Pronunciation quality evaluation model for English reading speech based on acoustic model adaption and support vector regression

Aiming at the case that the accuracy of Chinese students′ English pronunciation evaluated by standard American English acoustic model is low, a pronunciation quality evaluation model for English reading speech based on acoustic model adaption and support vector regression is proposed. Firstly, maximum likelihood linear regression and maximum a posterior algorithms are used to adjust acoustic model to match the English pronunciation of Chinese students. Then, different types of scoring features are extracted from three aspects: pronunciation accuracy, pronunciation fluency and reading completeness. Finally, support vector regression algorithm is introduced to combine these scoring features into an ultimate pronunciation quality score. Experimental results show that the proposed model can effectively improve the performance of pronunciation quality evaluation.