Aiming at the research hotspots of speech emotion recognition in human-computer intelligent interaction, noisy speech emotion recognition based on the multi-task constrained teacher-student model is designed as a research-oriented teaching experiment. In this experiment, the guiding role of the teacher model, the learning process of the student model and the constraining force of multi-level enhanced loss are observed. The design is based on the Wav2vec 2.0 teacher-student model and the multi-level enhanced loss mechanism. A speech enhancement auxiliary task is introduced into the student model, enabling it to acquire the feature representation ability of the teacher model through learning. In the testing phase, the student model directly extracts key emotional features from noisy speech for emotion classification. Finally, a large number of experiments are conducted to analyze the performance and robustness of the emotion recognition system. The experimental design based on the teacher-student model helps to improve students’ thinking ability, scientific research innovation and exploration awareness.