Research on Diversified Innovative Paths of Vocal Music Teaching in College Music Education Promoted by Information Technology

The virtual reality vocal singing learning mode can enable online visualization of teaching content, interactive virtualization of teaching environments, and interactivity of the teaching process. Firstly, Log-Gabor [27-28] is used to amplify the detailed local features of the vocal speech map [29] after grayscale from different scales and directions, and then ILBP is used to extract and reconstruct the texture features from different directions. Two-dimensional Log-Gabor has a better function for enhancing image texture details, and it can suppress the amplitude of low-frequency signals and enhance the amplification of high-frequency signals. The 2D Log-Gabor is expressed as follows: 1 g=(u,v)=exp{−log(u1/u0)22log(k/u0)2}exp{−v122σe2}

In Eq. (1), u₁,v₁ denotes the function u_i = (ucosθ+vsinθ, v₁ = usinθ+vcosθ) of θ ; θ denotes the direction of the filter; u₀ denotes the filter center and , k,σ_v denotes the bandwidth in the u₁, v₁ direction. Assuming that the spectrogram is G¯ , the calculation of the convolutional neural network result is unfolded with g_a,y(u, v) in the time domain, and the following convolutional expression is obtained: 2 Gv,y(a,b)=G¯*gv,y(a,b)

Where v denotes the transform scale (u ∈ {0,1,2,⋯,4}), corresponding to five scales v denotes the transform direction (v ∈ {0,1,2,⋯,7}), corresponding to eight transform directions * denotes the convolution operation library. The ILBP is introduced to extract the local texture features of the image screen corresponding to vocal singing in the constructed virtual system. The image pixel point matrix is set to I , and the texture feature map IPXLRP of the image is expressed as follows: 3 Ip,RtBP=∑i=0p−1f1(gi−gc)2i,fi(q)={ 1,q≥00,q<0

In Eq. (3), R denotes the radius of the region g_c denotes the pixel value of the central image point P denotes the total number of all pixel points in the region and g_i denotes the pixel value of each pixel point.ILBP is optimized by taking the traditional LBP as the basis, and ILBP is able to effectively improve the solution to the problem that the traditional algorithms have a weak ability of local detail expression when extracting image texture features. The improved LBP is expressed as follows: 4 { IP,RLBP=∑i=0P−1F1(gi−gc)2F1(gi−gc)={ 1,| gigc−1 |>E0,| gigc−1 |≤E E=(1/P)∑i=0P−1| gigc−1 |

In Eq. (4), E represents the average absolute percentage error between the surrounding pixel points and the center pixel point, and by setting a threshold value for the absolute percentage error and the average error size of each pixel point, the absolute error relationship between the surrounding pixel points and the center point can be more comprehensively embodied, and then the LBP value obtained by the calculation will be more accurate.

The multilevel residual structure [30] maximizes the preservation of the original information by interconnecting the original information of the first n convolutional layers with the convolutional layers in which they are located, and regulates the dimensions of the input original features by adding control coefficients, which ultimately effectively improves the recognition efficiency of the model and the convergence speed of the system. If the input when the residual structure is elicited is set as xi, and the output after the residual structure is introduced is x_i+n , then the corresponding multilevel residual structure obtained is: 5 xi+n=σ(wi+nF(xi+n−1)+bi+n+αxi)={ σ(wi+nF(xi+n−1)+bi+n),n≠1βxi,n=1

In Eq. (5), α,β denotes the control coefficient, which is used to limit the dimensionality of the input features. If the loss function is C, the corresponding obtained expression for the weights of the backpropagation is updated to Eq. (6): 6 ∂C∂xi=(∂C/∂xi+n)(∂xi+n∂xi)=[ (β∏k=1nwi+k+α)xi+T(w,b) ](∂C∂xi+n)

In the above equation, (β∏i=1wi,1+α)x denotes the lost feature term corresponding to the supplement of the multilevel residual structure; T(w,b) denotes the constant term of weights and bias. When solving the convolutional neural network layer by layer, the weight v will gradually decrease until it is close to 0, which leads to the gradient update of the back-passing close to 0, and eventually causes the phenomenon of feature loss. Due to the existence of multi-category features in the process of vocal music teaching recognition, the cross-entropy is introduced into the vocal music teaching recognition, and its corresponding expression is as follows: 7 C=−∑j=1iy^jlog(yj)

In the above equation, y^i denotes the true label of the j nd sample; y_i denotes the predicted output of the 1st sample in the constructed model. In addition, it is also necessary to select an appropriate optimization algorithm to minimize the loss function of the neural network. The most common method is to use the gradient descent method to minimize the loss of two numbers, in this regard, this paper proposes to use the RMSProp algorithm [31], which is expressed as follows: 8 { r:=ηr+(1−η)(∂C∂w)2w:=w−α∂C/∂wr+ε

In the above equation, r denotes the sliding rate of the squared value of the gradient w denotes the decay rate a denotes the learning rate ε denotes the constant term that prevents the denominator from being zero and η denotes the hyperparameters. In addition to this, the regularization method is used in order to avoid the phenomenon of overfitting. This method is an effective training strategy capable of randomly ignoring the neurons in the neural network results.

The research methods commonly used in the study of diversified and innovative ways of teaching vocal music in higher education music education include survey methods, quiz methods, experimental methods, and so on.

