Abstract

This study followed a mixed-methods approach to explore how Sri Lankan undergraduates viewed the use of ChatGPT Voice Mode (VM) and peer interaction (PI) in improving English-speaking proficiency. After experiencing both methods for one year, a questionnaire was distributed to 42 English as a second language (ESL) students to assess the limitations and strengths of each method. Preferences were almost evenly split when the participants had to choose between the two. ChatGPT VM was preferred for its non-judgemental and safe atmosphere, which helped students develop accuracy, confidence and reduce speaking anxiety. Furthermore, the participants stated that ChatGPT VM supported structured learning, particularly in grammar and vocabulary. As limitations of ChatGPT VM, complex answers provided, lack of naturalness and technical issues were cited. Difficulty in in comprehending pronunciation was viewed as another limitation. PI was preferred for its social quality and its ability to provide fluent, confident, and real-life conversations. Emotional support, authenticity and practical feedback were also cited as benefits of PI. However, minor concerns were raised regarding the varied proficiency levels of interlocutors. Therefore, as practical suggestions, educators and curriculum developers can integrate each method as complementary tools rather than mutually exclusive alternatives in order to enhance students’ overall learning outcomes.

Keywords: AI Mediated ESL Speaking, ChatGPT VM, Peer-Led Interaction, Sri Lankan University ESL Learners