Showing papers by "Sebastian Möller published in 2005"

Towards generic quality prediction models for spoken dialogue systems - a case study.

Abstract: In this paper, models are investigated which aim at predicting quality perceived during the interaction with spoken dialogue systems, on the basis of instrumentally or expert-derived interaction parameters. More specifically, it will be evaluated how generic model predictions are when going from one system or user group to the next. In two experiments, user quality judgments have been collected according to a recently standardized framework, and a large number of interaction parameters have been extracted. The results show that both changes in the user group and in the system configuration significantly impact prediction performance. Potential reasons for the observed limitations are discussed.

Quality of Transmitted Speech for Humans and Machines

Abstract: This chapter provides an overview of quality aspects which are important for telecommunication speech services. Two situations are addressed, the telephone communication between humans and the task-oriented interaction of a human user with a speech-technology device over the phone, e. g., a spoken-dialogue system. A taxonomy is developed for each situation, identifying the relevant aspects of the quality of service. The taxonomies are used for classifying quality features, as perceived by the users of the service, as well as parameters and/or signals which can be measured instrumentally during the interaction. For conversations between humans, relationships can be established between the parameters/signals and the perceptive quality features, thus allowing for prediction of the quality for specific application scenarios. Finally, future efforts necessary for establishing similar prediction models for task-oriented human-machine interaction over the phone are identified.