A Method of Speech Separation between Teachers and Students in Smart Classrooms Based on Speaker Diarization

The analysis of classroom teaching behavior is of great significance to improve the teaching quality of teachers, but there are few researches on teaching behavior and teaching model analysis based on speech processing. This paper proposes a method of speech separation between teachers and students based on speaker diarization, which can help educators distinguish the teaching model objectively by identifying and analyzing the speech acts of teachers and students in the classroom. Firstly, the speaker segmentation based on Bayesian distance is used to detect and segment the classroom audio. Then, based on the pre-trained speaker model, speaker detection and clustering are performed on multiple segments of speech processed by speaker segmentation. Finally, the speech acts distribution in the classroom is visualized, and the teaching model is distinguished based on the Student-Teacher analysis method. This paper uses classroom audio collected in the classrooms of two elementary schools as the experimental data. The experimental results show that the method proposed in this paper achieves 74.0% and 98.28% accuracy in speaker segmentation and speech recognition, respectively. And this experiment takes a Chinese open class as an example to analyze and discuss its teaching model in detail. The research can help teachers summarize classroom teaching effect in time, and improve the quality of classroom teaching.