Development of music teaching software based on neural network algorithm and user analysis

doi:10.21203/rs.3.rs-2725108/v1

Download PDF

Research Article

Development of music teaching software based on neural network algorithm and user analysis

https://doi.org/10.21203/rs.3.rs-2725108/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 12 Jun, 2023

Read the published version in Soft Computing →

You are reading this latest preprint version

At this stage, music teaching is facing an increasingly serious shortage of teacher resources. Therefore, it is particularly important to develop a music teaching software by using computer assisted music teaching activities. First of all, the operation method program of this system is carefully designed according to the principles of computer network technology. Using the performance characteristics of Fourier transform and its enhanced functions to extract music, the priority key system modules are designed according to the system structure framework and data processing program, and the main design code is provided. With the development of artificial intelligence technology, neural network has gradually become an important research method in this field. And compared with the traditional mechanical learning methods, the neural network based method has the advantages of simple algorithm mode, good universality, strong robustness, organic and mobility. With the rapid change of in-depth learning technology, music teaching software has shown great overall advantages in the accuracy and speed of detection. In addition, this paper analyzes the specific user level of music teaching programs, focusing on their interest in and specific acceptance of these music teaching programs, as well as the use of user feedback to develop specific and effective music teaching programs. Neural network algorithm and user analysis provide a new strategy for developing music teaching software.

neural network

User analysis

Music teaching

software development

With the increasing level of China's economic development, the living standard of the Chinese people has been significantly improved, the material life has also been significantly improved, and the people's spiritual life has also been constantly enriched [1]. Many people are looking for all kinds of music as an important way to cultivate emotions, because it can make many people feel calm. Although most people love music education at present, they rarely engage in music education. In addition, many music teaching, such as piano teaching, is often accompanied by piano teaching environment, time requirements and other factors, and needs to use various teaching methods one by one [2]. Therefore, At this stage, the popularity of such music is not high. For this reason, music learners need to develop a specific music education program, use computer automatic identification technology, automatically record the information learned by trainees, compare the music information in the software system, conduct on-site assessment of the impact of participants' music expression, find out the defects of participants' performance in the performance process, and improve the lack of students' music expression [3]. At the same time, this paper points out that neural network algorithm technology has a wide range of applications in user analysis of music teaching software development, such as mechanical fault detection, environmental safety monitoring, intelligent housing, mobile device monitoring, automatic navigation, intelligent robot application, animal protection, etc [4]. Automatic provision of voice signals and analysis and processing in these areas can improve people's work efficiency, and also reduce the cost of application when dealing with complex voice hypothesis tasks [5–6]. In addition, this paper also applies user analysis in music software development. With the continuous deepening of information technology, the relationship between people and the Internet is becoming more and more inseparable, and people will leave a variety of information on the Internet. Nowadays, the competition in the Internet market is becoming more and more fierce [7]. If you want to make a profit, you should think about how to continue to attract customers and improve the viscosity of users, which makes the analysis of users more important. These analyses are not simple and can be quickly determined, but need huge user data to support optimization and decision-making. Therefore, this paper attempts to analyze the user content of music teaching courses to promote a more detailed understanding of music teaching software [8].

The literature points out that the method of extracting audio features through machine learning algorithm is mainly through neural network, using MFCC spectrum and relevant original audio as the training set, designing neural network, conducting network training, studying the mapping relationship between real audio and sound features in network training and learning, and carrying out research to link the original audio characteristics and the signal generated by the sound, so as to enter the time domain of the sound signal [9]. In order to achieve better research results, these programs usually require less training packages and less computers as resources for research, but fundamentally, this method is to re learn by manually extracting the neural network of sound characteristics. Literature By studying the artificial art market, some interviewees will be able to see their previous evaluation of the music they listen to, while others will draw conclusions about the inequality and unpredictability of music appreciation and current popularity [10–11]. Good music will not be evaluated badly, and bad music will not be evaluated particularly well, but in other cases, any result may occur. The literature shows that a person's personal characteristics can be distinguished by the diversity of musical tastes, while others' personal characteristics can also affect a person's musical preferences [12]. People are usually affected by the surrounding information, which is the mechanism behind the formation of personal music listening habits and habits. The type of music platform selected by music users is more affected by the recommended music platform. In the research of music retrieval, some researchers investigated two main factors that affect music suggestions, which are the differences and similarities in music [13]. When measuring the diversity of music tastes, we also take into account the styles of different songs, and measure these characteristics through the common consumption behavior of users. The literature points out that in music teaching, the teacher is the teacher, and the recipient is the student [14]. The relationship between teachers and students needs to be well handled. In the process of China's music education reform, China is paying more and more attention to the dominant position of students. The reform of music education emphasizes the need to update teaching methods and teaching means in order to better cultivate students' innovation ability in the future [15].

