Embedded Compute Sound Sampling and Analysis System

Khaw, Si Kai (2026) Embedded Compute Sound Sampling and Analysis System. Final Year Project (Diploma), Tunku Abdul Rahman University of Management and Technology.

Text KHAW SI KAI_Full Text.pdf Restricted to Registered users only Download (1MB)

Abstract

This project presents the design and implementation of a real-time environmental sound analysis system on a low-cost embedded platform. Using a Raspberry Pi 4B paired with a USB microphone, the prototype continuously monitors acoustic environments, detects salient sound events via an RMS–EMA trigger, and classifies them with the pre-trained YAMNet deep learning model. Results are automatically logged into Excel with timestamps, confidence scores, top-3 predictions, and relative noise levels in dBFS, while waveform plots andWAV recordings provide additional traceability. Testing under multiple parameter configurations showed that an RMS threshold of 0.002 with an EMA decay of 0.995 delivered the best trade-off between sensitivity and stability. The system successfully recognized common environmental sounds such as speech, music, and vehicle noise, although misclassification was observed in cases of fireworks, impulsive sounds, and noisy backgrounds. While the system reports relative dBFS rather than calibrated dB SPL, it demonstrates the feasibility of embedded real-time noise monitoring. The work contributes to urban noise management, public health awareness, and smart city applications, and forms a foundation for future improvements such as calibration, dataset expansion, IoT integration, and the use of a built-in rechargeable battery for long-term autonomous deployment