EWNA 2026: Sensing Challenge

Details

Our goal for this product was to create a less intrusive secondary security system that has a secondary function of ML-powered assisted living. We had two goals in mind: one, to fill the gaps that current security systems currently have and provide an alternative for those who are more worried about personal security. Sound is one of the senses that provides crucial information about the environment, those without access to it are extremely disadvantaged and vulnerable to certain issues in the environment.This is where our product, “sound guardian” comes in. It is a local security device with a light array to detect and alert owners of audio anomalies, it acts as an assistive device as well as a secondary security system for existing security infrastructure. Standalone it can provide peace of mind as another way of recognizing danger.

The Sound Guardian is a ML-powered arduino sensor array that works with real time audio processing for audio anomalies to trigger sms notifications, light arrays and provides peace of mind through sms notifications and which acts as a way to bridge communication issues for those who are hard of hearing or elderly. It is compact and combines audio detection, machine learning, visual feedback and event messaging into one device

Files

sensing chall.png

Project Challenge

Portable Network Graphics (PNG) - 647.05 kB - 09/23/2026 at 18:12

Preview

wiring diagram2.png

Wiring Diagram

image/png - 56.34 kB - 09/23/2026 at 18:12

Preview

architecture diagram.png

ML Architecture Diagram

Portable Network Graphics (PNG) - 19.48 kB - 09/23/2026 at 18:13

Preview

Project Logs

Collapse

Build Log 3
Alex de la Cueva Tamanaha • 13 hours ago • 0 comments
Project Name: “Sound Guardian” Date/Time: 9/23/2026 / 10:00 AM

Current Phase: Model Training & Fine Tuning
- Trained ML model based on the ESC-50 dataset, targeting glass shattering, falls/thuds, and familiar/unfamiliar voices.
  - Gathered sample data for thud noises and falls, as well as voice recordings reading passages among our teammates for the familiar and unfamiliar voices.
- Added SMS Notification functionality
  - Successfully sends notifications to multiple numbers simultaneously
- Pivoted Back to RGB Matrix in order to have multiple matrices, one on-board, and one on light matrix, for outside and inside communication.
  - i.e. if someone falls, the light matrix is of no use to them, therefore sends a morse code sos signal to the other light to signal outsiders.
Build Log 2
JONENR • 14 hours ago • 0 comments

Project Name: “Sound Guardian” Date/Time: 9/22/2026 / 6:00 PM

Current Phase: Prototyping

Scrapped 8x8 LED Matrix for 8x13 LED Matrix present on Arduino Uno Q Board. Left us with no option but to draw certain images to differentiate between threats. Image display is completely implemented, with different functions to display different images, such as glass breaking or a fire alarm going off.

Fully implemented SMS messaging based off of amplitude of audio detection as proof of concept. Pivoted towards ML but still have not completely trained the model to use for threat detection. Found a dataset to train ML called ESC-50 to train model on, as well as our voices to get "familiar voices" and "unfamiliar voices" detected.

Arming/disarming placed on the backburner to focus on main problems with ML and audio detection.

The Mac processes audio from the USB microphone and sends the data over USB serial to the UNO Q. A Python relay running directly on the UNO Q's Linux side reads that serial data and forwards it over TCP into the App Lab container. This lets App Lab receive the microphone data for ML classification and then trigger alerts and LED responses.
Build Log 1
JONENR • 14 hours ago • 0 comments
Project Name: “Sound Guardian”

Date/Time: 9/22/2026 / 1:00 PM

Current Phase: Brainstorming

Started brainstorming ideas for what direction we wanted to steer our audio classification/threat detection project.
- Use given RGB LEDs on the 8x8 matrix to show how severe a potential threat may be
  - Fading Green/Blue for safe environment/idle
  - Pulsing Red for danger when detected
  - Use 8x8 matrix to draw out certain images when detecting specific threats
- Use SMS messaging to send a phone alert to whoever has connected to the device
  - Logs time and detected threat
  - Instant potential response
- Two options for audio detection
  - Use frequency filtering and harmonics to detect certain types of sound
  - Train ML model to detect sounds that signal potential threats
- Add (potentially remote) arming/disarming so that sound detection isn’t constantly running while audio such as the TV or guests are in the room the device is