Enhanced Environmental Awareness and Security for Smart Devices using WiFi-Sensing

With the current advancements generative AI. Applications across many fields are being integrated with AI agents to provide a better experience to the users. One of these applications is personal assistants which can be integrated with AI to support Natural Language Understanding (NLU). In this work we introduce the architecture and evaluation of an AI-powered smart home device. The role of the role of the presenteddeviceevice is to be a personal assistant which is accessible across different platforms (web, desktop, mobile, and wearable devices), using multiple communication methods (text, voice, notification). The device will be able to collect and process multi-modal data from different platforms including sensory data which is analyzed to predict human activity, bridging the digital and the physical worlds.We assess the performance for three key tasks: Human identification and activity tracking using wifi sensing, storytelling using large language models, and text-to-speech synthesis. We evaluate the three tasks using a combination of objective performance metrics and user studies. Statistical analysis was conducted to evaluate different state-of-the-art Large Language Models (LLM) and Text-To-Speech (TTS) models. We performed deep learning across different data learning paradigms including local, central, and federated learning ensuring privacy and high accuracy.