Stream Performance Analysis

Overview

Analysis of IRL streams to identify patterns that differentiate high-performing streamers from low-performing ones.

Requirements

• Python 3.8+
• ffmpeg (for audio/video processing)

Installation

python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

Usage

1. Configure Streams

Edit src/download.py and replace the placeholder URLs with your 4 selected streams:

2. Run Pipeline

python run_pipeline.py

This will automatically:

1. Download stream segments (15 minutes each)
2. Extract features from audio and video
3. Generate statistical analysis and visualizations

3. Review Results

• results/features.csv - Extracted features
• results/comparison_boxplots.png - Visual comparison
• results/analysis_summary.txt - Statistical summary

Signals Extracted

1. Vocal Energy Variance - Measures animation and reactivity in speech
2. Dead Air Ratio - Percentage of time with no speech activity
3. Visual Motion Intensity - Physical engagement via frame differencing
4. Speaking Rate - Percentage of time actively talking
5. Hook Effectiveness - Energy in first 60 seconds vs overall average

Assumptions

1. Sample representativeness: 15-minute segment represents full stream quality
2. Audio presence: Assumes streamers use microphones (not silent streams)
3. Performance classification: Based on viewership metrics (concurrent viewers, total views)
4. Niche consistency: All streams are IRL format

Limitations

1. Small sample size - feel limited by choice of samples
2. No chat analysis - missing critical engagement metric for the real live stream
3. Manual classification - subjective performance labels
   • Documented selection criteria for transparency
4. Segment sampling - 15 minutes may not capture full stream dynamics

Troubleshooting

ffmpeg not found
Install via package manager: brew install ffmpeg (macOS) or apt-get install ffmpeg (Linux)

All features are NaN
Video files may be corrupted. Verify with: ffmpeg -i data/high_1.mp4

No significant differences found
Normal with small sample size.

Project Structure

stream-labs/
├── src/
│   ├── download.py    # Stream downloader
│   ├── extract.py     # Feature extraction
│   └── analyze.py     # Statistical analysis
├── data/              # Downloaded VODs (gitignored)
├── results/           # Output files
├── requirements.txt
├── README.md
└── insight_summary.md