English Pronunciation Quality Evaluation Based on SAF-RNN
Yue Hu · 2025
Language learning systems that integrate Automatic Speech Recognition (ASR) technology have proven effective in enhancing foreign language pronunciation through targeted pronunciation practice. The present pronunciation quality evaluation with respect to Goodness of Pronunciation (GOP) utilizes the posterior probabilities of Acoustic models. However, such approaches suffer from generalization issues since they are employed for identifying a score metric for every phoneme instead of completeness or comparison through optimal exclamation of the words. Thus, this article proposes a Softplus Activation Function based Recurrent Neural Network (SAF-RNN) approach for the quality evaluation of English pronunciation. The feature extraction is performed to concentrate on important attributes of pronunciation and after fused by the neural network for effective quality evaluation. The experimentation of proposed SAF-RNN approach comprises of 100 college students such as 50 male and 50 female for evaluating their English pronunciation. The experimentation outcomes illustrate that proposed SAF-RNN approach reaches an optimal accuracy of$\mathbf{9 9. 5 6 \%}$as compared to other existing approaches. These identifications prove the effectiveness of introduced approach, demonstrating its capability to enhance pronunciation accuracy as well as fluency.