Implementation

Abstract

The ideas about adaptive dialogue management and speech-based emotion recognition as described in Chapters 3 and 4 constitute a firm groundwork as for theoretical aspects of the integration of emotions into adaptive SLDSs. This groundwork, however, features a large potential for improvement as well as a high degree of flexibility concerning an implementation. In this chapter, we identify approaches to improve the performance of our emotion and speech–emotion recognizers and we describe the implementation of our adaptive dialogue manager.

There exist a large variety of parameters which can be altered to increase the performance and robustness of speech-based emotion recognizers. In the following section, we address the optimization of our plain emotion recognizer and our combined speech-emotion recognizer. Optimizations of the recognition performance do not necessarily require a change of the recognizers. Instead, a post processing algorithm can be applied to reduce the recognition errors after the recognition process. Our approach to combining multiple plain emotion recognizers or combined speech-emotion recognizers to reduce the overall error rate is described in Section 5.2. In Section 5.3, we present an adaptive dialogue manager which is based on VoiceXML. This dialogue manager integrates the semi-stochastic emotional dialogue model as described in Section 3.8 to adapt the dialogue flow to the user’s emotional state.