Indoor Sound Classification Using Lightweight CNN with Real-Time IoT Integration

This paper presents a lightweight Convolutional Neural Network (CNN) for indoor sound classification, specifically designed to identify ten distinct classes: knock on the door, mouse click, keyboard typing, wooden door squeaks, opening the can, washing machine, vacuum cleaner, alarm clock, clock ticking, and broken glass from [1]. The sound signals were captured using a microphone and transformed from the time domain to time-frequency images through Continuous Wavelet Transform (CWT). The proposed CNN, with a compact model size of 2.1 MB, was benchmarked against several well-known pre-trained models, including DenseNet, EfficientNet, InceptionNet, MobileNet, NASNet, ResNet, VGGNet, and Xception. Experimental results demonstrate that our lightweight CNN significantly outperformed these models, achieving a validation accuracy of 80% in a ten-fold cross-validation framework. The best performance among the pre-trained models was observed with ResNet101V2, which achieved only 62% validation accuracy in a ten-fold cross-validation framework with a considerably larger model size of 163 MB. Furthermore, the proposed model was successfully deployed in real-time on various IoT and processor modules, including the Raspberry Pi, NVIDIA A100, Tesla V100, and Intel Xeon Platinum, where it demonstrated superior efficiency, running at just 35 milliseconds per inference on the Raspberry Pi edge device. The lightweight CNN’s high accuracy and low latency make it an ideal candidate for applications in smart home systems, security monitoring, and real-time audio event detection on resource-constrained devices.