The perceived utility of speech systems relies in part on the success with which they compete with more established computer interfaces. With the exception of certain tasks (such as dictation), speech interfaces have not made significant inroads in the desktop domain; on the other hand telephone-based applications are becoming established, as speech provides an effective high-bandwidth channel between human and computer. An emerging and possibly even more important domain is that of ``wearable'' systems consisting of small computers that can be easily carried on the person. While providing significant computing and communication power such systems have difficulty accommodating conventional interface devices such as keyboards, mouses and displays. An obvious alternative is speech, both for input and for output. The present paper describes an initial attempt to build such an interface in the context of a system for mobile inspection.
The task we chose was initially developed as part of the VuMan project at Carnegie Mellon University. The VuMan has been used for a limited technical inspection (LTI) of an amphibious assault vehicle for the USMC at Camp Pendleton, as a replacement for a clipboard and pencil procedure. The VuMan allows a mechanic to directly enter inspection data into a computer and has been shown to reduce inspection time by a half.
Despite this, the VuMan has a number of limitations, particularly a very low-bandwidth input device, the ``rotary mouse''. Input activity consists of circularly traversing hot-spots on a display using a dial on the device and clicking on spots corresponding to desired inputs. In the worst case, the user is shown the image of a keyboard and needs to enter data character by character using the mouse. Given this, speech seemed like an obvious enhancement to the task.