Typed VoIP Silence Prediction for Smartphone Energy Savings

Abstract

Previous research has shown that considerable energy savings may be made in Android phones by intelligently switching the WiFi radio from Constantly Awake Mode (CAM), to Power Save Mode (PSM) during periods of mutual silence in Voice over IP calls. This is because a WiFi radio in CAM consumes approximately one third of a smartphones battery life. Since a typical conversation can consist of up to 60 % silence, much of a conversation need not be transmitted, and thus wastes a considerable amount of battery power. This research shows that as much as 40 % of that energy can be saved by switching the radio to PSM during mutual silence periods. However, this technique relies heavily on accurately predicting the duration of such silence periods. This paper presents and compares a silence model which uses speaker identity to build a typology of silences to a naive un-typed strategy. This type model relies on recent research showing statistical difference between the durations of classes of silences based on speaker identity before and after a silence. Evaluation of this technique, however, demonstrates that such models do not show improvement over more naive strategies. Thus, a naive strategy is perfectly sufficient for accurate prediction of silence duration.