An intelligent, platform-independent skill that extracts speech content from online videos and converts it into structured, timestamped markdown transcripts and summaries.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install video-transcript-method
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install video-transcript-method using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Video Transcript Extraction Method?
The Video Transcript Extraction Method is a universal, platform-independent tool designed to strip speech from any online video (such as Bilibili, YouTube, Douyin, or podcasts) and structure it into a readable document. By prioritizing high-accuracy closed captions (CC) and falling back on advanced Whisper voice recognition when captions are unavailable, it achieves highly accurate, semantic transcriptions.
This skill, a core capability within the Openclaw Skills ecosystem, goes beyond simple speech-to-text. It handles the full pipeline—from video URL resolution and audio extraction to intelligent semantic segment chunking, chapter titling, and summarization. This makes video content highly searchable, archivable, and easy to review without sitting through hours of playback.
Video Transcript Extraction Method Use Cases
- Studying Lectures & Webinars: Easily convert online academic or technical video lectures into structured study guides with chapter markers and exact timestamps.
- Content Archiving & Repurposing: Extract high-quality text drafts and notable quotes from podcasts or video interviews for blog posts and social media.
- Meeting Minutes Automation: Process recorded virtual meetings or webinars to generate concise summary tables and action items.
- Quick Video Skimming: Read a semantic summary and key takeaways of a long video to determine if it is worth watching in full.
How Video Transcript Extraction Method Works
- Video Source Positioning & Subtitle Detection: The skill parses the video URL, gathers metadata (title, author, duration), and checks for existing CC subtitles using yt-dlp. If high-quality CC subtitles exist, it directly extracts them.
- Audio Extraction & Whisper Transcribing: If no CC subtitles are found, the skill extracts the audio track as an MP3 and runs OpenAI's Whisper model (defaulting to the medium model) to transcribe the speech with high accuracy.
- Semantic Segmentation & Key Point Extraction: The raw transcript is denoised, logically segmented by topic (rather than fixed time blocks), annotated with timestamps, and summarized into key points and golden quotes in a clean markdown format.
Video Transcript Extraction Method Setup
Ensure you have yt-dlp and ffmpeg installed on your system. To set up this Openclaw Skills method, run the following steps:
1. Download Video Audio
Use yt-dlp to extract the best quality audio track as an MP3 file:
yt-dlp -f "bestaudio" --extract-audio --audio-format mp3 -o "./output/audio.%(ext)s" "<VIDEO_URL>"
2. Detect Closed Captions (CC)
Before transcribing, query if the video already has user-uploaded subtitles:
yt-dlp --list-subs "<VIDEO_URL>"
3. Run Whisper Speech Recognition
If CC subtitles are not available, use the transcription script to run local Whisper voice recognition:
python scripts/transcribe.py --model medium --audio ./output/audio.mp3
Video Transcript Extraction Method Data Schema & Taxonomy
The skill organizes all output data into structured directories containing the raw audio files, metadata records, and formatted transcripts.
File Outputs
| File Type | Path/Pattern | Description |
|---|---|---|
| Audio Track | /output/audio.mp3 |
Extracted audio from the video source |
| Subtitles (CC) | /output/subtitles.srt |
Original CC subtitles (if available) |
| Structured Transcript | /output/transcript.md |
Final semantic-segmented markdown transcript with summary |
Output Template Structure (TF-01)
The generated markdown file follows a strict, verification-ready layout:
## Transcript Extraction Record
### Step 1: Video Source Positioning
- **Video URL**: [URL]
- **Platform**: YouTube / Bilibili / Douyin / etc.
- **Metadata**: Title, Author, Views, Release Date, Duration
### Step 2: Subtitle Detection
- **CC Subtitles Available**: Yes / No
- **Extraction Route**: CC Direct Extraction / Whisper Transcription
### Step 3: Semantic Segmentation
| Chapter | Time Range | Theme | Core Content |
| :--- | :--- | :--- | :--- |
| Chapter 1 | [00:00-03:15] | Intro | Summary of introduction |
### Step 4: Key Points Summary
| Method/Point | Core Philosophy | Concrete Operation |
| :--- | :--- | :--- |
| Key Takeaway 1 | Description | Detailed steps |
**Golden Quotes**: "Notable quote from video"
Video Transcript Extraction Method Advanced Features
- Hybrid Subtitle Extraction: Automatically optimizes resource usage by prioritizing publisher-verified CC subtitles over compute-heavy Whisper transcriptions.
- Dynamic Semantic Chunking: Segments the video text based on semantic logic and topic shifts rather than naive, fixed-length time divisions.
- Multi-Size Whisper Fallback: Automatically falls back on different Whisper model sizes (from tiny to large) based on system hardware resources and length requirements.
- Verification Checklist Pipeline: Guarantees high-quality deliverables by enforcing a 5-step checklist checking completeness, timestamp accuracy, semantic logic, core summarization, and formatting compliance.
SKILL.md
Loading
Related Openclaw Skills
GuanShi Intelligence Expert
An automated AI intelligence agent designed to search, filter, extract, and cross-validate industry data to produce structured, high-confidence intelligence reports.
GuanShi Data Expert
An AI agent skill designed for automated data cleaning, statistical testing, financial modeling, and scenario simulation following the CRISP-DM methodology.
Douyin Data Query Method (douyin-data-method)
A standardized data query framework to retrieve and analyze Douyin video, user, search, trends, and live data using the MaxHub API.
Financial Audit Domain
A specialized financial risk analysis and audit reference library designed for A-share non-financial listed companies, integrating 50 indicators and 5 fraud red lines.
diagram-workflow-method
An expert diagramming and workflow visualization engine for generating pixel-perfect SVG and Mermaid system architectures, sequence diagrams, state machines, and ER models.
Connector Hub
A unified entry point organizing 33 platform, authentication, and API connectors with executable Python scripts to easily build auditable Openclaw Skills.
Featured*
