An attempt to transcribe all GDC YouTube public videos
GDC-transcript is a pet project of PeDev and DK Liao in an attempt to transcribe publicly available videos in Game Developers Conference's YouTube channel.
The live tool is hosted at https://dklassic.github.io/GDC-transcript.
- How to Use
- Why
- How It Works
- Why Not Just YouTube Transcription?
- Project Structure
- Tools & Automation
- API
- Contributing
- Limitations
- Dependencies
- Future Work
The tool can be accessed at GDC-transcript. Enter a YouTube video ID to load the embedded video with synchronized subtitles from this repository.
You can also link directly to a specific video using a GET parameter:
https://dklassic.github.io/GDC-transcript/?v=<video_id>
For example: https://dklassic.github.io/GDC-transcript/?v=c2YRVWZupwo
For speakers who want to provide explicit permission or disallowance, or to add additional information about transcription of their talk, please submit an Issue or contact us via Google Form.
The goal is to help those who need assistance enjoying GDC content — not only non-English speakers, but also anyone dealing with poor audio quality, or anyone who wants to skim or search talk content without watching the full video.
For example, Integrating Narrative into Game Design: A Portal Post-Mortem is a great talk but the audio quality really suffers. A full transcript makes this content accessible to a much wider audience.
With full transcripts, text search becomes a powerful way to find relevant talks or quickly gauge whether a video is relevant. Since YouTube has removed community translation support, this project also aims to provide a foundation for community-driven subtitle translations.
Transcription is performed using OpenAI Whisper, a capable multilingual transcription model that handles accents and poor audio quality well. For more precise subtitle timing, we use the WhisperX fork by m-bain.
The pipeline:
- Audio is downloaded from YouTube using
yt-dlporpytube. - Large audio files (>24 MB) are resampled to a lower bitrate using
ffmpeg. - Audio is sent to the Whisper API or processed locally via WhisperX.
- The resulting
.srtsubtitle file is saved tostatic/src/subtitle/<video_id>.srt. - Plain-text transcripts are generated from the
.srtfiles and saved tostatic/src/transcript/. - During deployment,
build_metadata_json.pybuilds JSON data files consumed by the frontend.
The frontend is a static site with no build step — it embeds the YouTube player and overlays subtitles using youtube.external.subtitle.
- Better accuracy: Whisper outperformed YouTube auto-captions in our testing, especially with accents and poor audio.
- Better sentence structure: Whisper produces more readable, well-segmented transcripts.
- Plain-text access: Transcripts allow previewing or searching talk content without watching the full video.
- Translation support: Since YouTube removed community subtitles, this project provides an alternative path for community translations.
GDC-transcript/
├── index.html # Main web app (embedded player + subtitle viewer)
├── transcript.html # Transcript-only reader page
├── package.json # Node.js dependencies
├── API.md # JSON API documentation
│
├── static/
│ └── src/
│ ├── subtitle/ # .srt subtitle files (named by YouTube video ID)
│ ├── metadata/ # .meta JSON sidecar files (speaker permissions)
│ ├── transcript/ # Plain-text transcripts generated from SRTs
│ └── translation/ # Translated subtitles (subdirectories by language code)
│
├── tools/ # Python scripts and Jupyter notebooks
│ ├── auto_whisper.py # Main transcription pipeline
│ ├── build_metadata_json.py # Builds JSON API files for the frontend
│ ├── subtitle_to_transcript.py # Converts SRT to plain text
│ ├── download_audio.py # Batch audio downloader
│ ├── metadata_setup.py # Manages per-video .meta sidecar files
│ ├── update_srt_index.py # Re-numbers SRT index lines
│ ├── WhisperX_Youtube.ipynb # Jupyter notebook for local WhisperX transcription
│ ├── YouTube_Fetch.ipynb # Jupyter notebook for YouTube data fetching
│ └── gdc_video_list.txt # Master list of tracked GDC video IDs
│
└── .github/
├── CONTRIBUTING.md
├── workflows/ # GitHub Actions automation
└── ISSUE_TEMPLATE/ # Issue templates for requests and feedback
Current scale: ~1,759 subtitle files covering ~1,736 tracked GDC videos.