3.1 Neural network algorithm

The basic design and algorithm of convolutional neural network are also evolving through continuous development and application. Later, on this basis, through the constant exploration of algorithm researchers, a multilayer CNN neural network was innovatively proposed, such as LeNet-5. The network can recognize and distinguish the image content of handwritten numerals.

The neurons are interconnected in a way similar to the animal visual cortex tissue. The visual obstacle in the acceptable area of different neuron interactions can cover all visual areas that can be tested. Suppose the entrance variable neti of the i-th neuron in the input layer:

$${net}_{i}=\sum _{i=1}^{M}{x}_{i}+{\theta }_{i}$$

Ai is located at the threshold of i neurons. The corresponding algorithm is shown in Formula (2):

$${a}_{i}=f\left({net}_{i}\right)$$

In the process of neural network learning, nonlinear characteristics are mainly studied through hidden layer and output layer. Generally:

$${a}_{i}={x}_{i}$$

The input net_j including the seventh neuron is:

$${net}_{j}=\sum _{j=1}^{M}{x}_{i}+{\theta }_{j}$$

In the following formula, w_ij is the weight of the hidden layer and the range of the jth neuron. The corresponding output is a_j:

$${a}_{j}=f\left({net}_{j}\right)$$

The input network net_k of k neurons in the output layer is:

$${net}_{k}=\sum _{k=1}^{L}{w}_{jk}{a}_{j}+{\theta }_{k}$$

The output results can be summarized as follows:

$$Output=\frac{{map}_{size}-{kenerl}_{size}+2*padding}{stride}+1$$

The feature of maximum pooling is the maximum surface area reserved in each Fig. 2 * 2, and the average pooling is the average pixel of feature 2 * 2 reserved in each value. As shown in Fig. 1, the average number is obtained by keeping the average pixels in each attribute map. Although some channel information may lead to the loss of information about the value, there are no additional standard parameters during the pooling operation. In general, attribute loss has a small impact.

The neural network is composed of a multi-level complex function, which requires an activation function to ensure the nonlinearity of the network, because no matter how the linear function changes, its result is still a linear function. The existence of stimulation active layer is to activate neurons. Its primary role is to map the input of neurons to the output, which makes the feature expression of the model more detailed. In the early stage, neural networks are mainly used for Sigmaid as the operation function and output expression.

$$\sigma \left(x\right)=\frac{1}{1+{e}^{-x}}$$

Similar to Sigmaid, when the input is large or small, the gradient of Tanh is 0, which will cause slow network update. This is the output expression of Tanh:

$$\text{tanh}\left(x\right)=\frac{1-{e}^{-2x}}{1+{e}^{-2x}}$$

ReLU function is nonlinear near 0, and the decline rate is 0 or 1, so it can also be considered nonlinear. The output of ReLU is:

$$ReLU\left(x\right)=\text{m}\text{a}\text{x}(0,x)$$

When x ༞ 0, the read input is equal to the output. When x ≤ 0, the reread output is 0. Its conclusion is that the ReLU activation function will turn off the calculation of neurons when the input is less than or equal to 0, thus reducing the interaction between networks and reducing the weight of the model.

