Analysis of Multimedia Teaching Path of Popular Music Based on Multiple Intelligence Teaching Mode

Abstract

Pop music multimedia is one of the popular digital pop music types. Based on the multiple intelligences teaching model, a multimedia multiple intelligences teaching method of popular music is proposed. This method not only analyzes the characteristics of pop music in detail, but also fully considers other important characteristics of pop music. It teaches college students multimedia, purifies their hearts, improves their personality, cultivates their innovative consciousness, and promotes their healthy growth. In this paper, the multiple intelligences teaching model is introduced into the process of pop music multimedia teaching path. In the stage of music audio segmentation, the deep belief network algorithm is used to accurately carry out music multimedia teaching. Finally, the experimental analysis results show that the integration of pop factors into music teaching, the combination of pop music and quality education, and the creation of music that is suitable for students’ personality characteristics and the needs of aesthetic development can better serve the needs of students’ quality improvement.

This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

1. Introduction

With the continuous development of Internet technology and the improvement of technological progress, different Internet-based popular music multimedia began to emerge [1]. As one of the more popular types of digital pop music, the teaching and recognition of multiple intelligences of popular music multimedia have become the focus of domestic scholars’ research. However, how to obtain the specific content source from the initial popular music multimedia data lacking the definition of popular music content has become a huge challenge for the current popular music multimedia multi-intelligence teaching, because the popular music multimedia signal belongs to a way of chronological order, the multiple intelligences teaching mode can be used according to its invisibility. At present, other classification methods are relatively simple, and it is not accurate to obtain the characteristics of popular music multimedia.

The multiple intelligence teaching model is applied to the automatic classification process of popular music multimedia in this paper. This method can use the lyrics, word frequency, content, and meaning of popular music multimedia as current prior knowledge in the automatic classification process of popular music multimedia according to the features of automatic lyrics of popular music multimedia. The information gain method is used to obtain popular music characteristics of popular music multimedia content; this mode is used to limit the weight of lyrics and the semantic information of popular music, integrate lyrics with high meaning similarity, and construct various types of popular music multimedia teaching models.

2. Multiple Intelligence Teaching Mode

When multiple intelligences teaching usually uses a given sequence of observations $O=o_1,o_2,\dots ,o_T$, a certain λ = (A, B, π) parameter is used to multimedia, ensuring that the value of P(O|λ) can reach the maximum value in this method.

(1) According to the related definitions of forward probability and backward probability, we know$$\begin{array}{}\text{(1)}& PO|\lambda ={\displaystyle \sum _{i=1}^N{\displaystyle \sum _{j=1}^N{\alpha }_{t}i{a}_{ij}{b}_j{o}_{t+1}{\beta }_{t+1}j}},\hspace{1em}1\le t\le T-1.\end{array}$$

  When P(O|λ) being used reaches the maximum value, the training sequence of each experiment is limited, so the best method of estimating parameters cannot be achieved. In this case, the Baum–Welch algorithm uses recursion thought with P(O|λ), the part is very large, and finally the model parameter λ = (A, B, π) is obtained.

(2) The revaluation formula of the Baum–Welch algorithm is derived by recursion as$$\begin{array}{c}\text{(2)}& {\overline{\pi }}_i={\xi }_{1}i,{\overline{a}}_{ij}=\frac{{\displaystyle {\sum }_{t=1}^{T-1}{\xi }_{t}i,j}}{{\displaystyle {\sum }_{t=1}^{T-1}{\xi }_{t}i}},\\ {\overline{b}}_{jk}=\frac{{\displaystyle {\sum }_{t=1\text{\hspace{0.17em}and\hspace{0.17em}}{O}_t=V_k}^T{\xi }_{t}j}}{{\displaystyle {\sum }_{t=1}^T{\xi }_{t}j}}.\end{array}$$

  Among them, ${\xi }_{t}i,j$ represents the given training sequence O and model parameters. Λ, the state of multiple v-chains at time t and the time t + 1 is the state probability, ${\theta }_i{\theta }_i{\displaystyle {\sum }_{t=1}^{T-1}{\xi }_{t}i}$ represents the expected value of the number of transitions from state ${\theta }_t$ to state ${\theta }_i$.

