English as a Second Language (ESL) education faces momentous challenges including restricted personalized feedback and scalability constraints in large classrooms. This study developed and assessed an innovative AI-driven oral assessment tool that incorporates generative artificial intelligence with Internet of Things (IoT) technology to make over adaptive learning environments for individual learners. The research used a mixed-methods strategy, developing the tool using datasets of L2Arctic and Libri-speech, also assessing it through both qualitative human validation including ESL teachers and metrics of quantitative performance. Key indicators of performance constituted learning rate optimization, model accuracy and proportion balancing of dataset. The results have demonstrated that the G-ASR AI tool has gained 94.7% precision accuracy on datasets of native speaker and 86.6% on datasets of non-native speaker, with optimum performance by self-correction feedback and 60% AI to 40% ratio of teacher interaction. Human validation crosswise 24 ESL teachers and 240 students discovered large effect sizes (Cohen’s d > 1.6) crossways learning outcomes, specifically self-regulation abilities (d = 2.14) and metacognitive knowledge (d = 1.98).