RNN has also been constantly studied through algorithms in the fields of text translation, natural language processing, computer visualization, speech recognition and a large number of applications, which has greatly improved the network threshold of its application scenarios. At this time, the output of t at RNN is shown in Formula (11) (12):

$${a}^{t}={g}_{1}({W}_{aa}{a}^{t-1}+{W}_{ax}{x}^{t}+{b}_{a})$$

$${y}^{t}={g}_{2}({W}_{ya}{a}^{t}+{b}_{y})$$

It should be noted that the weight matrix W * * and threshold b * are shared at each time point, and the deep RNN network structure is obtained by combining, connecting and stacking multiple single-layer RNN structures.

Due to the problem of gradient disappearance, RNN can only be effectively memorized and processed to obtain relatively close information. It is impossible to effectively memorize remote information, which makes RNN training extremely difficult. The LSTM network is proposed to improve the LSTM network based on RNN. Different from the structure of RNN cycle layer foundation, LSTM uses three different structures designed at different times to control the model weight and output of the network related layer.

As the deep neural network uses network parameters for training and updating, for the deeper network layer, finally, the estimated loss value derived many times in the regression diffusion process is very low, which seriously affects the training process of the model. So that the GAN network can design more in-depth structure, there is also the problem of gradient loss. If the discriminator device model can accurately identify the real sample, then the gradient in the generator device model becomes a serious problem. The best discriminator is shown in Eq. (13):

$${D}_{G}^{*}=\frac{{p}_{data}\left(x\right)}{{p}_{data}\left(x\right)+{p}_{g}\left(x\right)}$$

The original GAN model uses KL divergence and JS divergence to quantify the difference between the two data, as shown in Formula (14) (15):

$$C\left(G\right)=-lb4+KL\left({p}_{data}\right||\frac{{p}_{data}+{p}_{g}}{2}+KL({p}_{g}\left|\right|\frac{{p}_{data}+{p}_{g}}{2}\left)\right)$$

$$C\left(G\right)=-lb4+2JS\left({p}_{data}\right|\left|{p}_{g}\right)$$

Including:

$$KL\left({p}_{1}\right|\left|{p}_{2}\right)={E}_{x-{p}_{1}}\left(lb\frac{{p}_{1}}{{p}_{2}}\right)$$

$$JS\left({p}_{1}\right|\left|{p}_{2}\right)=\frac{1}{2}KL\left({p}_{1}\right||\frac{{p}_{1}+{p}_{2}}{2}+\frac{1}{2}KL({p}_{2}\left|\right|\frac{{p}_{1}+{p}_{2}}{2})$$

3.2 Music extraction technology

Four pianos are used to identify the characteristics of the selected music. At the same time, the four pianos are divided into different playing forms and playing intensity, and include different audio circuits. Music information is collected in the form of time waves or non sound waves through Fourier transform.

The construction process is to revise the formula, revise the formula and extract the waveform. The waveform revision depends on the pitch and the fundamental frequency depends on the frequency. Therefore, it is necessary to revise the extracted formula again, as shown in (18).

$$S\left(w\right)=\left\{\begin{array}{c}A\frac{{a}_{1}}{{a}_{1}^{2}+{(w-{w}_{l})}^{2}}, {w}_{l}-{\alpha }_{1}\le w\le {w}_{l}+{\alpha }_{1}\\ A\frac{2}{|w-{w}_{l}|} , others\end{array}\right.$$

(18) Where: the corrected vibration coordinate value is S (w), the parameter value of the waveform curve near the adjusted waveform peak value is a1, A is the amplitude, and the specific audio or frequency multiplier refers to wt. For different sounds, they have different characteristics in time domain and frequency domain; In addition, the real audio data with the sampling rate of 44.1 KHz has a large amount of data, and direct feeding into the convolutional neural network requires a large amount of computing power and higher demand for computing power.

4.1 User Requirements Analysis

The first step in designing a music teaching program should be the analysis of users. Most of the users of music teaching software are students. With the development of computer technology, teachers have experienced convenience from online teaching, and students have benefited greatly from online learning.

