Smart home automation with wakeup word detection using ESP32

Gan, Pey Yi (2025) Smart home automation with wakeup word detection using ESP32. Final Year Project, UTAR.

PDF Download (1692Kb)

Abstract

The limitations of traditional smart home automation systems that rely on costly, commercial voice assistants, which often lack customization and raise privacy concerns. The project aims to develop a low-cost, user-friendly smart home automation system that enhances privacy and flexibility to control the lights. The method involved in this project is related to the continuous audio sampling via a microphone connected to the ESP32 microcontroller, with on-device wake word detection using TensorFlow Lite to recognize the trigger word "Marvin" locally, which reduces the reliance on cloud services. After wake word detection, voice commands are transmitted to the Wit.ai cloud service for natural language understanding that enables the system to interpret user intents and control commands. ESP32 then executes the user’s commands by managing the connected LED through the GPIO pins. Experiments evaluation showed that the wake word model had a high success rate in intent identification using the NLP service, minimal latency, and over 90% accuracy in controlled environments. This technique effectively protects users’ privacy by reducing the reliance on cloud services and cutting the expenses by utilizing affordable hardware. In comparison to traditional voice assistance, the project shows significant enhancements in system responsiveness and user convenience. At the end of this research, a flexible yet affordable and privacyconscious voice assistance solution is designed and built. This system offers consumer more control over their smart environment. Future work may explore expanding device compatibility and refining natural language processing to enhance system robustness further

Actions (login required)