This paper contains an analysis of caller response timing patterns in spoken dialog systems. The findings presented here are based on data from live commercial dialog systems. It is shown that caller responses after a system finished playing the prompt resemble a uni-modal distribution and can be modeled with a rational distribution function. This finding allows understanding when callers tend to respond to a system depending on the dialog state type. Based on these findings a no-speech timeout optimization method is being proposed.
Index Terms: spoken dialog systems, turn-taking behavior, caller response timing, timeout optimization
Bibliographic reference. Witt, Silke M. (2012): "Caller response timing patterns in spoken dialog systems", In INTERSPEECH-2012, 326-329.