4.2 Implementation platform and development environment

The speech recognition software in the classroom runs on the raspberry pie 4B. The raspberry pie is essentially a small integrated computer device based on the Linux operating system. It can play different functions by running different software at a lower cost. Raspberry pie can be used to edit files, browse pages, play online games, videos and videos. Through software design and expansion of equipment hardware, Raspberry Pie can also be used to make smart cars, electronic photo frames, cameras, etc. according to specific situations and personal needs.

Computer programs collect, recognize and enhance various environmental sounds, such as animal sounds, human sounds, mechanical sounds, environmental sounds, and so on, through sound collection devices (such as deployed microphones). As shown in Fig. 2, on the basis of successful voice recognition, sound processing is improved and environmental noise is eliminated. Figure 2 describes the situation of each unit in detail.

The evaluation subjects of music teaching focus on students, whose main task is to participate in education evaluation and ensure the successful results of education evaluation.

These contents include the following two points: ① Teacher's teaching situation: production of classroom content before vocal music course, vocal music teaching methods and skills, such as audio demonstration, pronunciation training, post course training, etc. ② Teacher quality: basic music knowledge, ideology and morality, and basic music skills.

4.3 Software testing

The system measures the pitch accuracy by comparing the listening results with the spectrum of the original singing model. The method to achieve this goal is: first, extract the short-term frame from the signal to be matched, then extract the basic frequency parameters by inverting the spectrum, and finally use the dynamic time warping (DTW) method to check the data, and calculate the distance between vectors twice in succession to determine the best result. DTW method is used to calculate the similarity, as shown in Fig. 3.

Table 1 shows the average distance between parameters and performance conditions. As shown in Table 1, the evaluation system is sensitive to the freedom of sound and breath, and it is known that the scoring system is not sensitive to the perception of volume. On this basis, the frequency trajectory is very important, followed by the smoothness of breath, and then the curve of sound intensity.

Table 1

Etude Scoring Results
characteristic parameter	intensity of sound	pitch	Breath stability	Final score	error
Machine rating	1.32	77.38	87.27	74.82	2.384%
Manual scoring	4.89	83.11	84.99	77.02	2.384%

The challenges posed by the information society in all fields of music education are quite obvious. The introduction of advanced science and technology, the use of innovative methods and vocal music teaching methods are the basic trend of discipline improvement and development. This paper proposes a voice teaching method based on deep learning, which applies multimedia technology to traditional voice teaching. This method goes beyond the traditional music education. It can provide students with an interactive teaching platform and self-learning platform, while providing skills guidance and correction functions, changing the monotony of voice teaching content, and greatly improving the awareness and effectiveness of the teaching process, interactivity and learning ability. It is not only novel, but also has high practical value.

In addition to audio testing, instrument testing is also required. After getting the same score, three performers were selected to play a piece of music on the piano at the same time. The data shown in Table 2 shows the score performed by the three performers after completion.

Table 2

Score of three players
group	Performance score 1	Performance score 2	Performance score 3	Performance score 4	Performance score 5	Average score
teacher	0.96	0.98	0.97	0.95	0.94	0.96
Student 1	0.87	0.90	0.87	0.89	0.87	0.88
Student 2	0.77	0.77	0.72	0.79	0.73	0.75

It can be seen from Table 2 that, from an objective point of view, the average score of a single performance or multiple performances, from high to low, is teacher and student 1 and student 2 respectively, which can reflect the actual level of performers. Therefore, the system has learned some lessons in alleviating the lack of music resources and promoting students' autonomous learning.

4.4 Validity and fit test of music teaching software

Efficiency is effectiveness, which indicates whether the measurement obtained effectively reflects the content of the investigation. The higher the efficiency, the more consistent the measurement results are with the matching height of the inspection content; The lower the efficiency, the worse the measurement results do not meet the matching height of the inspection content.

This paper classifies the project and studies a large amount of data, so it has good content validity. Next, it is effective in analyzing construct validity. Factor analysis is a common method to verify the validity of the form or questionnaire structure. Before analyzing various factors, a quantum experiment and a sub experiment are carried out first. KMO test is used to verify the correlation and deviation between 0–1; When the volume is close to 1; The greater the relationship between variables and the less the deviation, the better the analysis of factors. Therefore, KMO and B artlet spherical test results are shown in Table 3.

