A Study on the Practical Application of Multimedia Technology in College English Listening and Speaking Classroom Teaching

A flipped classroom means that knowledge transfer is done before class with the assistance of multimedia technology, and knowledge internalization is done during the class with the help of the teacher and classmates. Flipped classroom teaching mode is very different from the traditional classroom teaching mode, which completely reverses the traditional teaching structure. Students are required to learn the teaching content by watching the teaching video before class, and then consolidate and internalize the knowledge learned before class through a series of teaching activities in class, which is a brand new teaching mode, and it is necessary to carry out more in-depth practical research on it to test the effectiveness of its specific application.

On the basis of the field questionnaire survey and the principle of evaluation index construction, through an in-depth theoretical analysis of the teaching content of college English flipped classroom and the cognitive law of English learning, a total of 12 evaluation indicators were selected: the rationality of teaching content, classroom atmosphere, teaching logic, teaching skills, teachers’ sense of responsibility, the seriousness of homework correction, the results of students’ phased tests, the number of after-class homework corrections, the number of after-class tutoring, the number of students who left early, the number of students who were late, and the number of lectures. The 12 evaluation indicators are composed of the evaluation index system of college English listening and speaking classroom teaching.

Since the standard sparrow search algorithm adopts the same position update formula, the individuals in the population are gradually aggregated during the iteration process, which is easy to fall into the local optimum and difficult to obtain a large search speed [27-28]. In order to enable each individual in the population to search and hunt in the most suitable way according to the population characteristics and to realize the diversity of the population, the sparrow search algorithm was improved to increase the accuracy of the algorithm.

The sinusoidal search strategy was introduced to address the differences in the advantages and disadvantages of the positions of the two types of individuals, searchers and followers. The improved weight formula for the sinusoidal search strategy is as follows: (1) w=wmin+(wmax−wmin)⋅{ sin[ (fi′−fbest′fworst′−fbest′+1)⋅π2+π ]+1 }

Where: w_min,w_max is the minimum and maximum value of weights, respectively. fit is the fitness value of the i th sparrow in the population in the t rd iteration. fbestt is the optimal fitness value of the population in t iterations. fworstl is the worst fitness value of the population in t iterations. The individual update formula is as follows: (2) Xi,jt+1={ Xi,j′⋅exp(−1a⋅it) R2<STXp′+w⋅Q⋅L R2≥ST (3) Xi,jt+1={ Q⋅exp(xworstt−xi,jti2)i>n/2Xpt+1+w⋅|Xi,jt−Xpt+1|⋅A+⋅Li≤n/2 (4) Xi,jt+1={ Xbest′+w⋅β⋅|Xi,j′−Xbest′|fi>fgXi,j′+w⋅K⋅|Xi,j′−Xbest′|(fi>fw)+εfi=fg

Where: t is the current iteration number. X_i,j is the individual position information of the i rd sparrow individual on the j th dimension, j = 1,2,⋯,d;a ∈ (0,1] is the random number. P is the follower, X_P is the position of the follower. Q is a normally distributed random number. L is a 1×d-dimensional matrix and each element is 1. R_z(R_z ∈ [0,1]) and ST(ST ∈ [0.5,1.0]) represent the alarm signal value and the preset safety threshold, respectively. ε is the fitness variance of the sparrow population.

Diversity variance processing with the help of the concept of aggregation in biology, the introduction of population aggregation index A to solve the algorithm in the iterative process is very easy to reduce the diversity of the population and trapped in the local optimal problem, the expression is shown in equation (5): (5) A=ε−x¯−2x

Where: x¯ is the mean value of sparrow population fitness.

In order to avoid aggregated states appearing at the beginning of the iteration, Cauchy’s variation is used for the population. The improvement of the global optimal solution is expressed as follows: (6) X=Xbest+XbestoCauchy(0,0.5)

Where: Cauchy is a Cauchy random number, taking values from 0 to 0.5.