  Define the auxiliary function as$$\begin{array}{}\text{(3)}& Q\lambda ,\overline{\lambda }=\frac{1}{PO|\lambda }{\displaystyle \sum _{\text{all\hspace{0.17em}}S}PO,S|\lambda \ln PO,S|\overline{\lambda }}.\end{array}$$

(3) Among them, λ is the original model parameter, $\overline{\lambda }$ represents the model parameter to be solved $\overline{\lambda }=\overline{A},\overline{B},\overline{\pi }$, O represents the observation sequence used for training, $O=o_1,o_2,\dots ,o_T$, and S is a certain state sequence $S=q_1,q_2,\dots ,q_T$.

The multi-intelligence education model can not only be used to find a good enough state transition path, but also quickly calculate the output probability corresponding to the path. Meanwhile, the calculated amount required by the method of using the Markov model to calculate the output probability is much less than that in the total probability formula.

${\delta }_{t}i$ is defined as the maximum probability $o_1,o_2,\dots ,o_t$ produced by following a path $q_1,q_2,\dots ,q_t$ at time t and $q_1={\theta }_i$ which is ${\delta }_{t}i=\underset{q_1,q_2,\dots ,q_{t-1}}{\max }P{q}_1,q_2,\dots ,q_t,q_1={\theta }_i,o_1,o_2,\dots ,o_t|\lambda$.

The recursive form of the multiple intelligences teaching is as follows.

(1) Initialization:$$\begin{array}{c}\text{(4)}& {\delta }_{t}i={\pi }_i{b}_i{o}_1,\hspace{1em}1\le i\le N,{\psi }_{1}i=0.\end{array}$$

(2) Recursion:$$\begin{array}{c}\text{(5)}& {\delta }_{t}j=\underset{1\le i\le N}{\max }{\delta }_{t-1}i{a}_{ij}{b}_j{o}_t,\hspace{1em}2\le t\le T,1\le j\le N,{\psi }_{t}j=\underset{1\le i\le N}{\arg \max }{\delta }_{t-1}i{a}_{ij},\hspace{1em}2\le t\le T,1\le j\le N.\end{array}$$

(3) End:$$\begin{array}{c}\text{(6)}& P^{\ast }=\underset{1\le i\le N}{\max }{\delta }_{T}i,q_T^{\ast }=\underset{1\le i\le N}{\arg \max }{\delta }_{T}i.\end{array}$$

(4) Find the state sequence:$$\begin{array}{}\text{(7)}& q_t^{\ast }={\psi }_{t+1}{q}_{t+1}^{\ast },\hspace{1em}1\le t\le T-1.\end{array}$$

Among them, ${\delta }_{t}i$ represents the probability of accumulating the output value of the i-th state at time t, ${\psi }_{t}i$ represents the continuous state parameter of the i-th state at time t, $q_t^{\ast }$ is the state at time t in the optimal state sequence, and $P^{\ast }$ is the final output probability.

The multiple intelligence teaching model is a multilayered multiple intelligence teaching model that can automatically learn from above and below. This is the accumulation of multiple restricted Boltzmann machines (RBM). Its structure is shown in Figure 1, and every two adjacent layers constitute a restricted Boltzmann machine. From bottom to top, the hidden layer of the previous RBM is the visible layer of the next RBM. The output of the previous RBM is the input of the next RBM, so as to search to the last layer for each level.

[figure omitted; refer to PDF]

Through multimedia teaching and learning, the link weights and node offsets of each layer are obtained, and the network initialization is completed. The inverse conduction algorithm (BP) is used to fine-tune the deep trust network model monitored from top to bottom. Overcome the shortcomings of local optimization and long search time. Although the performance of the deep belief network model shows strong characteristic learning capabilities, from the above principles, Internet search requires a large amount of sample data to generate more parameter values [2, 3]. On the other hand, based on the popular music recommendation search problem, which lies in lack of a large amount of sample data, it is found that the generation of a large number of parameters takes a long time, which is not good for practical applications.

