An empirical investigation of voice as an input modality for computer programming

Abstract

Recently, automatic speech recognition systems have shown the potential of becoming a useful means of data entry and control. The most successful of these speech recognition systems accept an isolated utterance as input and use a task-oriented syntactically constrained vocabulary for increased recognition accuracy. At the same time, language-directed editors are beginning to be introduced into the programmer's workplace. A language-directed editor is an editor that has knowledge of the underlying syntax (and basic semantics) of a language. Program entry, then, is syntax driven and program editing may proceed on a syntactic (semantic) basis. This dissertation discusses the design, implementation, and results of a controlled experiment to evaluate voice versus keyboard (the standard input mode) in a language-directed editing environment. Twenty-four subjects input and edited program segments under control of a language-directed editor via the two input modes. Measures of speed, accuracy, and efficiency were used to compare the two modes of input. In general the results showed that the subjects were able to complete more of the input and edit tasks by keyboard (70%) than by voice (50-55%), but that keyboard input had a higher error rate than did voice input. Also, the use of voice was just as efficient as keyboard for the input of editing commands. These results must be viewed with the understanding that the subjects were novices with respect to voice input but were very experienced with keyboard input. In this light, it can be seen that voice holds much promise as a mode of input for computer programming.