Detecting snoring to inform night-time smartphone duty-cycle scheduling

Mobile sensing applications can infer user behavior at the cost of increased battery consumption due to the use of power-hungry sensors. To reduce power consumption, these applications can either cease sensing or perform duty-cycling at night when the user is sleeping; however, this approach works well only if user inactivity can be detected with certainty and with low power. In this paper we explore the viability of detecting human snoring as a strong indicator of user sleep periods and leverage that inference to perform aggressive night-time duty cycling. To detect human snoring, we collected ground-truth data from a sleeping participant and developed a threshold-based model that is able to identify periods of sleeping. This information was used to mark the onset of sleep, which in turn facilitated better duty-cycling and reduced power consumption.