Different from traditional machine learning methods, the characteristic of the multimedia education model is the original time-domain characteristics of the input signal, which automatically learns the input signal through hierarchical multimedia education. The general time domain features are the short-term average amplitude difference, short-term energy, short-term autocorrelation coefficient, and so on. In this article, short-term energy is used as the original input characteristics of the deep multiple intelligence teaching model. The short-term energy can reflect the process of the energy of the sound signal changing over time. Unvoiced and voiced syllables can be clearly distinguished by short-term energy characteristics. Set the popular music signal as x(n); the average energy of that short time is as follows:$$\begin{array}{c}\text{(8)}& E_n={\displaystyle \sum _{m=-\infty }^{\infty }{xnwn-m}^2}={\displaystyle \sum _{m=-\infty }^{\infty }x{m}^{2}hn-m}\\ =x{m}^2\ast hn.\end{array}$$

However, the characteristic dimension of the original time domain is high, in which there is a lot of redundancy and noise. Therefore, the input search data requires low-dimensional processing. The method of Principal Component Analysis (PCA) in this paper investigates the correlation between multiple variables by means of multiple statistical methods. It is studied that the internal structure of multiple variables is clarified through several main components through several principal components. After the multimedia teaching data is processed by dimensionality reduction, it is input into the multiple intelligence teaching mode for multimedia teaching, automatically learning the characteristic information of the data, and finally processing the output data with the Softmax multimedia education path function commonly used in deep learning. Complete the task of multimedia teaching path. Specifically, the multimedia teaching path structure of popular music based on multiple intelligence teaching mode is shown in Figure 2.

[figure omitted; refer to PDF]

A multiple intelligence teaching model corresponding to the shortcomings of the original network model is proposed in this paper, which is different from the original multiple intelligence teaching model. As shown in Figure 3, the stacked and downsampling layers in the network with deep beliefs join the multiple intelligence teaching model together. The second layer of the multiple intelligence education model has 5 nodes. There are 11 parameters that require multimedia teaching when there are 3 nodes on the third layer using all connections. With quota sharing and sampling, there are 4 parameters that require multimedia teaching. The output of the convolutional layer acts as a hidden layer node of the next Boltzmann machine with restrictions.

[figure omitted; refer to PDF]

The multi-intelligence education model is a machine learning algorithm that effectively integrates relevant characteristics into an ordered model [14]. In the education of diversity and multiple intelligences, in addition to the relevance of popular music and search terms, the similarities in popular music must also be considered. A variety of intelligent education modes are adopted to solve the problem of diverse and multiple intelligent education in popular music multimedia. Through the mechanical learning method, it has produced a lot of influence on the relevance of popular music and search terms and the similarity of popular music and transformed it into the features for training and test. The features are combined in different ways, and the effectiveness of the features is verified through the ranking effect. Figure 5 shows the framework of diversity ranking learning in popular music multimedia.

[figure omitted; refer to PDF]

First, given a set of query words $Q=q_1,q_2,\dots ,q_n$, each query word q has a corresponding pop music set $T_i={\text{tweet}}_{i1},{\text{tweet}}_{i2},\dots ,{\text{tweet}}_{im}$. Each piece of popular music ${\text{tweet}}_{ij}$ is manually marked with whether it is related to the corresponding query term $q_i$ and its corresponding subtopic (query intent), and the standard answer ${\text{rank}}_i$ for sorting the popular music collection $T_i$ is generated through this marked information. Each query word q has a ${\text{tweet}}_{ij}$ corresponding to pop music collection $T_i={\text{tweet}}_{i1},{\text{tweet}}_{i2},\dots ,{\text{tweet}}_{im}$. A series of features used to measure the correlation between each piece of popular music in the popular music collection $T_i$ and the query word $q_i$ and measure the similarity between each piece of popular music in the popular music collection T are designed and extracted to form a feature vector, ${\text{vector}}_{ij}$. Assuming that the ranking function in the diversity ranking is f, the goal of diversity ranking learning is to obtain the optimal ranking function f through manually labeled training data.

For the new query q and its corresponding pop music set T, the same features are extracted to form a feature vector adopted, and the sorting function f is used to generate the final diversity and multiple intelligent education results. We represent the definition formula (1) of the ranking function in the diversity ranking learning.$$\begin{array}{}\text{(9)}& f{x}_j^i,r_j^i=w_r^T{x}_j^i+w_d^{T}h{r}_j^i.\end{array}$$

Among them, ${\text{vector}}_{ij}$ represents the correlation feature vector between the query $q_i$ and the corresponding popular music ${\text{tweet}}_{ij}$, $r_j^i$ represents the similarity feature matrix between popular music ${\text{tweet}}_{ij}$ and the popular music set ranked in front of it, and the function h transforms the similarity feature matrix into feature vector by averaging multiple values corresponding to each feature. $w_r^T$ and $w_d^T$, respectively, represent the weights of the correlation feature vector and the similarity feature vector.

