This study examined the development and evaluation of an artificial intelligence (AI)–based system designed to detect emotional stress among students using thermal imaging and voice analysis. The primary goal was to develop a user-friendly software interface capable of real-time processing within a personal computer environment. Thermal and voice data were collected from 30 student participants in a simulated classroom setting to train and validate the AI model. The system integrated convolutional neural networks (CNN) for thermal classification and recurrent neural networks for voice sequence analysis to interpret physiological and acoustic indicators of stress. Results showed that the combination of thermal and voice inputs significantly improved the accuracy and reliability of emotional state recognition compared to single-input systems. The multimodal fusion model achieved 91.4% accuracy in classifying stress states, with a strong correlation between AI-generated and self-reported stress levels (r = 0.86, p < .001). The AI model also demonstrated consistent responsiveness and operational stability, supporting its potential application in classroom monitoring. Overall, the integration of thermal imaging and voice analysis presents a promising tool for helping educators understand students’ emotional well-being and enhance the learning environment.

Keywords: Artificial Intelligence (AI), thermal imaging, voice analysis, emotional stress detection, educational technology, multimodal fusion