Speaking skills remain one of the most challenging competencies to develop among EFL learners, particularly in junior high school settings where teacher-centered instruction and limited communicative practice contribute to low oral proficiency and high speaking anxiety. This study addresses the gap in empirical research on combining AI voice-interactive tools with structured role-play activities to improve students' speaking performance. Using a Classroom Action Research (CAR) design, the study was conducted at MTs Miftahul Huda Silir Wates during the 2025/2026 academic year, involving 29 eighth-grade students. The intervention integrated Google Gemini's voice-interactive features as a low-stakes AI speaking partner with structured role-play scenarios over two meetings in a single cycle. Speaking performance was assessed using an analytic rubric covering fluency, pronunciation, vocabulary, grammar, and comprehension. Results showed that the class mean score increased from 55.00 (pre-test) to 78.79 (post-test), with 82.76% of students surpassing the minimum mastery criterion of 75. The most notable gains were in fluency (+55.41%) and grammar (+64.77%). Qualitative observations also revealed reduced speaking anxiety and increased voluntary participation. These findings suggest that combining AI voice interaction with role-play provides an effective dual-scaffolding approach to develop oral skills and lower affective barriers in secondary EFL classrooms.
Copyrights © 2026