EASES: A Study of an English Speaking Assessment System Based on Enhanced AI Integration Modeling

English speaking assessment is of great significance in education and career development, but traditional manual evaluation methods often suffer from subjectivity and inconsistency. To address these issues and improve the accuracy and fairness of evaluations, this paper proposes and validates an Enhanced AI-based Speaking Evaluation System (EASES). The EASES framework combines Convolutional Neural Networks (CNN) for extracting audio features, Long Short-Term Memory (LSTM) networks for extracting language features, and an XGBoost model for comprehensive scoring. Experiments conducted on a dataset of 3100 speech samples show that the EASES framework significantly outperforms individual models in evaluation accuracy. Specifically, the EASES framework achieved an accuracy of 90.29% after 25 training epochs, markedly higher than the 80.98 % accuracy of the CNN model and the 82.98% accuracy of the LSTM model. The experiments validate the effectiveness and superiority of the EASES framework in English speaking assessment, providing more accurate and consistent scores, and contributing significantly to the efficiency and fairness of English speaking evaluations.