Table 3

KMO and B artlet spherical test
Kaiser Mean Olkin measurement of sampling adequacy		0.938
Bartlett's Sphericity Test	Approximate chi square	107574.87079909974
	Df	760
	Sig.	0

Indicates that the platform built is valid. The model fitting test measures the consistency between the theoretical model and the actual data. In this paper, NC, GFI and RMSEA3 absolute fitting indexes and NCI, CFI and TLI value-added fitting indexes are used to evaluate the model fitting. After reviewing the parameter meaning of the model, it is found that the corresponding effect between the theoretical model and the actual data is not good, and the model can be appropriately modified to improve the model fit. Model correction is to add or remove some parameters from the model.

Model correction usually adopts two methods: (1) Eliminate paths that are not important to parameters. In Amos, the absolute value of C.R. lower than 1.96 is considered insignificant, and deleting these insignificant paths is considered to improve the model fitting.

From the perspective of parameter estimation, the external reward of attitude expectation, the social factors of willingness to share and the convenience of sharing behavior did not pass the significance test, so they were removed from the model. Then, referring to the model correction index, the maximum M.I. value is obtained based on the theoretical significance. A new path coefficient is established step by step to improve the fitting of the model. The corrected model fitting indicators are shown in Table 4.

Table 4

Corrected fit test results
Fit index	Ideal value	actual value
NC	1 < NC < 3	2.433
GFI	GIF > 0.8	0.993
RMSEA	RMSEA < 0.06	0.034
NFI	NFI > 0.8	0.846
CFI	CFI > 0.9	0.937
TLI(NNFI)	TLI(NNFI) > 0.8	0.943

After the model is modified, the fitting indexes all conform to the standards, and the model is well fitted. Once modified, the proposed general index will conform to the standard and have a good style.

4.5 User experience feedback

Finally, from the perspective of user analysis.

Experience personnel: 20 students. There are 5 men and 10 women each time. The senior group, the junior group and the gender are divided into 4 groups.

Experimental results: The data collection and analysis results in the form of questionnaires and SAM scales are good, as shown in Table 5.

Table 5

Analysis of experience data of music teaching system
index	high quality	good	commonly	Poor
Immersion experience	97%	3%	0%	0%
Intelligent interaction	95%	5%	0%	0%
Fluency	98%	2%	0%	0%
Module perfection	87%	10%	3%	0%

It can be seen from Table 5 that the music teaching system has a very high standard in terms of intelligent participation and fluency, but the module integrity of the teaching software is still insufficient, and the software functions are also less, so we should pay attention to this part in the future, and the system basically meets the goal of the music teaching system. SAM uses the Emotional Self Reporting Scale to obtain highly simulated music interaction content, and conducts correlation analysis before and after in-depth learning intervention, as shown in Fig. 4.

This paper points out that the transfer of information technology in online music education is still an important concern in music teaching. Therefore, this paper designs a technology application software for music teaching, which aims to use the neural network information system to enhance students' interest in music lessons. Music creation software not only replaces teachers' teaching work to a certain extent, It also applies to professionals related to music creation. The experiments in this paper mainly summarize the following points: (1) According to the internal logic of computer neural network, the detailed network algorithm flow of the software system is designed and applied to the construction of the system. (2) Music feature extraction is a basic component of the system. This file converts the performer's voice information into a waveform that can be recognized by the computer, and corrects the image according to the transition feature. Therefore, the results show that the corrected music feature is corresponding to the performer's original music feature. (3) The system structure and key modules were designed and studied, and part of the program code was provided for this purpose. Finally, students were selected to carefully evaluate the effect of using the music software. After analyzing the effect of using the music application software, the results showed that the system was conducive to improving students' interest in learning music and the subsequent development of teachers' teaching.

Compliance with Ethical Standards

Conflict of interest

The authors declare that they have no conflict of interests

Ethical approval

This article does not contain any studies with human participants performed by any of the authors.

