1 Introduction

With the gradual commercialization of 5G mobile communication networks, new hot spots are gradually being replaced by the development prospects, capacity requirements and key technologies of the next-generation mobile communication technology 6G wireless transmission (Yuan et al. 2020). First, we will summarize the projects that 6G wireless transmission may participate in in the future, such as the new spectrum communication, MIMIO distributed collaboration, and the main technical direction of intelligent communication. The focus is on the integrated air-ground network (SCIN) based on deep satellite-earth integration. Based on two possible typical network topologies, different types of transmission links are analyzed and summarized (Kato et al. 2019). Finally, it analyzes and forecasts the key technologies that are in urgent need of breakthrough, such as optical multi-user access, efficient satellite-to-ground laser communication, and future innovative integrated optoelectronic networks, aiming to chart the direction for the next step of related research. Voice network analysis is part of the computer network protocol, which is the first to send human words through the communication network. The research of voice network analysis comes from the research project of the IAEA “Network Security Communication Project” (Hu et al. 2017). The goal of this project is to develop the feasibility of realizing secure, high-quality, low-bandwidth, real-time duplex digital voice communication through packet-switched computer network content. The main purpose of this research is to demonstrate high-quality, low-bandwidth, and encryption-enabled digital voice processing capabilities to meet the general needs of global encrypted voice communications (Xiao et al. 2020). In oral English teaching, attention should be paid to the fluency of spoken symbols, and real language materials should be selected from original novels, films and other literary works, and various phenomena and manifestations of the markup language should be analyzed, such as hesitation, suggestion, conversion, coherence, etc. Language features. In teaching, with the help of online language systems, many real language environments have been created (Li 2018). Students use oral English to organize various forms of communication, such as role playing, discussion topics, and simulated dubbing to improve students’ language skills and the ability to speak English in conversations (Cook et al. 2017). To further master the skills of language use. Teachers can collect students' written records and dialogues, summarize the characteristics of use, and form a voice prompt corpus to help improve the effectiveness of teaching. The main characteristics of the MOOC teaching system course are: (1) The scale of online learning is relatively large, and the number of students studying in this course is relatively large. (2) Diversified resources: MOOC teaching system courses integrate various types of digital resources and social network tools, which are richer and more diverse than traditional teaching. (3) Ease of use: The development of "online courses" teaching business can break the time and space requirements of traditional teaching, and at the same time enable people from all countries and regions in the world to learn courses they are interested in through network connections.

2 Related work

The literature introduces the development trend of mobile communication, and 5G has raised its technical level to an unprecedented level (Chen et al. 2018). The academia and the entire industry are focusing on whether the future 6G mobile communications has the potential to further improve basic technical indicators such as spectrum performance. This article attempts to answer this question systematically from the perspective of the structure of Shannon's information theory and its extended form (Wang et al. 2022). The results show that by increasing the number of transmitting and receiving antennas, introducing a new architecture, and finding ways to balance the effective balance block length (waiting time), error probability, and the minimum number of antennas, the development of core technologies is improved. In theory, there is no end to the improvement of system performance (Yuan et al. 2020). The literature introduces Shannon's information theory, Shannon’s information theory has many related research branches, and its results are huge (Amiri et al. 2021). Due to space constraints, this article will focus on the basic structure of the classic Shannon information theory and its extension to MIMO. On this basis, focusing on the improvement of frequency spectrum and power efficiency, the deepening of reliability and the lower waiting time and higher frequency range, theoretically discussed future 6G technical capability improvement methods. The literature introduces that online education has obvious advantages compared with previous teaching methods (Palvia et al. 2018). Students only need a computer connected to the Internet to study, saving round trip time and costs. Secondly, there is no need to rent classrooms, which reduces teaching costs and gives educational institutions more choices in the handling of educational funds. Finally, there are no geographical restrictions, and educational institutions can hire more foreign teachers. Many native English-speaking teachers can teach students from other countries online at home. Online English teaching is more forward-looking than traditional teaching (Zhang 2021). The literature introduces the educational system developed for the field of IELTS test teaching, which provides independent online media servers for voice and video services (Estaji and Ghiasvand 2019). There is no need to install third-party tools when using the system for teaching. The system has higher performance and stronger concurrency to make it smoother, which greatly improves the operation of the system and the user experience. The system is divided into three modules according to user roles: student module, teacher module and administrator module (Galindo-Dominguez 2021). Different roles correspond to different modules to perform their respective functions. The literature introduces the deep integration of space and earth networks based on satellites and the ground. According to the two possible network replication architectures of 6G, the characteristics of high-speed communication and technological development requirements as well as the speed of three different types of links are deeply summarized (Zhou et al. 2021). Then, it analyzes the key technologies that 6G needs to break through, such as multi-user optical phase matrix access, high-performance satellite-to-ground laser communication, and optoelectronic integrated networks.

