LLM-Based Teacher Tone Recognition for Educational Scenarios

Tone recognition plays a crucial role in E-education, enabling virtual teachers to enhance interactive experiences, provide personalized teaching, and flexibly s imulate diverse teaching styles, thus creating emotionally supportive learning environments. However, existing tone recognition research primarily focuses on general conversational contexts, overlooking the unique and fine-grained t one c haracteristics s pecific to educational settings. Educational tones, such as encouraging, guiding, critical, and neutral tones, are significantly m ore nuanced and differ markedly from everyday conversational tones. To address this gap, this study proposes a LLM-based teacher tone recognition method optimized for educational scenarios. Leveraging the strengths of large language models (LLMs) in language understanding and context analysis, we develop a teacher tone recognition method that processes and analyzes authentic classroom data to accurately classify various tone features. Experiments on 409 classroom utterances using four LLMs show that DeepSeek-V3 achieved the highest accuracy in teacher tone classification, with explicit task prompts significantly improving recognition performance.