Optimizing End-of-Turn Detection for Spoken Dialog Systems
Antoine Raux, Maxine Eskénazi · 2010
This paper presents an overview of our previously published work on the problem of end of turn detection in spoken dialog systems, which consists in determining whether the user has completed their turn as they are speaking it. Over the past few years, we designed two new models that exploit contextual features to significantly reduce system latency at the end of user turns without increasing turn-taking conflicts. We briefly present some experimental results on a publicly deployed spoken dialog system that confirm the effectiveness of our approaches to provide smoother and more efficient human-computer dialogs.