A Study on Correcting English Pronunciation for Non-English Major Students Based on Artificial Intelligence Speech Recognition Technology

This paper introduces a speech recognition and pronunciation correction system based on the Transformer model. The Transformer model uses the self-attention mechanism to capture the pronunciation timing and context information, and uses an end-to-end method to map the pronunciation signal to the correction suggestion. The model identifies the error type by calculating the pronunciation difference and generates personalized feedback. To improve performance, optimization strategies such as data enhancement and adding self-attention structure are adopted. The experiment uses a dataset containing 1,000 student samples and compares it with the recurrent neural network (RNN) and long short-term memory network (LSTM) models. The results show that the Transformer model-based system outperforms the traditional model in terms of accuracy (94.2%), recall (91.5%), F1 value (92.8), correction rate (87.6%), and other indicators, with an average response time of only 0.75 seconds. After using the system, students' pronunciation accuracy improved by more than 30%, and the improvement was more significant in the intermediate and advanced groups. More than 90% of students were satisfied with the system experience. In quiet and background noise environments, the Transformer model has strong stability and anti-interference ability, showing high robustness and practicality.