| name | youtube-transcript |
| description | Tải transcript (phụ đề/captions) từ YouTube video qua yt-dlp, convert VTT sang plain text sạch (deduplicated). Hỗ trợ manual subtitles + auto-generated, 60+ ngôn ngữ, chọn ngôn ngữ cụ thể.
Dùng khi user chia sẻ YouTube URL và muốn lấy nội dung text, hoặc nói "transcript", "phụ đề", "subtitles", "captions", "lấy text từ video", "nội dung video này nói gì", "tải phụ đề", "youtube transcript", "chuyển video thành text". Cũng dùng khi user paste link YouTube và muốn đọc/phân tích nội dung video mà không cần xem.
|
YouTube Transcript Downloader
Download transcript from YouTube video, convert to clean plain text.
Workflow
Follow this exact order. Stop and inform the user if any step fails.
Step 1: Verify yt-dlp
command -v yt-dlp
If not found, install:
- macOS:
brew install yt-dlp
- Linux:
sudo apt install -y yt-dlp
- Fallback:
pip3 install yt-dlp
If installation fails, tell the user to install manually.
Step 2: Get video info and check subtitles
VIDEO_URL="THE_URL"
yt-dlp --print "%(title)s" "$VIDEO_URL"
yt-dlp --list-subs "$VIDEO_URL"
Note what's available: manual subtitles (higher quality) vs auto-generated.
Step 3: Download subtitles
Try in order until one succeeds:
Manual subtitles (best quality):
yt-dlp --write-sub --skip-download --output "transcript_temp" "$VIDEO_URL"
Auto-generated subtitles (fallback):
yt-dlp --write-auto-sub --skip-download --output "transcript_temp" "$VIDEO_URL"
Both produce a .vtt file.
If neither works, inform the user that video này không có phụ đề.
Step 4: Convert VTT to plain text
Use the bundled conversion script to deduplicate and clean the VTT output:
VIDEO_TITLE=$(yt-dlp --print "%(title)s" "$VIDEO_URL" | tr '/' '_' | tr ':' '-' | tr '?' '' | tr '"' '')
VTT_FILE=$(ls transcript_temp*.vtt 2>/dev/null | head -n 1)
python3 skills/youtube-transcript/scripts/vtt-to-txt.py "$VTT_FILE" "${VIDEO_TITLE}.txt"
Then clean up the temporary VTT file:
rm "$VTT_FILE"
Step 5: Confirm to user
Tell the user the file name and location. Offer to read/display the content if they want to review it.
Why deduplication matters
YouTube auto-generated VTT files show captions progressively with overlapping timestamps, producing many duplicate lines. The conversion script uses a seen-set to preserve speaking order while removing duplicates.
Language selection
By default yt-dlp downloads all available subtitle languages. To target a specific language:
yt-dlp --write-auto-sub --sub-langs vi --skip-download --output "transcript_temp" "$VIDEO_URL"
Common codes: en (English), vi (Vietnamese), ja (Japanese), ko (Korean).
Error handling
Read references/error-handling.md for solutions to common issues (private videos, geo-blocking, SSL errors, missing subtitles).
