AbstractsComputer Science

Intuitive speech interface technology for information exchange tasks

by Hansjörg Hofmann

Institution: Universität Ulm
Department: Ingenieurwissenschaften und Informatik
Degree: PhD
Year: 2015
Record ID: 1104472
Full text PDF: http://vts.uni-ulm.de/docs/2014/9294/vts_9294_13968.pdf


Smartphones are considered as people’s companions and help users to get instant access to the Internet anytime and anywhere. However, the manual use of smartphones is only appropriate in situations, where the actual smartphone use is in focus. In so-called dual-task scenarios people perform two tasks in parallel and often different priorities are given to each task. For instance, if a driver would like to send an email while driving, it is more important to drive safely than to use the smartphone manually to send the email. In order to optimize the performance of both tasks intuitive spoken dialog systems (SDSs) can help to access the Internet as secondary task by little impairing the primary task performance. Until today there is no persuading, consistent and sophisticated speech dialog concept, which allows users to control the Internet by speech. In the thesis at hand different speech-based human-machine interaction concepts for Internet access by speech as secondary task in a dual-task scenario are compared. Due to the urgency and the high risks the automotive environment has been chosen as example scenario. Information exchange tasks (such as hotel bookings or sending emails) require the design of multi-turn dialogs and the functionality to provide users with proactively incoming information. As these tasks demand a lot of attention of the user and thereby might impair the primary task performance, this research focuses on the development of SDSs to perform information exchange tasks. In order to find the most appropriate speech interface for the performance of multi-turn dialogs and to handle proactively incoming events several in-car SDS concepts were designed, prototypically implemented and evaluated in driving simulator studies in terms of usability and driver distraction. Based on the result the thesis at hand provides first clear guidelines for in-car SDS developers, which help to design Internet enabling speech interfaces in the driving environment.