This research aims to eliminate data processing bottlenecks (manual transcription and labeling) in a system that generates personalized advice for teachers.
We developed a system that utilizes multimodal AI to automatically execute the entire process—from inputting classroom audio data, through speech analysis and LLM-based reconstruction of teacher-student interactions, to generating concrete advice—in an **end-to-end** manner.
This enables the immediate provision of feedback that previously took hours, realizing practical and highly timely reflection support to enhance teachers' instructional skills. This system significantly contributes to streamlining teacher training.
