The purpose of this paper is to suggest effective ways to develop an English-speaking testing tool using a speech recognition system. To achieve this purpose, this paper examines the recent development of speaking tests based on three categories: 1) human conduct and human scoring, 2) machine conduct and human scoring, and 3) machine conduct and machine scoring. It also analyzes and suggests more effective scoring schemes for current English speaking tests. Finally, this study also suggests detailed features that should be considered in creating an English-speaking testing tool to test learners' communicative competence using speech recognition systems: 1) enter predictable data, 2) add a 'Helper function' and a 'special signal function' to the database, and 3) limit conversation based on a 'Tree-Structure' to allow more student-directed conversation in the process of designing the speech recognition engine. |