Deep learning algorithms can provide bidirectional mapping of high-dimensional data space and low-dimensional nested structure, which can effectively solve the inverse mapping problem that most nonlinear dimensionality reduction methods don’t have, and provide a new idea for realizing the accurate assessment of the quality of college English listening and speaking teaching based on flipped classroom. Deep recurrent neural network (DRNN) is built on the basis of recurrent neural network, which is a variant of recurrent neural network, in order to enhance the expressive ability of the model, deep recurrent neural network is stacked multiple recurrent neural networks on top and bottom, set up multiple recurrent layers, and the outputs of the lower layer are used as inputs of the upper layer [29-30]. The structure of deep recurrent neural network is shown in Fig. 1. The traditional artificial neural network (DNN) is only made of a simple stack of multiple layers of BP neurons, with full connections between layers, but no connections between neurons in each layer, as shown in Fig. 1(a). Assuming that the DNN network contains N hidden layer, the input layer is counted as layer 0, and the output layer is layer N + 1, o^k(i) is the output of the i th neuron of the neural network in layer k(k = 0,1,…,N), M_k represents the number of neurons in the network in layer k, W^k(i,j) is the weight of neuron i in layer k as a percentage of the input of neuron j in layer k + 1, and b^k(i) is the bias factor of the i th neuron in layer k. ϕ represents the Sigmoid function or Tanh function, the specific choice of the function should be decided according to the DNN network usage scenario: (7) { sigmoid:y=1/(1+e−x)tanh:y=2/(1+e−2x)

The neural network input-output relationships for each layer of the DNN are shown below: (8) ok+1(i)=ϕ(∑j=1MkWk+1(i,j)ok(i)+bk+1(i)) (9) o0(i)=x(i)

x(i) : Neurons in the input layer.

Figure 1.

Comparison of structural principles between DNN and DRNN

In the forward propagation process of the DNN neural network, the input information is transmitted from the input layer to the output layer through the hidden layer layer-by-layer abstraction transformation, and the state of the neurons in each layer is only affected by the neurons in the upper layer. Backward propagation returns the error signal between the final output value of the network and the desired output value layer by layer along the original connection pathway, and the weights of the neurons in each layer are modified by repeated generation selection to minimize the error.

The DRNN network is improved to address the shortcomings of the traditional deep neural network DNN layer in which neurons are not connected to each other and have no tracking characteristics for time changes. The overall structure of the network is similar to that of DNN, which contains an input layer, an output layer and several hidden layers. Different from the BP neurons used in the hidden layers of DNN network, LSTM neurons are used in DRNN to keep the training error and realize the multi-scale transmission of the error along the time and layers. The principle of LSTM neuron structure is shown in Fig. 1(b). The tuple states s_t and h_l are introduced into the LSTM and the flow control of the temporal information is realized by adding o_l control gates: the forgetting gate, the input gate, and the output gate. The LSTM The final state passed to the next layer is 3. Similar to the traditional artificial neural network forward propagation process, the DRNN extracts training samples, inputs oto of a given sample into the network, and the information is transmitted to the output layer by layer-by-layer transformations and the error between the actual output and the known standard output of the training samples is calculated. When the information is transmitted between the hidden layers, the output of the previous hidden layer is used as the input of the next hidden layer. DRNN backpropagation contains the same error as the traditional BP neural network along the layer propagation in addition to the error along the time backpropagation (BPTT), but in practice, due to limitations on the amount of arithmetic can not be taken into account in front of an infinite number of time states, it is often used to truncate the BPTT.

In summary, although traditional artificial neural networks can be used to solve static nonlinear classification problems, it is difficult to track the characteristics of data in the time dimension. While the current moment output oiN+1 of DRNN neural network not only contains the influence of the current moment input x, but also takes into account the influence of the information perceived in the previous moment S_t–1 , the use of DRNN network realizes the quality of college English listening and speaking classroom teaching detection.

In order to effectively construct a university English teaching quality evaluation model, a total of 9,000 questionnaires from university B were collected as data sets in this study. They were randomly divided equally into three different data sets (F, G, H), with the first 2,000 copies of each data set as the training set and the last 1,000 copies as the testing set. For the collected data, due to the different data scale of the evaluation indexes of the same sample, it is easy to cause the network to crash if directly input into the ISSA-DRNN model, so it is necessary to normalize the raw data as follows: (10) yi'=yi−yminymax−ymin

where yi' is the normalized data, y_i is the original data, and y_max and y_min are the maximum and minimum values of the same index of the original data, respectively.

