Embedded Intelligence for Smart Home Using TinyML Approach to Keyword Spotting

Current research in home automation focuses on integrating emerging technologies like Internet of Things (IoT) and machine learning to create smart home solutions that offer enhanced convenience, efficiency, and security. Benefits include remote control of household devices, optimized energy usage through automated systems, and improved user experience with real-time monitoring and alerts. In this study, a TinyML (Tiny Machine Learning)-based keyword spotting machine learning model and system is proposed which enables voice-based home automation. The proposed system allows users to control household devices through voice commands with minimal computational resources and real-time performance. The main objective of this research is to develop the TinyML model for resource-constrained devices. The system enables home systems to efficiently recognize specific keywords or phrases by integrating voice control for enhanced user convenience and accessibility. In this research, the different voice keywords of users of different age groups have been collected in the home environment and trained using machine learning algorithms. An IoT-based system is then developed utilizing the TinyML model to recognize a specific voice command and perform home automation tasks. The model has achieved 98% accuracy with an F1 score of 1.00 and 92% recall. The quantized model uses Latency of 5 ms, 7.9 K of RAM and 43.7 K of flash for keyword classification, which is the best fit for any resource-constrained devices. The proposed system demonstrates the viability of deploying a keyword spotting model for home automation on resource-constrained IoT devices. The research helps in building efficient and user-friendly smart home solutions, enhancing the accessibility and functionality of home automation systems.