Person Identification and Gender Recognition from Footstep Sound using Modulation Analysis

Abstract

We describe a person identification system that is based on classifying the sound of the footstep. The classification is done on the spectrotemporal modulations of sound that are estimated using a model of auditory processing. We describe how different footsteps form a unique footprint in the spectrotemporal modulation domain and how this representation captures the user specific signatures. Using this representation, we achieved higher than 60% accuracy in identifying 9 people with three different shoes and two floors. The study demonstrates the efficacy of the spectrotemporal features in the tasks examined.