Speech Recognition's Application In Computer-assisted Language Learning

Many Applications of computer-assisted language learning system lack of evaluation & feeedback of learning speech. In language pronunciation learning, however, most help for learners comes from effective feedback. Based on speech recognition technology, evaluations could be given by calculating similarity of speech and reference speech or reference model. These methods are HMM and feature comparision based scoring.After some introduction and research of various fundamental theories of speech recognition and scoring methods, a pronunciation error detection algorithm that use HMM force alignment to detect missed and mistakenly read phonoms is proposed. Then we bring forward a pitch synchronous overlap-add algorithm based automatic prosody modify method that synthesize speech with reference prosody and learner's timbre. Recommendations for improvement queried from correction knowledge base offer learners a timely help.Then we design and implement an English pronunciation learning software. Functions such as multi-type pronunction scoring, error detection and prosody feedback are provided. System can help users to improve their English pronunciation effectively.