feat(bilibili): 优先走官方 player API 直拉字幕

之前 BilibiliDownloader.download_subtitles 走的是 yt-dlp 的 writesubtitles 路径,对 B 站签名/Cookie 的兼容性差,常常空手而归,落到音频下载 + Whisper 转写的慢路径。

新增 bilibili_subtitle.BilibiliSubtitleFetcher:
- /x/web-interface/view?bvid=... → 拿 cid
- /x/player/wbi/v2?bvid=...&cid=... → 拿 subtitle 列表(subtitle_url 已带 auth_key)
- 优先级:人工中文 > AI 中文 > 任意中文 > 任意非空
- fetch JSON body 解析为 TranscriptResult
- 通过 CookieConfigManager 自动注入 SESSDATA cookie(AI 字幕必需)

bilibili_downloader.download_subtitles 顺序改为:先试新 fetcher,失败再回退到原 yt-dlp 路径。NoteGenerator 的字幕优先逻辑无需改动——它本来就调 download_subtitles。

效果:
- B 站视频如果有字幕(人工或 AI),直接秒拿,跳过音频下载 + 转写
- 完全绕开 MLX Whisper 不可用 / 模型未下载 等转写器问题
- 拿不到字幕时仍可走原音频转写路径

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
huangjianwu
2026-05-07 11:55:50 +08:00
co-authored by Claude Opus 4.7
parent 3bd8b670ca
commit 702b57c165
2 changed files with 174 additions and 0 deletions
@@ -8,6 +8,7 @@ from typing import Union, Optional, List
import yt_dlp
from app.downloaders.base import Downloader, DownloadQuality, QUALITY_MAP
from app.downloaders.bilibili_subtitle import BilibiliSubtitleFetcher
from app.models.notes_model import AudioDownloadResult
from app.models.transcriber_model import TranscriptResult, TranscriptSegment
from app.utils.path_helper import get_data_dir
@@ -155,6 +156,15 @@ class BilibiliDownloader(Downloader, ABC):
:param langs: 优先语言列表
:return: TranscriptResult 或 None
"""
# 1) 优先走 B 站官方 player API(直拉,无需下视频;AI 字幕需 SESSDATA cookie
try:
result = BilibiliSubtitleFetcher().fetch_subtitles(video_url)
if result and result.segments:
return result
except Exception as e:
logger.warning(f"player API 直拉字幕异常,回退到 yt-dlp: {e}")
# 2) Fallback:原 yt-dlp 路径(更脆弱,遇到签名/Cookie 问题失败概率较高)
if output_dir is None:
output_dir = get_data_dir()
if not output_dir: