Chapter 13. Optimizing Accuracy and Recovering from Errors


High recognition accuracy is at the core of speech interface usability. Accuracy is key for all user interfaces; the system must correctly "understand" the user's intent and must respond appropriately. However, accuracy takes on special importance for voice user interfaces, for a number of reasons.

First, the impermanent and dynamic nature of spoken interactions makes it hard for the user even to identify errors when they happen, much less take appropriate action to correct them. Second, the technology is imperfect; there will be errors, especially in difficult acoustic environments, such as those that have background noise or cellular channel distortions. Therefore, it is essential to do whatever is possible in the dialog design to prevent errors and recover effectively when they do happen.

And problems with accuracy are exacerbated by the fact that callers usually can't understand why the system made the error. Although some errors may be obvious (for example, "Boston" and "Austin" sound similar and may get confused), the reasons for most recognition errors are obscure. As a result, errors typically cause confusion for callers as well as reduce their faith in the "rational behavior" of the system. Every error degrades the usability of the system. The best-designed and most graceful error recovery approaches cannot compensate for poor accuracy.

Evidence gathered from deployed systems shows that recognition accuracy is highly correlated with both task completion and user satisfaction. This is also supported by laboratory data. Using a VUI experimental framework called PARADISE, researchers at AT&T analyzed three applications: two voice-enabled e-mail systems and a train information system. They found that recognition accuracy was the largest contributor to user satisfaction (Kamm, Walker, and Litman 1999).

In an experiment conducted at the Centre for Communication Interface Research (CCIR-1 1999) at the University of Edinburgh, user satisfaction was assessed for the same speech system at four accuracy levels. The researchers found that subjective attitudes toward the system improved as the accuracy increased.

There are a number of contributors to recognition accuracy. Many of them are related to acoustic modeling and search techniques and are independent of the individual application. However, the way you design your dialogs and grammar affects both accuracy and perceived accuracy. This chapter discusses dialog design guidelines that can contribute to accuracy. We then describe approaches for error recovery: handling those cases when accuracy fails.



Voice User Interface Design 2004
Voice User Interface Design 2004
ISBN: 321185765
EAN: N/A
Year: 2005
Pages: 117

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net