Segmental and suprasegmental pronunciation training via AI: An exploration of university students’ perceptions and attitudes

Abstract

Artificial Intelligence (AI) has emerged as a transformative tool in language learning, particularly for improving pronunciation. Although recent studies have highlighted AI’s effectiveness in pronunciation training, there has been limited in-depth investigation into university-level English majors' perceptions of using AI for pronunciation practice, especially focusing on two key features: segmental and suprasegmental aspects. Therefore, this research aimed to explore the perceived benefits, challenges, and attitudes of English as a Foreign Language (EFL) students, specifically university-level English majors, using AI to practice these two pronunciation features. A mixed-methods design was employed, combining a questionnaire with semi-structured interviews and journal diaries. The researchers constructed a questionnaire focusing on segmental and suprasegmental features among 70 participants recruited by purposive sampling technique. This questionnaire was constructed by the researcher with critical reviews from six experts in the TESOL field. Qualitative data from semi-structured interviews and journal diaries were collected from 5 participants recruited by convenience sampling technique. The qualitative instruments were theoretically based on the Tripartite Model of Attitudes (Rosenberg & Hovland, 1960). Findings revealed that students found AI tools helpful for improving segmental pronunciation features, especially long vowels, but encountered challenges such as technical inaccuracies. Suprasegmental features, like word stress, showed moderate improvement, while pauses, speech chunking and connected speech received insufficient feedback. Despite these challenges, most students expressed enthusiasm for AI’s convenience and personalized support. This study presents implications for theory and offers practical insights for teachers, English majors, and AI developers.