Evaluation for Liaison of Spoken English: a Sugeno Integration Approach

The liaison evaluation for spoken English is one of the key problems for computer aided spoken language learning. Though a lot of factors affect the performance of a spoken language evaluation algorithm, there are mainly two factors that contribute to the most of the obstacles, i.e. the natural casualness of spoken language and the unstable performance of existing speech processing systems. In this paper, the Sugeno integral is introduced to address the problems faced in liaison evaluation. It models the casualness of the spontaneous speech sound and the unstableness of the speech processing system via fuzzy measures and reliabilities respectively, and then fusing them under the Sugeno integral framework to output the linguistic rather than quantitive score. The experiment shows that, under the 58% average recognition rate of the system, our liaison evaluation model get very reliable and stable results for both closing and opening test, which encourage us to apply this model to other tasks of the spoken language evaluation in the future