Influence of different dialogue situations on user²s behavior in spoken corrections

This study analyzes the acoustic-prosodic features observed in spoken corrections in a task-oriented spoken dialogue. While previous studies have been found that spoken corrections are often predicted as hyperarticulate speech and related to some acousticprosodic events, their close analysis were not shown in terms of various dialogue situations in practical spoken dialogue systems and tasks. In this study, a set of spoken dialogue data was collected through both a fully operational spoken dialogue system and a human-human dialogue task with the same task setup. In addition, the influence of a concurrent task of speaking and car-driving on the prosodic features is investigated. The results of statistical analysis and a preliminary experiment of automatic classification showed that the characteristics of the prosodic features differ in the different dialogue situations while some finding are consistent with the observation in the previous studies in human-human and human-machine interactions.