Towards Conversational Speech Recognition for a Wearable Computer Based Appointment Scheduling Agent
Wong, Benjamin A.
McGuire, R. Martin
MetadataShow full item record
We present an original study of current mobile appointment scheduling devices. Our intention is to create a conversational wearable computing interface for the task of appointment scheduling. We employ both survey questionnaires and timing tests of mock scheduling tasks. The study includes over 150 participants and times each person using his or her own scheduling device (e.g., a paper planner or personal digital assistant). Our tests show that current scheduling devices take a surprisingly long time to access and that our subjects often do not use the primary scheduling device claimed on the questionnaire. Slower devices (e.g., PDAs) are disproportionately abandoned in favor of devices with faster access times (e.g., scrap paper). Many subjects indicate that they use a faster device when mobile as a buffer until they can reconcile the data with their primary scheduling device. The findings of this study motivated the design of two conversational speech systems for everyday--use wearable computers. The Calendar Navigator Agent provides extremely fast access to the user's calendar through a wearable computer with a head-up display. The user's verbal negotiation for a meeting time is monitored by the wearable which provides an appropriate calendar display based on the current conversation. The second system, now under development, attempts to minimize cognitive load by buffering and indexing appointment conversations for later processing by the user. Both systems use extreme restrictions to decrease speech recognition error rates, yet are designed to be socially graceful.