diff --git a/CHANGELOG.md b/CHANGELOG.md
index f5a244c3a..c9dc1ae45 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,12 @@
 
 本项目所有重要变更记录于此。格式参考 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)，遵循 [语义化版本](https://semver.org/lang/zh-CN/)。
 
+## [Unreleased]
+
+### Fixed
+
+- 抖音转写改为从实际视频提取音轨，避免误用独立配乐资源导致漏掉口播；旧任务重试时备份并停用旧音频元数据及转写缓存。
+
 ## [2.4.4] - 2026-06-23
 
 ### Security
diff --git a/backend/app/downloaders/douyin_downloader.py b/backend/app/downloaders/douyin_downloader.py
index d0bc9d29b..8f29090dd 100644
--- a/backend/app/downloaders/douyin_downloader.py
+++ b/backend/app/downloaders/douyin_downloader.py
@@ -10,6 +10,7 @@
 from pydantic import BaseModel
 
 from app.downloaders.base import Downloader
+from app.downloaders.local_downloader import LocalDownloader
 from app.downloaders.douyin_helper.abogus import ABogus
 from app.enmus.note_enums import DownloadQuality
 from app.models.audio_model import AudioDownloadResult
@@ -224,19 +225,13 @@ def download(
                 output_dir = self.cache_data
             os.makedirs(output_dir, exist_ok=True)
 
-            output_path = os.path.join(output_dir, "%(id)s.%(ext)s")
-
             video_data = self.fetch_video_info(video_url)
-            output_path = output_path % {
-                "id": video_data['aweme_detail']['aweme_id'],
-                "ext": "mp3",
-            }
-            url = video_data['aweme_detail']['music']['play_url']['uri']
-            # 下载音频
-            audio_data = requests.get(url)
-            with open(output_path, 'wb') as f:
-                f.write(audio_data.content)
-            print(url)
+            # music.play_url 是配乐资源，不一定包含视频口播；必须提取实际视频音轨。
+            video_path = self.download_video(video_url, output_dir=output_dir)
+            output_path = os.path.join(
+                output_dir, f"{video_data['aweme_detail']['aweme_id']}_video_audio.mp3"
+            )
+            LocalDownloader().convert_to_mp3(video_path, output_path)
             tags = []
             for tag in video_data['aweme_detail']['video_tag']:
                 if tag['tag_name']:
@@ -252,8 +247,9 @@ def download(
                 video_id=video_data['aweme_detail']['aweme_id'],
                 raw_info={
                     'tags': video_data['aweme_detail']['caption'] + ''.join(tags),
+                    'audio_source': 'video_track_v1',
                 },
-                video_path=None  # ❗音频下载不包含视频路径
+                video_path=video_path
             )
         except Exception as e:
             raise e
diff --git a/backend/app/services/douyin_audio_cache.py b/backend/app/services/douyin_audio_cache.py
new file mode 100644
index 000000000..57b58eab9
--- /dev/null
+++ b/backend/app/services/douyin_audio_cache.py
@@ -0,0 +1,21 @@
+"""Retire transcripts derived from the old Douyin music download path."""
+import json
+from pathlib import Path
+from uuid import uuid4
+
+
+def retire_legacy_douyin_cache(audio_cache: Path, transcript_cache: Path) -> bool:
+    if not audio_cache.exists() and not transcript_cache.exists():
+        return False
+    try:
+        data = json.loads(audio_cache.read_text(encoding="utf-8"))
+        if data.get("raw_info", {}).get("audio_source") == "video_track_v1":
+            return False
+    except (OSError, ValueError, AttributeError):
+        pass
+    # Keep old results recoverable, but prevent retries from reusing music transcripts.
+    suffix = f".legacy-music-{uuid4().hex}.bak"
+    for path in (transcript_cache, audio_cache):
+        if path.exists():
+            path.rename(path.with_name(path.name + suffix))
+    return True
diff --git a/backend/app/services/note.py b/backend/app/services/note.py
index f493c9104..789216bde 100644
--- a/backend/app/services/note.py
+++ b/backend/app/services/note.py
@@ -1,4 +1,5 @@
 import json
+from app.services.douyin_audio_cache import retire_legacy_douyin_cache
 import logging
 import os
 from dataclasses import asdict
@@ -133,6 +134,10 @@ def generate(
             audio_cache_file = NOTE_OUTPUT_DIR / f"{task_id}_audio.json"
             transcript_cache_file = NOTE_OUTPUT_DIR / f"{task_id}_transcript.json"
             markdown_cache_file = NOTE_OUTPUT_DIR / f"{task_id}_markdown.md"
+            if platform == "douyin" and retire_legacy_douyin_cache(
+                audio_cache_file, transcript_cache_file
+            ):
+                logger.info("已备份旧抖音配乐转写缓存，将从视频音轨重新转写")
             # 1. 获取字幕/转写：优先缓存 → 平台字幕 → 音频转写
             transcript = None
 
