Design and optimization of higher vocational English listening and speaking training system based on speech recognition technology

At present, in English teaching in higher vocational colleges, the traditional mode is difficult to meet the diverse learning needs of students, especially the cultivation of listening and speaking ability. The introduction of speech recognition technology provides a new way to solve this problem. The system designed in this study is based on hierarchical architecture, including front-end presentation layer, business logic layer, data processing layer and back-end storage layer, and has three functional modules: speech recognition, listening and speaking training and feedback and evaluation. Aiming at the applicability challenge of the existing speech recognition algorithm in higher vocational education, this paper puts forward the idea of improving the algorithm, including hybrid enhancement strategy, dual-path antagonistic learning model, Transformer-LSTM hybrid encoder and hierarchical knowledge distillation, so as to improve the robustness and generalization ability of the model. The experimental results show that the improved model outperforms the baseline model in key metrics including Word Error Rate (WER), Real-time Factor (RTF), model size, and Pronunciation Correlation Coefficient (PCC). Specifically, WER is reduced by 23%, RTF is decreased by 46%, the model size is compressed by 60%, and PCC is increased by 19%. Furthermore, the robustness of the improved model in various noisy environments is significantly enhanced. This study provides strong support for the informatization reform of listening and speaking teaching in higher vocational English education, demonstrating the application potential of speech recognition technology in the educational field.