Annotating Software Requirement Specifcations with Mobile Speech Recognition

Abstract

Requirements are crucial in software engineering and there is ample support for how to elicit and document them; however, relatively little support exist for requirements maintenance. We argue that lightweight methods for annotating requirements are needed and present a system based on speech recognition to enable it. This paper describes the system design and a set of experiments and user tests to validate its use. For more realistic evaluation our system has been adapted to the commercial requirements management tool SystemWeaver. Results show that the accuracy of free text speech input is not high enough to enable free-form addition and edits to requirements. However, requirements identification and annotation is practical by extending the system with string distance calculations to the set of requirements being matched. Lookup times on industrial requirements data was deemed practical in user tests, in particular for annotations on the go and during requirement reviews.