Speaker
Description
Generative AI voice chat offers EFL learners accessible opportunities to practise speaking, yet its pedagogical value depends on how learners use its affordances and respond to its limitations. This study examines Vietnamese university students’ expectations of AI-supported speaking practice before a classroom activity and their reported experiences, perceived benefits, and challenges afterwards. An exploratory classroom-based mixed-methods survey was conducted at two time points. The pre-activity survey received 114 responses, while 78 students completed the post-activity survey. Quantitative data were analysed descriptively, and open-ended responses were examined through inductive–deductive content analysis. Findings from the two datasets were integrated at the aggregate level.
Before the activity, students primarily expected AI to support vocabulary and language accuracy, pronunciation and fluency, idea development, content organisation, and question-and-answer preparation. Afterwards, students reported perceived benefits in developing ideas, organising responses, practising pronunciation, building confidence, and rehearsing question-and-answer interactions. However, they also encountered speech-recognition errors, premature interruptions, rapid responses, limited vocabulary, prompting difficulties, technical disruption, and interaction that sometimes felt mechanical or emotionally limited. Some students also expressed concerns about dependence on AI. The findings suggest that AI-supported speaking activities require explicit prompting guidance, turn-taking support, reflection, and continued teacher and peer interaction. The study contributes to discussions of learner agency, AI literacy, and responsible AI-supported speaking pedagogy in TESOL.
Keywords: generative AI, EFL speaking, voice chat, learner agency, AI literacy
