Computer Analysis-based Pronunciation Quality Assessment In Language Learning System

The focus of modern English education has changed, which concerned from students'grammar and reading comprehension skills to listening and speaking skills. Oral English training and evaluation thereof have abstracted more and more attentions. At the same time, because of network technology and speech signal process technology's great development, as well as a crush increase number of oral English examinees, it is necessary to build and develop a computer design-based oral English test system.In the project, one of important functions of computer analysis-based pronunciation quality assessment was accomplished, which means following standards. Trainers would choose a demo from an established speech training bank depending on their requirements. When test starts, the system extracts two feature coefficients. Through comparing them, calculating a Euclidian distance of feature coefficients between standards and trainings, it can judge the pronunciation quality. All these referred theories include: speech signal pre-processing, speech clustering algorithm, feature extraction, and distortion measurement.Following key innovations were included. On one hand, in order to decrease the system's error rate and get closer to experts'judgments, a new feature coefficient was adopted. On the other hand, a special method to calculate the Euclidian distance between standards and trainings was proposed, so as to give a reasonable score finally. Its achievement can also benefit other fields of language test to some degree.