Analysis of overlapping speech and emotions for interaction quality estimation

Abstract

Designing of indicators for problem detection during the dialogue is an important aspect for improving the systems user-adaptability in such rapidly expanding spheres as Spoken Dialogue Systems (SDSs) and Social Robotics (SR). Moreover, it is important not only for human-robot/computer spoken interaction (HCSI), but also for human-human conversation (HHC) as a way of service quality improvement in call centres. There are metrics, which may be used for both fields: HCSI and HHC. One of these metrics is customer/user satisfaction (CS), which modification is called Interaction Quality (IQ). Analysis of CS/IQ and human behaviour in some problematic situations (according to CS/IQ score) may be useful for further developing of an SDS that would be more human-like. Our research is focused on such essential parts of speech as emotions and overlapping speech. In this paper we analyse an impact of these speech features on automatic IQ estimation for HHC. Afterwards, we compare the obtained results with the result for HCSI.