Acoustic Detection of Consecutive Stages of Spoken Interaction Based on Speaker-Group Specific Features

Authors: Ronald Böck, Olga Egorow, Andreas Wendemuth

Abstract:

Spoken human interaction comprises a variety of tasks and behaviours. Since the state of a human is affected by various circumstances, it can change during an interaction – for example, depending on the task at hand. Such interaction stages occur in human-human interaction as well as in human-computer interaction, and therefore are object of automatic detection and classification. In this paper, we investigate several feature sets with respect to their suitability to the task of interaction stages classification.