The multiple intelligence teaching model is a new tool to solve the quality evaluation of popular music multimedia teaching by means of optimization methods and is especially suitable for the study of evaluation problems. Its essence is based on the multiple intelligence teaching model in popular music multimedia teaching path analysis, to establish a popular music multimedia teaching path analysis method using linear function hypothesis space.

Assuming the analysis of popular music multimedia teaching path, the popular music multimedia set is $x_i,y_i,i=\mathrm{1,2},\dots ,nx\in R^{d}y\in -1,+1$. In the $d$ dimensional space, the general form of popular music multimedia teaching path analysis function is $gx=w\cdot x+b$, and the specific classification surface equation is$$\begin{array}{}\text{(10)}& w\cdot x+b=0.\end{array}$$

To ensure that the classification surface of all popular music multimedia can be classified correctly, it must be satisfied:$$\begin{array}{}\text{(11)}& y_{i}w\cdot x+b-1\ge 0,\hspace{1em}i=\mathrm{1,2},\dots ,n.\end{array}$$

As mentioned above, if the above conditions are met, the smallest classification surface is the best classification surface. The matching of string similarity can convert the solution problem of optimal classification into the following constraint optimization problem; that is, the calculation function $w^2$ is satisfied under the restraint in formula (11).$$\begin{array}{c}\text{(12)}& \phi w=\frac{1}{2}{w}^2=\frac{1}{2}w\cdot w.\end{array}$$

The minimum value: this problem is further transformed into Lagrange function:$$\begin{array}{}\text{(13)}& Lw,b,a=\frac{1}{2}w\cdot w-{\displaystyle \sum _{i=1}^n{a}_i{y}_{i}w\cdot x_i+b-1}.\end{array}$$

The optimal classification function obtained by solving is$$\begin{array}{c}\text{(14)}& fx=\mathrm{sgn}w\ast \cdot x+b\ast =\mathrm{sgn}{\displaystyle \sum _{i=1}^n{a}_i^{\ast }{y}_i\cdot x_i\cdot x}+b\ast ,\end{array}$$where sgn () is a symbolic function. In the case of strict analysis of popular music multimedia teaching path in popular music multimedia, it can be processed according to formulas (10)∼(14); however, when popular music multimedia cannot conduct the analysis of popular music multimedia teaching path, slack variables need to be introduced; namely, the condition is satisfied by adding a slack variable ${\xi }_i\ge 0$ in (11), so that (10) can be used.$$\begin{array}{}\text{(15)}& y_{i}w\cdot x+b-1+{\xi }_i\ge 0,\hspace{1em}i=\mathrm{1,2},\dots ,n.\end{array}$$

The corresponding objective function is also transformed into the minimum value:$$\begin{array}{}\text{(16)}& \varphi w,\xi =\frac{1}{2}{w}^2-C{\displaystyle \sum _{i=1}^n{\xi }_i}.\end{array}$$

As a typical social media platform, popular music multimedia has its rich user information, which may help solve the problem of popular music multimedia education. Therefore, it is necessary to consider the characteristics of users among popular music publishers. The user characteristics of popular music performers include the geographic location of the user, whether the user is authenticated, the user language, the number of popular songs released by the user, the number of friends of the user, the number of followers of the user, and the number of times the user is grouped into other users.

4.5. Experiment and Result Analysis

26 songs and 5 kinds of popular music are selected in different styles on popular music websites, and a total of 200 pieces of popular music are provided for this paper. Perform preprocessing, feature extraction, multiple intelligence calculations, and angle cosine cluster analysis for the abovementioned popular music.

In the preprocessing stage, segmentation and subframe processing are performed first. For popular music, its local segment is similar to the entire popular music, and a popular music segment needs to be truncated with a segment length of about 3–5 minutes to reduce information redundancy and speed up the calculation. In this experiment, the initial popular music is divided into 15 segments, and 70% of the data of each segment is extracted as the signal processing data source.