Data Availability

Data will be made available on request.

K. Aune, “Feminist spirituality as lived religion: How UK feminists forge religio-spiritual lives,” Gender & Society, vol. 29, no. 1, pp. 122-145, 2015.
E. Georgii-Hemming, & M. Westvall, “Teaching music in our time: student music teachers' reflections on music education, teacher education and becoming a teacher,” Music education research, vol. 12, no. 4, pp. 353-367, 2010.
K. R. Fitzpatrick, “A mixed methods portrait of urban instrumental music teaching,” Journal of Research in Music Education, vol. 59, no. 3, pp. 229-256, 2011.
A. D. Almási, S. Woźniak, V. Cristea, et al., “Review of advances in neural networks: Neural design technology stack,” Neurocomputing, vol. 174, pp. 31-41, 2016.
Z. Wang, H. Li, & X. Zhang, “Construction waste recycling robot for nails and screws: Computer vision technology and neural network approach,” Automation in Construction, vol. 97, pp. 220-228, 2019.
X. Xu, D. Cao, Y. Zhou, & J. Gao, “Application of neural network algorithm in fault diagnosis of mechanical intelligence,” Mechanical Systems and Signal Processing, vol. 141, p. 106625, 2020.
J. K. Corley, Z. Jourdan, & W. R. Ingram, “Internet marketing: a content analysis of the research,” Electronic Markets, vol. 23, no. 3, pp. 177-204, 2013.
J. A. Borges, & D. Tome, “Teaching music to blind children: new strategies for teaching through interactive use of Musibraille software,” Procedia Computer Science, vol. 27, pp. 19-27, 2014.
T. Kim, J. W. Kim, & K. Lee, “Detection of sleep disordered breathing severity using acoustic biomarker and machine learning techniques,” Biomedical engineering online, vol. 17, no. 1, pp. 1-19, 2018.
F. Khan, I. Tarimer, H. S. Alwageed, et al., “Effect of Feature Selection on the Accuracy of Music Popularity Classification Using Machine Learning Algorithms,” Electronics, vol. 11, no. 21, p. 3518, 2022.
N. Mahmoodian, A. Schaufler, A. Pashazadeh, et al., “Proximal detection of guide wire perforation using feature extraction from bispectral audio signal analysis combined with machine learning,” Computers in biology and medicine, vol. 107, pp. 10-17, 2019.
G. Nave, J. Minxha, D. M. Greenberg, et al., “Musical preferences predict personality: evidence from active listening and facebook likes,” Psychological science, vol. 29, no. 7, pp. 1145-1158, 2018.
M. Müller, A. Arzt, S. Balke, et al., “Cross-modal music retrieval and applications: An overview of key methodologies,” IEEE Signal Processing Magazine, vol. 36, no. 1, pp. 52-62, 2018.
K. R. Fitzpatrick, “A mixed methods portrait of urban instrumental music teaching,” Journal of Research in Music Education, vol. 59, no. 3, pp. 229-256, 2011.
K. Salvador, & V. Pasiali, “Intersections between music education and music therapy: Education reform, arts education, exceptionality, and policy at the local level,” Arts Education Policy Review, vol. 118, no. 2, pp. 93-103, 2017.

Download PDF

Journal Publication

published 12 Jun, 2023

Read the published version in Soft Computing →

Reviewers agreed at journal
03 Apr, 2023
Editor assigned by journal
25 Mar, 2023
First submitted to journal
22 Mar, 2023

You are reading this latest preprint version

Development of music teaching software based on neural network algorithm and user analysis

Status:

Journal Publication

Version 1

Abstract

Figures

1 Introduction

2 Related Work

3 Neural Network Algorithm And Music Extraction Technology

3.1 Neural network algorithm

3.2 Music extraction technology

4 Development Of Music Teaching Software

4.1 User Requirements Analysis

4.2 Implementation platform and development environment

4.3 Software testing

4.4 Validity and fit test of music teaching software

4.5 User experience feedback

5 Conclusion

Declarations

References

Status:

Journal Publication

Version 1