with (
+            patch(
+                "apps.api.app.services.plan_generator_service.SQLAlchemyEditPlanRepository",
+                return_value=plan_repo,
+            ),
+            patch(
+                "apps.api.app.services.plan_generator_service.SQLAlchemyEditPlanClipRepository",
+                return_value=clip_repo,
+            ),
+        ):
+            db = MagicMock()
+            svc = PlanGeneratorService(db, asset_repo=asset_repo)
+            svc._plan_repo = plan_repo
+            svc._clip_repo = clip_repo
+
+        return svc, asset_repo
+
+    def test_formal_generation_shuffles_asset_ids(self):
+        """正式生成路径下 asset_ids 应被打乱，多次调用顺序应不同"""
+        svc, _ = self._make_service_with_asset_repo()
+
+        template = _make_template("one_take")
+        # 6 个 clip 容纳 6 个素材
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[
+                {"clip_type": ClipType.MAIN, "order": i, "min_duration": 3.0, "max_duration": 5.0} for i in range(6)
+            ],
+        )
+
+        asset_ids = ["a1", "a2", "a3", "a4", "a5", "a6"]
+
+        # 收集多次调用中 distribute_assets 收到的 asset_ids 顺序
+        captured_orders = []
+        with patch(
+            "apps.api.app.services.plan_generator_service.distribute_assets",
+            side_effect=lambda clips, asset_ids, *a, **kw: captured_orders.append(list(asset_ids)),
+        ):
+            # mock _sort_assets_by_smart_score 返回固定顺序，验证 shuffle 会打乱
+            with patch.object(
+                svc,
+                "_sort_assets_by_smart_score",
+                side_effect=lambda ids: list(ids),  # 原样返回
+            ):
+                with patch.object(
+                    svc,
+                    "_fetch_asset_scene_points",
+                    return_value={},
+                ):
+                    for _ in range(10):
+                        svc.generate_from_template(
+                            template=template,
+                            clip_configs=clip_configs,
+                            asset_ids=list(asset_ids),  # 每次传新列表
+                            random_preview=False,  # 正式生成
+                        )
+
+        assert len(captured_orders) == 10
+        # 每次 order 应该是 asset_ids 的一个排列
+        expected_set = set(asset_ids)
+        for order in captured_orders:
+            assert set(order) == expected_set
+
+        # 10 次调用中应至少出现 2 种不同顺序（概率 > 99.9%）
+        unique_orders = set(tuple(o) for o in captured_orders)
+        assert (
+            len(unique_orders) >= 2
+        ), f"Expected shuffled orders to vary, but got only {len(unique_orders)} unique order(s): {unique_orders}"
+
+    def test_formal_generation_does_not_mutate_original_list(self):
+        """shuffle 不应修改调用方的原始 asset_ids 列表"""
+        svc, _ = self._make_service_with_asset_repo()
+
+        template = _make_template("one_take")
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[
+                {"clip_type": ClipType.MAIN, "order": i, "min_duration": 3.0, "max_duration": 5.0} for i in range(4)
+            ],
+        )
+
+        original = ["a1", "a2", "a3", "a4"]
+        original_copy = list(original)
+
+        with patch("apps.api.app.services.plan_generator_service.distribute_assets"):
+            with patch.object(svc, "_sort_assets_by_smart_score", side_effect=lambda ids: list(ids)):
+                with patch.object(svc, "_fetch_asset_scene_points", return_value={}):
+                    svc.generate_from_template(
+                        template=template,
+                        clip_configs=clip_configs,
+                        asset_ids=original,
+                        random_preview=False,
+                    )
+
+        assert original == original_copy, "Original asset_ids list should not be mutated"
+
+    def test_preview_random_mode_unaffected_by_shuffle(self):
+        """预览随机模式不走 shuffle 路径，行为不变"""
+        svc, _ = self._make_service_with_asset_repo()
+
+        template = _make_template("one_take")
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[
+                {"clip_type": ClipType.MAIN, "order": i, "min_duration": 3.0, "max_duration": 5.0} for i in range(4)
+            ],
+        )
+
+        asset_ids = ["a1", "a2", "a3", "a4"]
+
+        captured_orders = []
+        with patch(
+            "apps.api.app.services.plan_generator_service.distribute_assets",
+            side_effect=lambda clips, asset_ids, *a, **kw: captured_orders.append(list(asset_ids)),
+        ):
+            for _ in range(5):
+                svc.generate_from_template(
+                    template=template,
+                    clip_configs=clip_configs,
+                    asset_ids=list(asset_ids),
+                    random_preview=True,  # 预览随机模式
+                )
+
+        assert len(captured_orders) == 5
+        # 预览模式下 random.shuffle 不应被调用（在 _distribute_assets 的 if not random_selection 块内）
+        # 所以 asset_ids 应该保持调用方传入的顺序（可能已由上层 shuffle 过）