In this paper, we choose to extract the time domain energy, frequency domain energy, spectral envelope, and MFCC features of the audio signal. If the triangle filter is selected, the center frequency is taken as M24. The spectral envelope is using formulas.$$\begin{array}{}\text{(17)}& fm,\hspace{1em}m=\mathrm{1,2},\dots ,M.\end{array}$$

The length produced by Hilbert transform is 4098. As shown in Figure 6, there are three different styles of popular music spectrum. The envelope shows a big difference in the spectral envelope diagram, especially in the low frequency part.

[figure omitted; refer to PDF]

In this experiment, the abovementioned feature vector and feature vector are integrated to perform Hausdorff feature dimension calculation. Through frame segmentation and window processing, the simplified data segment still has many features. Therefore, for each feature of popular music, the first 20% is a 2056-dimensional vector formed into a data set.

Before using sandwich cosine for clustering, it is necessary to regularize the obtained vector. For two popular pieces of music with similar styles, the vector sequence is not a corresponding frame, but only a table. Since the current frame indicates the time sequence in the first popular music, in order to improve the similarity of popular music, it is necessary to count the similarity count of each segment while ignoring the order of the feature vector.

The input vector is $X=x_1,x_2,\dots ,x_n$, $Y=y_1,y_2,\dots ,y_n$, the output result is $d_{\text{sum}}$, and the algorithm is described in Algorithm 1:

Algorithm 1: Baum–Welch algorithm..

  $d_{\text{sum}}$;

 for (i = 1: n) {

 $d_i=abs{x}_i-y_i$;

 min = i; for (=2: n) {

 $d_j=abs{x}_i-y_j$;

 if $d_j<d_i$;

 $d_i=d_j$;

 min = j;

 }

 }

 $d_{\text{sum}}=d_{\text{sum}}+d_i$;

 }

This algorithm describes the calculation of the distance between the element of the vector X and the element of Y and counts the number of minimum distances. Calculate the abovementioned distance of the sandwich cosine, and use the K-Manas method to cluster to obtain the analysis result. The results of multi-intelligence as sandwich cosine clustering are shown in Table 1. Correctly classify the four popular music styles, and display the percentage statistics in Table 2.

Table 1

Multi-intelligence clustering result.

Clustering cluster	Total popular music	Clustering result	Style of popular music,
			M1 (sentimental)	M2 (passionate)	M3 (quiet)	M4 (lonely)
Type 1	30	T	21	24	20	18
		F	1M3 4M4	0	2M1 2M3	3M1 3M3
Type 2	30	T	21	23	19	24
		F	3M4	1M1 2M3	5M1 2M4	2M3
Type 3	30	T	21	24	22	23
		F	2M2 3M3	2M4	1M1 2M3 1M4 2M1 1M3
Type 4	30	T	16	24	23	21
		F	2M2 3M3 5M4	2M4	2M1 1M3	2M1 3M4

Table 2

Statistics of the classification accuracy of the four popular music styles.

Style of popular music	Type 1	Type 2	Type 3	Type 4
	Accuracy (%)	Accuracy (%)	Accuracy (%)	Accuracy (%)
Sentimental	81	89	81	61
Passionate	100	89	93	93
Quiet	85	73	85	89
Lonely	77	93	89	81

From the experimental results, we know that the clustering effect of popular music is better than other styles. The other three kinds of popular music are obviously different from this. The other three popular music styles are somewhat similar. There are also sad types. Please be quiet. The result was classified incorrectly. However, overall, the clustering effect is ideal. The popular music multimedia teaching path function, as a popular music perception ability required for the development of artificial intelligence, can bring users a real entertainment experience. Based on the current problems of popular music multimedia teaching path, in-depth research is conducted on traditional machine learning methods, in combination with the popular music multimedia teaching capabilities of the multiple intelligence teaching model, optimizing the multiple intelligence teaching model’s role in the music multimedia teaching process, which requires more samples to find the optimal model parameters and solve the problem of the time it takes for the algorithm to run. Finally, the multiple intelligence teaching mode and traditional methods have a better effect on the time consuming of multimedia teaching.

