Integrating Deep Learning and Natural Language Processing to Enhance Spoken English Proficiency Among EFL Students in Higher Education

Students of EFL must possess robust spoken English skills to converse effectively and maintain engagement in their language learning process. To prevent communication breakdowns and diminish interest in utilising the target language, it is essential to rectify verbal errors and enhance pupils' confidence in speaking, which is intricately linked to their linguistic proficiency. This study examines how to optimise corrective feedback, including students' preferences across all competency levels, notwithstanding the controversy surrounding its efficacy. To support this, is employed a preprocessing technique that involved normalisation to enhance the input quality of EFL learners' speech. Furthermore, argument topics were derived from EFL learning videos on YouTube utilising an automated LDA algorithm. The research introduces a hybrid optimisation model that integrates a RCN, which utilises dynamic capsule routing to capture spatial hierarchies and a recurrent layer to analyse temporal patterns in audio data. The experimental results indicate that the proposed model outperforms traditional architectures such as RNN, FNN, CNN, DNN, and basic RCN, achieving an accuracy of 97.45%, precision of 95.78%, recall of 93.91%, and an F1-score of 97.56%. These findings illustrate the potential of AI to enhance spoken error correction and, consequently, the communicative proficiency of EFL students in English.