A Multimodal Dialogue System for Interacting with Large Audio Databases in the Car

Abstract:

The variety and complexity of audio storage devices nowadays available in automotive systems turn selecting audio data into a cumbersome task. How can users comfortably access particular audio data in a variety of media carriers containing large amounts of audio data while actually pursuing the driving task? Browsing logical and physical hierarchies can heavily increase driver distraction. This paper proposes a speech-based approach that significantly facilitates accessing database items. It consists of two interaction concepts: the concept for categorybased search requiring pre-selecting a category (e.g. artist, title, genre, etc.), and the concept for category-free search allowing the user to search globally across all categories. In both concepts search space comprises all media carriers. The user may directly address audio data items by saying the corresponding name. However, evidence was taken from address book data that users do not perform well when it comes to remembering the exact name of an item. We therefore create additional wording variants to allow for fault-tolerant search. This is achieved by means of filter- and recombination rules. Thus users, who do not remember an item correctly are still able to find it when speaking only parts thereof.


Year: 2007
In session: Anwendungen im Kraftfahrzeug
Pages: 202 to 209