| Script | Purpose |
|---|---|
auto_whisper.py |
Downloads audio from YouTube and sends it to the Whisper API to produce an .srt file. Handles files >24 MB by reducing bitrate with ffmpeg. |
build_metadata_json.py |
Builds per-video JSON files consumed by the frontend. Reads .srt files, fetches YouTube metadata, reads .meta sidecars, and outputs json/ data files. |
subtitle_to_transcript.py |
Converts a single .srt file to a plain .txt transcript. Supports en, zh_tw, and ja. |
subtitle_to_transcript_all.py |
Batch version of the above for all subtitles. |
metadata_setup.py |
Creates or updates .meta JSON sidecar files, recording reviewer info and explicit speaker permissions. |
update_srt_index.py |
Re-numbers sequential indices in an SRT file after manual edits. |
download_audio.py |
Batch-downloads audio from a list of video IDs using concurrent threads. |
fetch_id_and_name_from_playlist.py |
Fetches video IDs and titles from a YouTube playlist. |
fetch_id_and_view_from_list.py |
Fetches view counts for a list of video IDs. |
| Workflow | Trigger | Purpose |
|---|---|---|
deploy.yml |
Push to gh-pages |
Builds JSON data files and deploys the site to GitHub Pages. |
auto_whisper.yml |
Manual | Transcribes a video by ID and opens a PR with the new .srt file. |
transcript.yml |
Manual | Converts a subtitle to plain-text transcript and opens a PR. |
reviewed_metadata.yml |
Manual | Marks a video as reviewed in its .meta file and opens a PR. |
allowance_metadata.yml |
Manual | Records explicit speaker permission for a video. |
disallowance_metadata.yml |
Manual | Records explicit speaker disallowance for a video. |
update_srt_index.yml |
Manual | Re-numbers SRT indices and opens a PR. |
all_transcript.yml |
Manual/Scheduled | Batch-generates transcripts for all subtitles. |
issue_check.yml |
Issue opened/edited | Warns if a transcript-request issue has unchecked checkboxes. |
Install Python dependencies:
pip install pytube yt-dlp openai requests
# ffmpeg must also be installed on your systemTranscribe a single video:
python tools/auto_whisper.py <video_id>Convert a subtitle to a plain-text transcript:
python tools/subtitle_to_transcript.py <video_id>The frontend consumes a set of JSON files generated at deploy time. See API.md for full documentation.
| Endpoint | Description |
|---|---|
GET /json/subtitles.json |
Array of all transcribed video IDs |
GET /json/reviewed.json |
Array of all reviewed video IDs |
GET /json/titles.json |
Dictionary mapping video IDs to titles |
GET /json/<video_id>.json |
Full metadata for a specific video |
Please create a Pull Request following the Contributing Guide. Feel free to open an Issue if a video you need has not been transcribed yet.
Key guidelines:
- Review subtitles, not transcripts. Transcripts are generated from subtitles, not the other way around.
- Before working on a new feature, open an issue first to discuss whether it's desirable.
- PR titles should follow the format:
review: <YouTube_Video_ID>orreview: <Talk Title> (<video_id>).
GDC-transcript uses Whisper's large model for best accuracy, which costs money to run. If you'd like to support the project financially, you can do so via Ko-Fi.
Thank you to everyone contributing to GDC-transcript!
- Hallucinations: Whisper occasionally makes things up. A filler phrase like "This initial prompt states that this is a game development transcript" may appear during music-only segments.
- Fast speakers: When a speaker talks very quickly, subtitle alignment can become unreliable.
- Forced English: Language detection is forced to English to ensure English segments are correctly transcribed. This can cause Whisper to mistranslate or misalign non-English segments. A future improvement would process these segments in their original language.
- subtitles-parser — SRT parsing (Neos3452's fork for dot-separator support)
- youtube.external.subtitle — External subtitle overlay for YouTube embeds
- OpenAI Whisper — AI transcription model
- WhisperX — Whisper fork with improved subtitle timing
- yt-dlp / pytube — YouTube audio download
- Allow video resize
- Show reviewer information
- Action to generate transcript for reviewed subtitles
- Action to generate summary for reviewed subtitles
- Action to generate translations for reviewed subtitles
- Action to generate summary for reviewed translations
