Video Transcriber

Turn a YouTube video into a clean text transcript and a ready-to-paste prompt for any AI assistant (Gemini, ChatGPT, NotebookLM, Claude, etc.) — so you can go from a recording to an outline, summary, and key takeaways in minutes. Curated prompt templates for sermons, security talks, IT/SRE talks, and other structured spoken content.

---

Privacy & cost

Everything runs on your Mac after install. yt-dlp is the only network call (downloading from YouTube); whisper.cpp does the transcription on-device, and the prompt is copied to your local clipboard. No API keys, no telemetry, no per-minute fees. The AI step (paste into Gemini, ChatGPT, NotebookLM, or Claude) uses whichever assistant you already have.

---

⚡ Quickstart (Mac, Homebrew already installed)

brew install whisper-cpp ffmpeg yt-dlp

git clone https://github.com/TheDeLay/video-transcriber.git
cd video-transcriber
chmod +x bash/transcribe.sh

./bash/transcribe.sh "https://www.youtube.com/watch?v=YOUR_VIDEO_ID"

When it finishes, your Desktop has <title>_<date>.txt (transcript) and <title>_<date>_prompt.md (AI prompt — also auto-copied to your clipboard).

Don't have Homebrew? Install it from brew.sh, then run the steps above. Want every step explained? See bash/install.md — full walkthrough, troubleshooting, customization.

--latest for regular content

For series content (weekly podcasts, Sunday services, daily uploads, etc.), skip URL lookup:

# Most recent upload from a channel
./bash/transcribe.sh --latest "https://www.youtube.com/@SomeChannel/streams"

# Most recent upload whose title contains a keyword (case-insensitive, scans 30 most recent)
./bash/transcribe.sh --latest --filter "KEYWORD" "https://www.youtube.com/@SomeChannel/streams"

---

Two installation paths

Path	For	What you install
bash/	Anyone on a Mac. Easiest.	Homebrew + 3 brew formulas. No Python.
python/	Developers who want to extend it.	Python 3.11+, pip, a virtualenv.

Both paths produce identical outputs.

Platform support: macOS is the primary tested platform. Linux is expected to work for both paths but isn't validated in this release. Windows isn't supported yet — the bash path won't work natively, and the Python path likely works under WSL but is unverified. See CHANGELOG.md for what's tested per release.

How it works

YouTube URL  →  yt-dlp  →  audio.wav  →  Whisper  →  transcript.txt
                                                      ↓
                                             prompt-template.md
                                                      ↓
                                             prompt.md (clipboard)
                                                      ↓
                                             Paste into your AI
                                                      ↓
                                       Outline + Summary + Takeaways

yt-dlp is the only network step; everything after the audio download runs locally.

Customizing the AI prompt

prompt-template.md (at the project root) is the default prompt. It produces an outline, an insightful summary, and 5–7 takeaways for any structured talk — sermons, lectures, conference talks, podcasts, interviews. It also runs a transcription-hygiene pass first so the AI flags speech-to-text errors instead of building conclusions on top of them.

You can swap in a different template three ways:

# Bash path — flag (highest precedence)
./bash/transcribe.sh --template templates/infosec-leader.md "URL"

# Bash path — env var
PROMPT_TEMPLATE=templates/infosec-leader.md ./bash/transcribe.sh "URL"

# Python path
video-transcribe --template templates/infosec-leader.md "URL"

Or set it as your shell default:

# Add to ~/.zshrc or ~/.bashrc
export PROMPT_TEMPLATE="$HOME/path/to/video-transcriber/templates/infosec-leader.md"

Curated alternatives

Template	For
templates/church-leader.md	Sermons / spiritual talks where the user is a pastor or ministry leader.
templates/infosec-leader.md	Security conference talks, vendor briefings, threat intel, post-mortems.
templates/it-leader.md	SRE / DevOps / platform talks, vendor roadmaps, ITSM training, architecture reviews.

See templates/README.md for what each one adds and the convention if you want to write your own and contribute it back.

License

MIT — see LICENSE.