Zweistufige kontextsensitive Sprecherklassifikation am Beispiel von Alter und Geschlecht

This dissertation describes an approach on how to exploit the so called paralinguistic information contained in the speech for estimating the speakers’ age and gender. The technical constraints are given by mobile speech-based dialog systems, which shall acquire user models in a non-intrusive fashion. According to the main speaker characteristics age and gender, the approach was developed under name AGENDER as a part of the project m3i (Mobile Multi-Modal Interaction). It belongs to the project COLLATE (Computational Linguistics and Language Technology for Real Life Applications), which is sponsored by the German Federal Ministry for Research and Education (BmBf) and is being carried out by the German Research Center for Artificial Intelligence (DFKI) and the University of the Saarland. The target applications of AGENDER involve mobile shopping as well as pedestrian navigation systems. A central issue of these applications is the interaction style, which is multimodal i.e. consists of gestures, speech, handwriting, and a combination of these. On the basis of the user models that are provided by AGENDER, the shopping assistant is able to make a specific selection of products (e.g. digital cameras): When the speaker is recognized as being a female person, the system can e.g. choose a camera that is especially designed for women. Analogously, the navigation system can adapt the selection of alternative routes: When the user is a child, a tourist guide could choose sights especially interesting for kids. The vital interest in the AGENDER technology from telecommunications industry yielded another application domain, namely telephone-based services. Since call centers like that used for shopping or service hotlines cause high costs for the provider, the telecommunications industry is interested in solutions to improve the efficiency. Analogously to emotion recognizers, the AGENDER technology can be integrated into Automatic Call Distribution (ACD) systems. Besides this, telephone-based services involve spoken dialog systems (SDSs). Unlike call centers, SDSs are provided by speech-enabled computers instead of human agents. Railroad information systems currently belong to the best-known examples of such services but more and more product information and shopping systems are emerging in the market. The ambitions for improving SDSs through the AGENDER technology are related to the increase of consumer satisfaction: On the basis of the speaker model, the product shall be selected to meet the requirements of the respective customer group. At the same time, the dialog shall be adapted accordingly.