A study on the application of optimization algorithms in vocal music teaching strategies

Abstract

Abstract Subjective interference is a common difficulty in vocal music teaching, and human ear audition cannot fully objectively analyze the students’ problems in vocal practice due to the influence of environment and other factors. This paper takes the convolutional neural network as the vocal music recognition algorithm and the Mel spectrum as the vocal music feature extraction algorithm and constructs the vocal music analysis model based on the optimization and improvement of the two algorithms. Then select the support vector machine, the nearest neighbor node, Wavenet, LSTM, GAN, SAGAN, CLDNN_BILSTM, and other models, as well as this paper’s model, for comparison experiments. Finally, the model was utilized in the vocal education classroom to evaluate the singing practice of four students. It is found that the MSE value of Arousal’s algorithm in this paper is the lowest, and the R2 values of 0.51197 and 0.71058 are the highest in the test of the MFCC vocal music feature dataset. Valence’s model in this paper has the MSE value of 0.51996, which is still the lowest, and the R² value of 0.76946, which is still the highest. This paper’s model has the best performance and results. The average rate of professional singers is 61 beats, and the model calculates the average singing rate of the four students as 77, 66, 63, and 61 beats. The first three still have a large gap compared to the standard level, and the student D level is higher. The problem of student practice analysis and vocal feature extraction and recognition in vocal teaching can be solved using new ideas and methods provided in this study.