Webex Meeting Speaker Labeling Tool (MVP)

An offline Python tool that post-processes Webex meeting recordings to identify and label speakers using audio diarization and video analysis.

Features

• Audio Diarization: Identifies distinct speakers from audio tracks
• Video Analysis: Detects faces and tracks lip movements to associate speakers with visual identities
• Speaker Naming: Automatically extracts participant names from meeting introductions
• Multi-format Output: Generates labeled SRT subtitles and structured JSON metadata
• Annotated Video Generation: Creates video with face detection boxes and speaker-labeled subtitles
• Web UI Viewer: Interactive interface to view videos with synchronized transcripts
• Auto-Transcription: Built-in Whisper ASR support when transcript is not available
• CPU-Optimized: Runs locally without GPU requirements
• Privacy-First: All processing happens on your machine

Use Cases

• Post-process Webex meetings recorded in conference rooms (single microphone, multiple participants)
• Generate accurate speaker-labeled transcripts
• Create searchable meeting records with speaker attribution

System Requirements

• Python 3.8+
• CPU-based processing (no GPU required)
• ~4GB RAM for typical meetings
• Disk space for temporary processing files

Installation

# Clone the repository
git clone <repository-url>
cd webex-speaker-labeling

# Create virtual environment
python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

Quick Start

Basic processing:

python process_meeting.py \
  --video path/to/meeting.mp4 \
  --transcript path/to/transcript.vtt \
  --output-dir ./output

Auto-transcription (no transcript file):

python process_meeting.py \
  --video path/to/meeting.mp4 \
  --output-dir ./output \
  --asr-model base

Generate annotated video with face detection:

python process_meeting.py \
  --video path/to/meeting.mp4 \
  --output-dir ./output \
  --annotated-video

Launch web UI viewer:

python process_meeting.py \
  --video path/to/meeting.mp4 \
  --output-dir ./output \
  --web-ui

Or view already-processed videos:

python view_videos.py --output-dir ./output --port 5000

Then open http://localhost:5000 in your browser.

Output Files

• meeting_labeled.srt: Subtitle file with speaker names
• meeting_labeled.json: Structured metadata with speaker segments, timestamps, and confidence scores
• meeting_annotated.mp4: (Optional) Video with face detection boxes and speaker-labeled subtitles

Project Structure

webex-speaker-labeling/
├── src/
│   ├── audio/           # Audio processing and diarization
│   ├── video/           # Face detection and tracking
│   ├── fusion/          # Audio-visual alignment
│   ├── naming/          # Speaker name extraction
│   ├── output/          # Output generation
│   ├── visualizer.py    # Video annotation with face boxes
│   └── web_ui.py        # Flask web viewer
├── process_meeting.py   # Main CLI entry point
├── view_videos.py       # Standalone web UI viewer
├── config.yaml          # Configuration parameters
└── requirements.txt     # Python dependencies

Configuration

Edit config.yaml to customize:

• Video frame sampling rate
• Diarization parameters
• Face detection thresholds
• Name extraction patterns

Limitations (MVP)

• Up to 10 participants
• Meeting duration up to 2 hours
• Processing time: ~1-2x meeting duration
• Best results when all participants are visible on camera
• Single-room, single-microphone setups

Future Enhancements

• VS Code extension integration
• Electron desktop app wrapper
• Real-time processing capabilities
• GPU acceleration support

Privacy & Security

• 100% Local Processing: No data leaves your machine by default
• Optional Cloud Services: Can be configured but disabled by default
• No Data Collection: No telemetry or usage tracking

License

[Your License Here]

Contributing

Contributions welcome! Please read CONTRIBUTING.md for guidelines.

Support

For issues or questions, please open a GitHub issue.