diff --git a/backend/tests/test_douyin_audio_source.py b/backend/tests/test_douyin_audio_source.py
new file mode 100644
index 000000000..f970e3e47
--- /dev/null
+++ b/backend/tests/test_douyin_audio_source.py
@@ -0,0 +1,52 @@
+import json
+import tempfile
+import unittest
+from pathlib import Path
+from unittest.mock import patch
+
+from app.downloaders.douyin_downloader import DouyinDownloader
+from app.services.douyin_audio_cache import retire_legacy_douyin_cache
+
+
+class DouyinAudioSourceTests(unittest.TestCase):
+    def test_audio_comes_from_video_not_music(self):
+        downloader = DouyinDownloader()
+        detail = {
+            'aweme_id': '123', 'item_title': 'demo', 'caption': '', 'video_tag': [],
+            'music': {'play_url': {'uri': 'https://example.com/wrong-music'}},
+            'video': {'duration': 85000, 'cover': True,
+                      'cover_original_scale': {'url_list': ['cover']}},
+        }
+        with tempfile.TemporaryDirectory() as folder, \
+             patch.object(downloader, 'fetch_video_info', return_value={'aweme_detail': detail}), \
+             patch.object(downloader, 'download_video', return_value='/video.mp4'), \
+             patch('app.downloaders.douyin_downloader.LocalDownloader.convert_to_mp3') as convert, \
+             patch('app.downloaders.douyin_downloader.requests.get') as request:
+            result = downloader.download('https://www.douyin.com/video/123', output_dir=folder)
+            convert.assert_called_once_with('/video.mp4', str(Path(folder) / '123_video_audio.mp3'))
+            request.assert_not_called()
+            self.assertEqual(result.raw_info['audio_source'], 'video_track_v1')
+            self.assertEqual(result.video_path, '/video.mp4')
+
+    def test_old_cache_backed_up_and_new_cache_reused(self):
+        with tempfile.TemporaryDirectory() as folder:
+            audio, transcript = Path(folder) / 'audio.json', Path(folder) / 'transcript.json'
+            audio.write_text(json.dumps({'raw_info': {}}))
+            transcript.write_text('old transcript')
+            self.assertTrue(retire_legacy_douyin_cache(audio, transcript))
+            self.assertFalse(audio.exists())
+            self.assertFalse(transcript.exists())
+            backups = list(Path(folder).glob('*.bak'))
+            self.assertEqual(len(backups), 2)
+            self.assertIn('old transcript', [p.read_text() for p in backups])
+            audio.write_text(json.dumps({'raw_info': {'audio_source': 'video_track_v1'}}))
+            transcript.write_text('new transcript')
+            self.assertFalse(retire_legacy_douyin_cache(audio, transcript))
+            self.assertEqual(transcript.read_text(), 'new transcript')
+
+    def test_orphan_transcript_not_reused(self):
+        with tempfile.TemporaryDirectory() as folder:
+            audio, transcript = Path(folder) / 'audio.json', Path(folder) / 'transcript.json'
+            transcript.write_text('old transcript')
+            self.assertTrue(retire_legacy_douyin_cache(audio, transcript))
+            self.assertFalse(transcript.exists())