WalkID: Towards Context Awareness of Smart Home by Identifying Walking Sounds

Abstract

Smart Home provides great conveniences to people by connecting various IoT devices (e.g., smart TVs, thermostats and health care devices) in the network. The voice assistant is utilized as the pivotal hub, which takes the user’s commands and controls the IoT devices accordingly. However, the voice assistant can only passively receive the user’s clear voice commands and is lack of the capability to automatically fulfill the users individual needs such as the personal environment setting (e.g., light, temperature and volume) and health monitoring. In this work, we equip the voice assistant with the context-awareness ability to provide personalized services automatically by recognizing the user’s identity. In particular, we develop a system to identify the user based on the walking sounds. The proposed system first recognizes the walking sounds based on the cluster-based walking detection to detect the presence of a user and then identifies the user based on machine-learning algorithms. Moreover, the far-field microphone arrays available on voice assistants are leveraged to further improve the user identification performance to achieve over 90% accuracy.