A Fuzzy Pronunciation Evaluation Model for English Learning

The evaluation of pronunciation for spoken English is one of the key problems for computer aided spoken language learning. While the most of researchers focus on the improvement of speech recognition to build a reliable evaluation system, there still needs a model that fuses the reliabilities of existing speech processing systems and the learner personalities into the evaluation system. In this paper, the Sugeno integral techniques are introduced to solve this problem. At first, the English phonemes that are hard to be distinguished (HDP) from Chinese language learner are collected and are grouped into different HDP sets. Then, the system reliabilities for distinguishing the phonemes within a HDP set are computed based on the standard speech corpus and are integrated with the phoneme recognition results under the Sugeno integral framework. The fuzzy measures are given for each subset of speech segments that contains at least 10 occurrences of phonemes within a same HDP set. Finally, the linguistic evaluation results are given by the Sugeno integral model based on system reliability and fuzzy measures. The experiment taken on Sphinx-4 shows that, under the 84.7% average recognition rate of the system, our pronunciation evaluation model get reliable and stable results for 3 test corpora