<![CDATA[classroom time, speaking anxiety, and insufficient authentic interaction often restrict learners’ opportunities to develop fluency and confidence. Although AI-based speaking tools have increasingly been used in language learning, previous studies have emphasized individual AI-mediated practice and measurable proficiency outcomes, while limited attention has been given to human–AI collaborative speaking practice and learners’ experiences in the Indonesian EFL context. This study investigated the effect of human–AI collaborative speaking practice using SmallTalk2Me on learners’ speaking skills and anxiety, while also exploring learners’ perceived benefits, challenges, and views of AI and human interaction. A sequential explanatory mixed-methods design was employed with 20 intermediates Indonesian EFL learners. Quantitative data were collected through IELTS-based speaking pre-tests and post-tests and a Foreign Language Speaking Anxiety questionnaire, while qualitative data were obtained from semi-structured interviews with five participants. The results showed significant improvements in all speaking components: fluency increased from 5.17 to 6.32, lexical resource from 5.68 to 6.50, grammatical accuracy from 4.84 to 5.66, and pronunciation from 5.48 to 6.33. Paired-sample t-tests confirmed significant gains across all components, with p = 0.000. Learners’ speaking anxiety also decreased significantly from 3.20 to 2.75, t(19) = -13.42, p = 0.000. Human–AI collaborative speaking practice effectively enhanced learners’ speaking performance and reduced anxiety by combining AI-generated feedback with meaningful human interaction.]]>
