A demo-ready project that generates highlights and a semantic summary from any video (sports, lectures, generic).

• Upload a video (MP4/MOV) or provide a YouTube link
• Picks key segments using a multi-signal scoring pipeline:
  • Scene detection (PySceneDetect)
  • Audio energy peaks (librosa)
  • Semantic importance from transcript embeddings (Sentence‑BERT)
• Exports:
  • highlights.mp4
  • timestamps.json (why each moment was chosen)
  • a readable summary (bullets + paragraph)

• Python 3.10+ recommended
• ffmpeg installed and available in PATH

Check:

ffmpeg -version

python -m venv .venv
# mac/linux
source .venv/bin/activate
# windows (powershell)
# .venv\Scripts\Activate.ps1

pip install --upgrade pip
pip install -r requirements.txt

First run will download pretrained models (Whisper + Sentence‑BERT). That’s expected.

streamlit run app.py

python -m pipeline.cli   --video "path/to/video.mp4"   --mode lecture   --target_seconds 75

Outputs go to: outputs/<run_id>/

• If export fails: ensure ffmpeg is installed.
• If transcription is slow: choose tiny model in the UI.
• If video has no speech: the semantic module falls back and uses audio/visual cues.