Automatically extract slides and transcribe audio from lecture videos.

Use the Google Colab notebook for the easiest experience - no installation required:

The Colab notebook provides a user-friendly interface with:

• Form-based configuration (no command-line needed)
• Automatic GPU acceleration for Whisper
• One-click download of results
• Google Drive integration for easy file access

• Smart Slide Detection: Grid-based peripheral uniformity analysis to identify actual slides vs UI elements
• Letterbox Detection: Automatically detects and crops letterbox bars from recordings
• Sensitive Change Detection: Catches animated slides and text appearing
• Fast Transcription: Uses Deepgram API by default (or Whisper as fallback)
• Batch Processing: Process entire folders of lecture videos
• PDF Output: All slides compiled into a single PDF for easy LLM intake

• Python 3.9+
• FFmpeg (for audio extraction)
• Deepgram API key (free tier available at deepgram.com)

git clone https://github.com/graysonmc/Lecture-Processor.git
cd Lecture-Processor

# Create virtual environment (using Anaconda)
conda create -n lecture-processor python=3.11
conda activate lecture-processor

# Or using venv
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Create a .env file in the project directory with your Deepgram API key:

DEEPGRAM_API_KEY=your-api-key-here

Get a free API key at https://deepgram.com ($200 free credits)

For users who prefer running locally instead of Colab:

Put your lecture videos in ~/Desktop/Raw Lectures/, then run:

python process_lecture.py --batch

Output goes to ~/Desktop/Processed Lectures/<video_name>/:

• slides.pdf - All slides in one PDF
• transcript.txt - Full transcription

python process_lecture.py lecture.mp4

# Slides only (skip transcription)
python process_lecture.py --batch --slides-only

# Custom output directory
python process_lecture.py --batch --output-dir /path/to/output

# Adjust slide detection sensitivity (lower = more sensitive)
python process_lecture.py --batch --threshold 20

# Use local Whisper instead of Deepgram
python process_lecture.py --batch --whisper-model base

1. Letterbox Detection: Requires black bars on sides (positive signal for recorded slides)
2. Rectangle Mapping: Detects slide boundaries within the letterbox
3. Grid Analysis: Maps 18x12 grid onto detected slide region
4. Peripheral Uniformity: Checks that edges/corners have consistent white background
5. Change Detection: Sensitive to new text appearing (catches animated slides)

~/Desktop/Processed Lectures/
├── Lecture-01/
│   ├── slides.pdf
│   └── transcript.txt
├── Lecture-02/
│   ├── slides.pdf
│   └── transcript.txt
└── ...

pip install pytest
pytest tests/ -v

• No slides detected: Video may not have letterbox bars, or slides may have dark backgrounds
• Too many duplicates: Lower the threshold with --threshold 15
• Missing slides: Raise the threshold with --threshold 30
• Transcription fails: Check your DEEPGRAM_API_KEY is set correctly

MIT