Fluency.my: An Artificial Intelligence-Driven Multimodal Presentation Coaching Tool for Malaysian University English Test Preparation

Abstract

Malaysian learners preparing for the Malaysian University English Test face restricted access to individualised speaking rehearsal and feedback, alongside anxiety that can discourage oral practice. This study evaluated whether Fluency.my, a locally oriented multimodal speaking coach, provided a usable practice environment, reduced immediate self-reported speaking anxiety, and supported automated analysis of learner recordings. [MY INFERENCE] Design science research and multimodal learning analytics framed the platform as an educational artefact whose value depended on the joint operation of linguistic, vocal, and visual feedback. A mixed-methods pilot evaluation was conducted with 10 students at Management and Science University in Malaysia after each participant completed a platform session. The analysis descriptively examined System Usability Scale responses, adapted Foreign Language Classroom Anxiety Scale scores collected before and after practice, and speech-recognition and computer-vision performance metrics. The participants’ mean System Usability Scale score was 74.5. The mean adapted Foreign Language Classroom Anxiety Scale score decreased from 24.2 before practice to 19.1 afterwards, corresponding to the reported 21.1% immediate reduction. The speech-recognition pipeline recorded a word error rate of 11.4%, while face detection reached 94.2%. [MY INFERENCE] The study’s originality lies in integrating examination-aligned language coaching with speech, voice, and non-verbal analysis for Malaysian learners. [MY INFERENCE] Educators should use the platform as a supervised rehearsal aid while retaining human judgement for consequential speaking assessment.