|
This paper introduces a new architecture for the design and
development of speech user interfaces. In this design, the speech
input and output capabilities are implemented using the smallest
reasonable components and combining these components into
sophisticated interfaces by hierarchical composition,
cross-referencing, customization and extension. To define legal
spoken input, a speech user interface component comprises any
number of input expressions each of which concisely defines
syntactic, semantic and activation constraints plus support for
composition by cross-reference to other input expressions.
Similarly, components contain output expressions with equivalent
compositional capabilities for dynamic generation of spoken
output. The component design follows the Model-View-Controller
architecture. The architecture has been applied to the development
of multi-modal desktop applications and speech-only dialog
systems. The benefits of this fine-grained component-based
architecture include simplification of speech application
development, enhancement of speech user interface consistency by
automatic generation of common patterns, and improved
maintainability and readability of speech software. In analyzing
the architecture, we explore some of the fundamental differences
between the requirements to support graphical interfaces and
speech interfaces.
|
