Knight Watch: Ubiquitous Computing Enhancements To Sleep Quality With Acoustic Analysis

TL;DR

Knight Watch tackles snoring through a pillow-adjacent wearable that performs on-device snoring detection via a CNN on the Arduino Nano Sense 33 BLE, paired with a Raspberry Pi for audio processing and a cloud analytics pipeline. The approach demonstrates a two-tier processing architecture, enabling on-device inference with haptic remediation and cloud-enabled sleep-pattern insights. Key contributions include an end-to-end embedded ML deployment, a BLE data stream, and cloud-based sleep analytics, illustrating the viability of ubiquitous computing in sleep health. The work also highlights hardware memory and data-collection constraints that shape practical deployment and guide future improvements.

Abstract

This project introduces a wearable, non-intrusive device for snoring detection and remediation, designed to be placed under or alongside a pillow. The device uses sensors and machine learning algorithms to detect snoring and employs gentle vibrations to prompt positional changes, thereby reducing snoring episodes. The device is capable of connecting via an API to a cloud-based platform for the analysis of snoring sleep patterns and environmental context. The paper details the development from concept to prototype, emphasizing the technical challenges, solutions, and alignment with ubiquitous computing in sleep quality improvement.

Figures (3)

• Figure 1: A picture of the faster model's (0.337 ms) accuracy using the test dataset.
• Figure 2: A picture of the embedded devices on a breadboard.
• Figure 3: A diagram of the IHI and Ubicomp cross team architecture.