Music development cannot be detached from the development of computer technology, which will not only affect the music creation and performance process but will also inevitably have an impact on the development of means for music theory education. Based on the four focuses of music theory knowledge, the realization of computer music data expression through MIDI, the establishment of the music generation model by integrating RNN, GAN, and VAE, and saving the four-dimensional music information matrix in the music dataset, the MIDI file data feature extraction is conducted to get the monophonic and chord sequences of all musical instrument data sets, and the optimization of MIDI music generation models by WaveNet predictions. By analyzing the automatic composition results of music model RNN-VAE-GAN by using objective criteria, detecting the pitch variation of the main melody produced by RNN-VAE-GAN, and measuring the consistency of the note pitch variation of the MIDI music monophony composed by the model with the pitch variation path of the notes. According to the test results combining pre- and post-test and control experiments, comparing the effect of music learning between the experimental group and the control group, P-value of the core music literacy of the two groups was 0.005, and P-value of music learning interest of the two groups was 0.001, indicating a significant difference, which demonstrates a high level of progress in computer-assisted pre-school music theory teaching model application practice.
