English Pronunciation Error Detection Method Based on Multiple Model Fusion
Liang Meng · 2023
In order to solve the problem of correcting accuracy of English pronunciation error automatic correction system, this article proposes a model training strategy based on multi-model fusion, which can make full use of the training data and solve the problem of the imbalance of positive and negative samples caused by the lack of true pronunciation error data to a certain extent. Glds-svm and GMM based method are fused to further improve the performance of pronunciation error detection. The experimental results show that the equal error rate of the fusion system of GLDS-SVM and GMM-UBM on the simulation and real test sets is 9.92% and 16.35%, respectively. The experimental results indicate that GLDS-SVM has obvious advantages over traditional radial basis function kernel method in terms of model space and operation speed.