An GMM-HMM Based Intelligent Scoring Mechanism for Oral English Pronunciation

Abstract

To improve the ability of intelligent detection of oral English pronunciation errors, this paper proposes an intelligent scoring model for pronunciation based on related speech recognition principles. The scheme is based on automatic speech recognition technology. First, MFCC is used to process the speech signal from the oral test system. Second, the corpus is standardized by the posterior probability scoring method based on the Hidden Markov Model(HMM), to fit the distribution of speech feature observation vectors. Finally, the pronunciation error detection is recognized according to the spectrum difference. The simulation results show that the mechanism presents high accuracy in the automatic evaluation of oral pronunciation quality, which helps to improve the self-learning ability of English learners.