5. Conclusions

In order to change the traditional educational concept, establish a new educational concept, innovate music multiple intelligences education methods, and improve the overall quality and ability of music teachers, this paper puts forward an analysis method of multimedia teaching path of popular music. By applying the multiple intelligences teaching mode to pop music multimedia teaching, this paper analyzes the actual situation of pop music teaching, summarizes the importance of integrating the quality education path in pop music teaching, discusses the ways of quality education path in pop music teaching from many aspects, better serves the needs of students’ quality improvement, and obtains the optimal parameters of HMM model. The accuracy of teaching rate is improved. Finally, the results of example analysis show that the introduction of multiple intelligences teaching mode in pop music multimedia teaching, combined with the melody characteristics of pop music multimedia, can improve the accuracy of pop music multimedia teaching. In the follow-up, we will take music as the carrier, rich music art content, colorful music expression forms, and related music culture as students’ experience, and explore the main objectives and main processes of learning.

Acknowledgments

This study was sponsored by XinXiang University.

References

[1] Y. Ma, "Research on the arrangement and visual design of aerobics under the new situation," International Core Journal of Engineering, vol. 5 no. 9, pp. 170-173, DOI: 10.6919/ICJE.201908_5(9).0026, 2019.

[2] Q. Q. Zhao, Q. Qi, "The application of computer multimedia technology in aerobics music orchestration," Applied Mechanics and Materials, vol. 651-653, pp. 1958-1961, DOI: 10.4028/www.scientific.net/amm.651-653.1958, 2014.

[3] J. Ledger, "Australian festival of chamber music wagner arrangement - conducted by j ledger," Journal of Community & Applied Social Psychology, vol. 4 no. 1, pp. 31-45, 2013.

[4] R. L. Turner, "Encyclopedia of appalachia: music2013237Encyclopedia of appalachia: music. Knoxville, TN: university of Tennessee press last visited february 2013. Gratis," Reference Reviews, vol. 27 no. 6, pp. 40-41, DOI: 10.1108/rr-02-2013-0048, 2013.

[5] N. Sato, "The factor of ga, ko, and zoku in the music of the ancient world: a case study upon an acculturation," Journal of Solid State Chemistry, vol. 228 no. 33, pp. 14-19, 2015.

[6] M. Fernandez-Santiago, "Of language and music: a neo-baroque, environmental approach to the human, infrahuman and superhuman in richard powers’ orfeo," Anglia, vol. 137 no. 1, pp. 126-146, DOI: 10.1515/ang-2019-0008, 2019.

[7] Julie, Ra, "Crossing the border into ‘classic music’ as a ‘pop musician’: the case of sting," Journal of the Musicological Society of Korea, vol. 16 no. 3, pp. 11-39, 2013.

[8] D. Yue, "Misreadings in the search for commonality between Chinese and western culture," Springer Singapore, vol. 13 no. 10, pp. 6545-6553, DOI: 10.1007/978-981-10-1116-0_28, 2016.

[9] X. Gu, Z. Y. Liu, B. Liu, B. E. Qi, S. Wang, "A cross cultural analysis of musical timbre perception:comparison between Chinese and western culture," Journal of Clinical Otorhinolaryngology, vol. 30 no. 20, pp. 1589-1592, 2016.

[10] K. Xue, M. Yu, "New media and Chinese society, communication, culture and change in Asia," Shaping Music Consumption in China’s New Media Eera: Uuse, Eexchange, and Iidentity, pp. 239-254, DOI: 10.1007/978-981-10-6710-5, 2017.

[11] S. Chithra, M. S. Sinith, A. Gayathri, "Music information retrieval for polyphonic signals using hidden Markov model," Procedia Computer Science, vol. 46, pp. 381-387, DOI: 10.1016/j.procs.2015.02.034, 2015.

[12] Y. Chen, "Automatic classification and analysis of music multimedia combined with hidden markov model," Advances in Multimedia, vol. 2021 no. 8,DOI: 10.1155/2021/7824001, 2021.

[13] C. C. Chen, D. J. Hong, S. C. Chen, Y. Y. Shih, Y. L. Chen, "Study of multimedia technology in posture training for the elderly," Engineering, vol. 5 no. 10, pp. 47-52, DOI: 10.4236/ENG.2013.510B010, 2016.

[14] B. Kostek, "Music information retrieval-The impact of technology, crowdsourcing, big data, and the cloud in art," Journal of the Acoustical Society of America, vol. 146 no. 4,DOI: 10.1121/1.5137234, 2019.