| Requirement | Why | Notes |
|---|---|---|
| Python 3.9+ | Runs the pipeline | |
| ffmpeg | Whisper decodes audio through it | Must be on PATH |
| Ollama | Generates the summaries | Must be running locally |
| Git | Cloning the archive |
git clone https://github.com/willtheorangeguy/Shell-Game-Transcripts.git
cd Shell-Game-Transcripts
pip install -r requirements.txt
pip install git+https://github.com/openai/whisper.git
Whisper runs on the CPU by default, and on a full episode that is slow enough to matter. For an NVIDIA GPU, install the CUDA build of PyTorch instead:
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118
This is the single change with the largest effect on how long a run takes.
ollama pull llama3.1:8b
Ollama must be running when the summarisation stage executes — it is contacted over HTTP on localhost, not embedded.
requirements.txt pulls in| Package | Used for |
|---|---|
yt-dlp |
Downloading episode audio |
requests |
HTTP fetches for feeds and published transcripts |
openai-whisper |
Transcription |
transformers |
Tokenizer used to size summarisation chunks |
ollama |
Client for the local summarisation model |
google-api-python-client |
YouTube metadata lookups |
language-tool-python |
Grammar and spelling cleanup |
torch |
Whisper’s runtime — replace with the CUDA build for GPU |
mutagen |
Writing ID3 tags onto downloaded audio |
Quickstart to run one year, or Configuration to change the models first.