AI-based IoT device for smart nation deployment

Abstract

Hospitals are essential institutions that address the population’s healthcare needs. The workloads of nurses can be wide-ranging and stressful. Research has shown that a positive working environment helps nurses perform better at work, which translates to improved patients’ experiences. Hence, this project aims to create a device based on machine learning applications to improve nurses’ workplace efficiency. This project seeks to implement a real-time audio classification system on the new platform of Google Coral. It should be able to classify four categories of sounds, namely shouting, coughing, speech, falling people, or objects. When deployed at hospital wards, the system acts as an additional safeguard for patients’ wellbeing by monitoring the surrounding soundscape for any health or safety concerns. The implementation of this project involves integrating software and hardware. It requires audio data pre-processing, feature extraction, and convolutional neural network model training. The trained model is tested as a sound acquisition and inferencing system on PC. For deployment to hardware, the model is converted to a lightweight, compatible form and run on the Linux based Coral device.