A Study on the Recognition of English Pronunciation Features in Teaching by Machine Learning Algorithms

Abstract

A better understanding of students" English pronunciation features would be a useful guide for teaching spoken English. This paper first analyzed the English pronunciation features and extracted Mel-frequency cepstral coefficients (MFCC) features from the pronunciation signal. Then, the support vector machine (SVM) method was used to identify the cases of incorrect and correct pronunciation. To further improve the recognition effect, deep features were extracted using deep brief network (DBN) as the input of the SVM, and the parameters of both DBN and SVM were optimized by the sparrow search algorithm (SSA). Experiments were conducted on the dataset. The results showed that the MFCC-SSA-SVM algorithm had better recognition performance than the MFCC-SVM algorithm. The DBN-SVM algorithm had higher recognition correctness and accuracy than the MFCC-SSA-SVM algorithm, while the SSA-DBN-SVM method had 88.07% correctness and 85.49% accuracy, indicating the best performance. The results demonstrated the reliability of the proposed method for English pronunciation feature recognition; therefore, it can be applied in practical spoken language teaching.