SafeSpace is an infrastructure for mental rest in women’s shelters. By using privacy-first, edge-based audio intelligence, it recognizes patterns like vocal tension and physical energy shifts. This provides a responsive environment where needs are heard without the social risk of speaking out. It reduces the weight of hyper-vigilance, allowing residents to move out of survival mode and reconnect with their own identity and wellbeing.

1. Edge (Detection Device)
2. Python
3. openwakeword - wake word detection (hey jarvis)
4. sounddevice - microphone audio recording
5. tensorflow_hub + YAMNet - audio classification model (screaming, crying, glass, etc.)
6. numpy - audio waveform processing
7. librosa - audio utilities
8. requests - sending HTTP alerts to the staff server
9. threading - running Layer 1 and Layer 2 concurrently

1. Python + Flask - web server and REST API endpoints (/register, /alert, /resolve)
2. Flask-SocketIO - real-time WebSocket push to the browser dashboard

1. Clone the repository.
2. Run the staff dashboard: cd staff pip install -r requirements.txt python app.py
3. Run the edge device client: cd edge pip install -r requirements.txt python main.py

Note: The staff dashboard and the edge device client will run on different computers.

1. Runs Python natively
2. Built-in WiFi : connects directly to staff dashboard network
3. Always-on audio capture, ultra low power draw
4. Discreet form factor : mounts on any wall in any room

1. Laptop microphones differ in sensitivity, directionality, and noise handling. Shelter environments have overlapping voices, echoes, and background noise that are not replicated.
2. Training and testing data may not fully represent real distress signals across accents, languages, emotional intensity, or speaking styles.

For real-world shelter deployment, we plan to replace the laptop-based setup with a Raspberry Pi Zero 2 W + INMP441 MEMS Microphone (~$25–35 CAD per unit). This standalone sensor runs Python natively, connects over WiFi, and mounts discreetly on any wall. When a distress sound is detected, the alert flows directly from the sensor to the staff dashboard over the same network.