<![CDATA[Developing effective speaking proficiency, particularly in pronunciation, accuracy, and fluency, remains a persistent challenge for EFL learners in Indonesian higher education. In response, this study investigates the impact of integrating ELSA Speak, a voice-based AI application, on enhancing these core aspects of speaking performance. Employing an explanatory sequential mixed-methods design, the research involved 23 Indonesian university students over an eight-week intervention. Quantitative data were collected through pre-tests and post-tests using CEFR-aligned speaking assessments, with paired-samples t-tests employed to examine the statistical significance of performance gains. Test validity was ensured through expert review and alignment with internationally recognized rubrics. Results indicated a statistically significant improvement in students’ overall speaking scores, particularly in pronunciation and accuracy. To complement these findings, qualitative data were gathered through semi-structured interviews and analyzed thematically. Students reported that ELSA Speak provided constructive, personalized feedback, context-rich materials, and a flexible, motivating learning environment. However, limitations were noted, including restricted access to premium features, internet connectivity issues, and low self-regulation, especially in managing independent learning. These findings underscore both the pedagogical benefits and practical challenges of integrating AI-powered tools like ELSA Speak into EFL speaking instruction. ]]>