Prior to the lesson, pre-test papers were distributed to Class A (experimental class) and Class B (control class) and the students were guided to actively complete them. The quiz scores were collected, compared, and analyzed to test whether there was any difference in the initial abilities of the two classes. An independent samples t-test was conducted on the scores of the two classes. The results of the significance analysis of the pre-test scores of the experimental and control classes are shown in Table 1. The table shows that the mean score of class A (experimental class) is 67.05 with a standard deviation of 16.27.The mean score of class B (control class) is 65.72 with a standard deviation of 15.14.The mean scores of the two classes are similar. The independent samples t-test of the pre-test scores of the two classes has a probability of significance p-value of 0.5248>0.05, indicating that there is no significant difference between the pre-test scores of the students in the experimental and control classes. The two classes had basically the same level of knowledge of musical instruments before the experiment. It indicates that there is no significant difference in the effect of vocal music teaching between the two classes of students before the experiment.

After the new class was taught, a post-test paper was distributed to Class A (experimental class) and Class B (control class), and students were guided to actively complete it. The quiz scores were collected and compared and analyzed to test whether there was a significant difference in the mastery of knowledge and skills in the experimental class after using the teaching resources of the VR technology-based vocal music education teaching platform. The results of the significance analysis of the pre- and post-test scores of the experimental class are shown in Table 2. As can be seen from the table, the significance probability P-value of the pre- and post-test scores of the experimental class is 0.000<0.05, indicating that there is a significant difference between the pre- and post-test scores of the students in the experimental class. Comparing the average scores of the pre- and post-tests, it is found that the students’ scores increased by an average of 11.89 points after using the teaching platform for vocal music education based on VR technology, and the standard deviation decreased from 16.27 to 11.36, which is a decrease of 4.91, indicating that there is a significant increase in the students’ learning effect of the vocal music after using the teaching of the VR platform, which suggests that the use of the VR technology can significantly increase the efficiency of the extraction of the vocal features, and reduce the difficulty of recognition, thus increasing the students’ mastery of vocal music knowledge and reducing the difficulty of vocal music feature extraction and recognition.

At the end of the classroom instruction, an independent sample t-test was conducted on the posttest scores of the two classes. The results of comparing the posttest scores of the experimental and control classes are shown in Table 3. As can be seen from the table, the p-value of the probability of significance of the posttest scores of the experimental and control classes is 0.000<0.05, which indicates that there is a statistically significant difference between the posttest scores of the experimental and control classes. In addition, the mean value of the posttest of the experimental class was 78.94 points, and the mean value of the posttest of the control class was 66.03 points, and the experimental class had an increase of 12.91 points compared with the control class, and the mean score of the posttest of the experimental class was significantly higher than that of the control class. The comprehensive survey results found that after the VR teaching platform was applied to the vocal music teaching classroom, the students’ learning effect and interest were significantly improved. This is obviously correlated with the application of the platform to the extraction and recognition of obscure and difficult vocal features in teaching. Because the application of the platform can more easily help students understand and master the knowledge related to vocal music, making their learning process easy and interesting, thus changing their learning attitude towards vocal music and increasing their motivation for vocal music learning.

The results of the survey on students’ satisfaction with the application of the VR teaching platform for vocal music teaching are shown in Table 4. The survey results showed that the mean of students’ overall satisfaction (very satisfied and satisfied) with the first part of the information technology-driven vocal music teaching and diversified innovations in music education was 92%, and that overall, the application of the teaching platform for vocal music education based on VR technology was appropriate. The student satisfaction survey on teachers’ choice of instructional media found that 95% of the students were satisfied that the instructional media helped them understand the music appreciation content.89% of the students were satisfied with the instructor’s choice of instructional media, which they felt stimulated their interest in music.82% and 99% of the students were satisfied with the instructor’s use of instructional media in the classroom because it allowed them to be more focused on the classroom content while enhancing their understanding and memorization of the music content. In terms of students’ feelings about the third part of the difference between this lesson and previous lessons, 96% of the students thought that the teacher’s way of teaching this lesson would allow students to learn more and enjoy the class better than previous lessons. Because more than 95% of the students thought that the use of the platform effectively solved the process of extracting and recognizing vocal features, and more carefully and patiently solved the difficulties they encountered in the learning process, the role of the platform was embodied to a greater extent. Taken together, the students were very satisfied with the application of the VR technology teaching platform to assist the teaching of vocal music in music education. This finding reflects the positive role of the VR technology-based vocal education teaching platform in the classroom.

After the lesson, this paper conducted an interview survey with the two music teachers who listened to the lesson, and conducted interviews and exchanges for the effect of music teaching practice, the performance and effect of the use of the teaching platform of vocal education based on VR technology, student participation, and opinions on the design of music teaching, and the following is a collation of the contents of the interviews of Teacher A and Teacher B. The interview survey of the teachers’ satisfaction with the application of the VR teaching platform for vocal teaching after the lesson is as Table 5 shown. Through the interviews and exchanges with the two listening teachers, it can be observed that their evaluation of the effect of music teaching practice is more positive. First, they pointed out that the achievement of teaching objectives was good, students showed higher learning motivation in class, the classroom atmosphere was more active, and students’ experience of music practice was improved. Second, the teachers were able to reasonably and effectively combine traditional and modern VR technology-based teaching platforms for vocal music education to address the key and difficult points in teaching, and to promote students’ feelings, imitations, experiences, and cooperative abilities in music. In addition, in terms of teaching design, the teachers’ teaching objectives were clear and focused, and the overall teaching design was clear and unambiguous, providing students with a good learning experience. Overall, teachers were satisfied with the overall teaching effectiveness.