Development and application of a speech recognition-based pronunciation assessment system for university English oral teaching

This study investigates the application of advanced speech recognition technology to address challenges in university English pronunciation teaching. An automated pronunciation assessment system was developed based on a transformer-driven acoustic model and multidimensional scoring metrics, enabling real-time evaluation of pronunciation accuracy, fluency, and prosody. A controlled experiment involving 120 non-native English learners compared the system with traditional teacher-led assessments. The experimental group demonstrated significant improvements in phonetic accuracy, with a marked reduction in pronunciation errors, and better alignment of speech rhythm compared to the control group. The system also provided consistent, objective feedback and reduced assessment time by over 80%, making it highly efficient and scalable for large-scale educational environments. Furthermore, this research integrates artificial intelligence into language education, offering innovative tools to support personalized, data-driven, and adaptive learning. The findings highlight the system's potential to enhance the quality of oral English instruction by addressing the inefficiencies and subjectivity of traditional methods. This study contributes to the development of smart education by providing a practical and effective solution for improving students' speaking skills, paving the way for broader applications of AI technologies in language learning and beyond.