In this study, we proposed a method for generating advice based on classroom speech analysis and an automatic evaluation method using LLM-as-a-judge to support teachers. Using the LLM's Attention mechanism, which mimics teachers and children, we estimated the influence of teachers' speech, and generated advice by integrating the estimated results. Furthermore, we proposed a method for objectively evaluating the quality of the generated advice using automatic evaluation by the LLM. Experimental results showed that the advice generated by integrating the impact estimation tended to be highly evaluated by the automatic evaluation by LLM. Although the effectiveness of the proposed method was suggested, improvement of the automatic evaluation and its reliability are future issues to be addressed.
