A Streamlit web app that takes a YouTube video URL, transcribes its audio using OpenAI Whisper, and generates a concise summary using a HuggingFace Transformers summarization model. Runs entirely on CPU -- no GPU or API keys required.
- Download -- the audio track of a YouTube video is downloaded via
pytubefix. - Transcribe -- OpenAI Whisper (through Haystack's
WhisperTranscriber) converts speech to text. - Summarize -- a HuggingFace summarization pipeline (
sshleifer/distilbart-cnn-12-6) condenses the transcript into a short summary. - Display -- the video and its summary are shown side-by-side in a Streamlit UI.
| Component | Library |
|---|---|
| Web UI | Streamlit |
| NLP Pipeline | Haystack (v1.x) |
| Speech-to-Text | OpenAI Whisper |
| Summarization | HuggingFace Transformers |
| Video Download | pytubefix |
# Clone the repo
git clone https://github.com/Ishan007-bot/Yt-Video-Summarizer.git
cd Yt-Video-Summarizer
# Create a virtual environment and install dependencies
python -m venv venv
venv\Scripts\activate # Windows
# source venv/bin/activate # macOS / Linux
pip install -r requirements.txt
# Run the app
streamlit run yt_summary.pyCheck out the app by opening https://ishanbot-yt-video-summarizer.hf.space, paste a YouTube URL, and click Submit.
| Variable | Default | Description |
|---|---|---|
WHISPER_MODEL |
base |
Whisper model size (tiny, base, small, medium) |
SUMMARY_MODEL |
sshleifer/distilbart-cnn-12-6 |
HuggingFace summarization model |
YT_DOWNLOAD_DIR |
/tmp/yt_downloads |
Directory for downloaded audio files |
.
├── yt_summary.py # Main Streamlit app
├── model_add.py # Custom Haystack invocation layer (HF summarization)
├── summary.py # Standalone CLI summarization script
├── requirements.txt # Python dependencies
├── .streamlit/ # Streamlit config (headless, no email prompt)
├── LICENSE
└── README.md
Distributed under the MIT License. See LICENSE for more information.
If you like this project, drop a star to the repo!