In order to compare the iteration speed and best performance of the two algorithms, ISSA and SSA, 2 test functions are selected for comparison. To ensure fairness, the parameters of the algorithms need to be set consistently. The expressions of the 2 high dimensional functions Ackley and Schwefel are as follows: (11) F1(x)=−20exp(−0.21m∑i=1mxi2)−exp[ 1m∑i=1mcos(2πxi) ]+20+e (12) F2(x)=∑i=1dxi2

The images of the 2 high-dimensional functions and the adaptation change curves of their 2 algorithms are shown in Figs. 2 and 3, where (a)~(b) are the Ackley function and the Schwefel function, respectively. From Fig. 2 and Fig. 3, it is obvious that ISSA has faster convergence speed and higher convergence accuracy, and also ISSA is greatly improved when dealing with the multi-peak function problem.

Figure 2.

High-dimensional function graph

Figure 3.

Fitness change curve

To verify the superiority of DRNN, the curves of loss value and accuracy between DRNN and DNN are shown in Fig. 4, where (a)~(b) are the loss value and accuracy, respectively. It can be seen that the loss value of DRNN is small, the accuracy is higher, and the evaluation performance is significantly better than that of the DNN model. This is because the DRNN model not only can greatly retain the effective features of the data and prevent the features from being wrongly rejected, but also can repeatedly extract the effective features, thus improving the performance of the model.

Figure 4.

Curve of loss value and accuracy between DRNN and DNN

In order to verify the superiority of ISSA-DRNN for the evaluation of English teaching quality, it was compared with three methods, namely DNN, DRNN and SSA-DRNN, and each of them was run independently for 10 times to take the average value. The average accuracy and standard deviation of the English teaching quality evaluation models based on different algorithms in dataset A are shown in Table 1, and the iterative training time consumed by different approaches is shown in Table 2. The average accuracy of the English teaching quality evaluation model is shown in Figure 5. Comparative analysis of Fig. 5, Table 1 and Table 2 yields the following conclusions: 1)

As can be seen from Fig. 5, the evaluation accuracy of the ISSA-DRNN-based university English teaching quality evaluation model is much higher than that of the other 3 models on datasets A, B and C.

In summary, the English evaluation model based on ISSA-DRNN can be used effectively for predicting and evaluating teaching, and it can provide guidance for the quality assessment of English flipped teaching using multimedia technology.

Figure 5.

Average accuracy of English teaching quality evaluation model

With the rapid development of multimedia technology, new teaching modes have gradually emerged into the English classroom. Flipped classroom accelerates the application of multimedia technology in English listening and speaking classroom teaching, students can combine the actual needs of personalized teaching, eliminating the limitations of the traditional teaching mode, and promoting the advancement of English teaching methods. Based on this study, this study takes college English flipped classroom as an example to analyze teaching practices and explore its application effect in current college English classroom teaching.

6.1.3

Data analysis

1)

Analysis of differences before intervention

SPSS14.0 statistical analysis software was used to analyze the data, and then to analyze whether there was any difference between the two groups of students’ English listening level (A1), basic knowledge level (A2), English reading level (A3), and total knowledge competence (A4) before the experiment, and the results of the pre-intervention variability analysis are shown in Fig. 6, where the lines indicate the upper 1/4 value, the upper 1/2 value, the median line, the lower 1/2 value, and lower 1/4 value. From the data analysis in the figure, it can be seen that the mean values of the pretest English listening knowledge, pretest English basic knowledge and pretest English reading knowledge of the experimental group and the control group are extremely close to each other, in which the mean values of the total pretest English knowledge competence are 2.38 and 2.47 respectively, which indicates that the level of the experimental group’s English knowledge competence and that of the control group’s English knowledge competence basically maintains at the same stage. The standard deviations of the pre-test English listening knowledge, pre-test English basic knowledge and pre-test English reading knowledge of the experimental group and the control group are also extremely close to each other, in which the standard deviations of the total pre-test English knowledge competence are 0.41 and 0.39, which shows that the internal dispersion of English scores of the two classes is roughly the same.

Figure 6.

