Research And Application On Digital Speech Evaluation System

For many reasons, many people can not speak well or talk to a foreigner freely, although they have learned English for a long time and worked very hard. Pronunciation teaching with computer is one of the important ways to help improve users' pronunciation. Automatic Speech Recognition (ASR) based English pronunciation teaching system is the application of ASR in the field of English pronunciation teaching. It is well-known that the useful feedback can give the learners great help in language pronunciation leaning. The valid feedback from Computer-assisted Pronunciation Learning System (CAPL) can instruct the learners to pronounce correctly. This issue refers with many other subjects, such as speech recognition, language teaching, phonetics, linguistics and so on. Many issues on ASR based mandarin pronunciation teaching system are studied in this thesis. We do some research on pretreatment, feature extraction and automatic scoring arithmetic, we also analyze their characteristics. We show mainframe of the system, and discuss some key technologies and problems.At last, we design and implement an English Pronunciation Learning System. According to the characteristic of English pronunciation, we set an automatic scoring arithmetic. In this system, we use logarithm prior probability based on HMM and pronunciation length detection, and we integrate these two methods and use it in the system. For a better comprehension, we make linear transformations on the real score, and turn it to percent score which user can understand. In addition, we proclaim an evaluating mechanism for the system. This system can help users to improve their English pronunciation effectively.