Difference analysis before intervention

The independent sample test of the total pre-test of English knowledge ability of the experimental group and the control group is shown in Table 3, and according to the statistical analysis of the data, the probability of the significance of the variables in the Levene test of the variance equation is (sig) 0.754, which is greater than 0.05, so the variance of the scores in the results of this pre-test is equal. The equal variance data in the total pre-test English knowledge ability Sig.(2-tailed) is 0.837 which is greater than 0.05, so there is no significant difference between the experimental group and the control group’s English knowledge ability. The combined results of data analysis show that there is no significant difference between the total pretest English knowledge ability of the experimental group and the total pretest English knowledge ability of the control group.

Table 3.

Independent sample testing for total pretest

Index		Levene test of variance equation				Mean test
		F	Sig	T	Df	Sig(2-tailed)	Mean score different	Standard mean error
A1	Equality	0.108	0.734	-1.947	54	0.044	0.04	0.12
	Inequality			-1.947	65.937	0.044	0.04	0.12
A2	Equality	0.129	0.684	0.379	54	0.248	0.03	0.09
	Inequality			0.379	65.873	0.248	0.03	0.09
A3	Equality	3.147	0.758	2.764	54	0.009	0.02	0.06
	Inequality			2.764	63.839	0.009	0.02	0.06
A4	Equality	0.057	0.754	0.044	54	0.837	0.09	0.12
	Inequality			50.058	65.867	0.837	0.09	0.12

2)

Post-intervention differential analysis

Based on the principle of independent samples t-test, the difference analysis between the two groups was carried out, and the comparative analysis of the results of the two groups after intervention is shown in Figure 7. The average value of post-test English listening knowledge, post-test English basic knowledge and post-test English reading knowledge in the experimental group was higher than that of the control group, and the average value of the total post-test English knowledge ability in the experimental group was 3.49, and the average value of the total post-test English knowledge ability in the control group was 2.57.

Figure 7.

The results of the two groups after intervention were compared

The independent samples test of the total pre-test of English knowledge and ability of the experimental and control groups are shown in Table 4. The results of data analysis show that the probability of significance of the variables in the Levene’s test of the variance equation is (sig) 0.236, which is greater than 0.05. Therefore, the variance of the scores in the results of this pre-test is equal. The variance in the total posttest English knowledge proficiency Sig.(2-tailed) is less than 0.05, so there is a significance of statistical analysis in this between-group comparison, and there is a big difference in the level of English knowledge proficiency between the experimental group and the control group. Combining the results of the data, it can be seen that there is a significant difference between the improvement of the total posttest English knowledge competence level of the experimental group and the total posttest English knowledge competence level of the control group.

Table 4.

Independent sample detection

Index		Levene test of variance equation				Mean test
		F	Sig	T	Df	Sig(2-tailed)	Mean score different	Standard mean error
A1	Equality	3.817	0.415	-6.338	54	0.027	1.04	0.23
	Inequality			-6.338	52.433	0.027	1.04	0.23
A2	Equality	2.138	0.136	-3.728	54	0.019	0.99	0.18
	Inequality			-3.728	63.373	0.019	0.99	0.18
A3	Equality	6.431	0.024	4.016	54	0.014	0.87	0.26
	Inequality			-4.016	54.339	0.014	0.87	0.26
A4	Equality	0.733	0.236	-5.403	54	0.037	0.92	0.24
	Inequality			-5.403	62.389	0.037	0.92	0.24

After testing the superiority of the combination algorithm, the English listening and speaking teaching mode of the integrated flipped classroom is evaluated by combining the evaluation indexes constructed above, taking the data of the questionnaire filled in by the students as the real value and the evaluation value of the combination algorithm as the predicted value, and then comparing and analyzing the two, so as to reflect the current status of English listening and speaking teaching in the integrated flipped classroom, and the results of the assessment of the quality of teaching are shown in Table 5, in which the data ranges from 1, 2, 3, 4, 5, corresponding to poor, poor, fair, good and excellent, respectively. As can be seen from the data in the table, the true value is distributed between 3 and 4.5, which indicates that the quality of its teaching mode is good. In addition, the error of both is less than 0.5%, which means that the evaluation model in this paper can realize accurate assessment of teaching quality, play an important guiding role in intelligent monitoring of teaching quality, and easily improve the level and quality of college English listening and speaking classroom teaching.