3 Voice network analysis and oral English teaching model design for 6G wireless transmission technology

3.1 6G wireless transmission technology

When the transmission speed is not large compared to the capacity, there is an encryption method, and when the code length tends to be infinite, the possibility of transmission errors tends to zero. Shannon proved in this article that the capacity of the C channel is as follows:

$$C=\underset{x}{max}\{H(x)-H(x\mid y)\}$$
(1)

The expression of the noise channel (AWGN) capacity:

$$C=B\cdot {\mathrm{log}}_{2}\left(1+\frac{P}{{\sigma }^{2}}\right)\triangleq B\cdot {\mathrm{log}}_{2}(1+\mathrm{SNR})$$
(2)

Shannon’s information theory is not structured, and it is impossible to design a communication system with maximum power (or transmission rate). Shannon’s student Gallager proposed a finite length random code theory, which advanced Shannon’s theory. For discrete binary memory (B-DMC) channels, the Chernof limit can prove that if the code length is n and a randomly selected code is used, the average error probability P has upper and lower limits:

$${P}_{e}<{2}^{-n({R}_{0}-R)}\text{ or }{P}_{e}<{2}^{-n{E}_{r}(R)}$$
(3)

R is the encoding speed Ro is the cut-off speed

$${R}_{0}\triangleq -\log_{2}\int\limits_{{\mathcal{C}}^{M}}{\left(\frac{1}{|\mathcal{M}|}\sum_{x\in {\mathcal{M}}}\sqrt{p(y\mid x)}\right)}^{2}{\text{d}}y$$
(4)

The maximum cut-off rate is the main problem of channel capacity research. Many innovations in the construction of communication systems have been carried out simultaneously with this research. Academia once believed that after determining the capacity of the C channel, Ro remains unchanged. People’s understanding of Ro has changed with Pinsker’s research on plug-ins. When the channel coding is divided into two levels of internal and external connections, the channel capacity remains unchanged for C, but R will gradually increase. As shown in Fig. 1

Fig. 1
figure 1

Cut-off rate Ro and Gallager index Er(R)

Full size image

The core problem in the channel capacity approximation is to maximize the cutoff rate R. Many innovations in the construction of communication systems have been carried out simultaneously with this research. Academia once believed that after determining the capacity of the C channel, Ro remains unchanged. People's understanding of Ro has changed with Pinsker's research on plug-ins. As shown in Fig. 2, when the encoding of the channel is divided into internal and external codes, the capacitance of the C channel remains constant, while Ro will gradually increase.

Fig. 2
figure 2

a Channel combination and b split diagram

Full size image

The virtual channel shown in the figure is based on the above two conditional probabilities, which proves that C(W) < C(w) < C(W), and the equivalent cutoff rate satisfies:

$${\overline{R}}_{0}({W}_{2})\triangleq \frac{1}{2}[R({W}_{2}^{-})+R({W}_{2}^{+})]>{R}_{0}(W)$$
(5)

For the AWGN channel, Shannon starts from the error index shown in Eq. (3), and after detailed derivation, obtains the approximate expression of the best error probability P when the block length is n and the coding speed is R. In the following statement, this article rewrites its expression in the following form:

$$R\approx C-\frac{1}{\sqrt{n}}\sqrt{\frac{2+\mathrm{SNR}}{2\cdot \mathrm{SNR}(1+\mathrm{SNR})}}{\Phi }^{-1}({P}_{e}^{\mathrm{opt}})$$
(6)

Polyanskiy et al. studied the feedback problem of progressive channel capacity under the condition of finite block length from a broader perspective. For DMC channels without shared discrete memory, the progressive capacity is summarized as follows:

$$R=C-\sqrt{\frac{V}{n}} {\Phi }^{-1}({P}_{e})+o(n)$$
(7)

In summary, for a given error probability Pe and coding rate R, the required channel coding block length should be greater than the following estimated value:

$${n}^{*}\approx V\cdot {\left[\frac{{\Phi }^{-1}({P}_{e})}{C-R}\right]}^{2}$$
(8)

The above results are relatively accurate only when n > 100. The approximate term o(n) in formula (7) can be approximately ignored.

The development of multi-antenna channel capacity theory originated from the initial research work of Foschini and others. Someone gave a more complete MIMO channel capacity analysis method and laid a theoretical foundation for the development of contemporary mobile communication systems with higher spectrum efficiency and larger capacity.

Assuming that the MIMO wireless link has Nt transmitting antennas and Nr receiving antennas, the wireless link model can be generally described as:

$$y=Hx+n$$
(9)

It is easy to calculate the mutual information about the transmitted signal X and the received signal y, that is, the channel capacity as follows:

$$C_{{{\text{MIMO}}}} \left( {\varvec{H}} \right) = B \cdot {\text{log}}_{2} \left[ {{\text{det}}\left( {{\varvec{I}} + \frac{1}{{\sigma^{2} }}{\varvec{HQ}}_{x} {\varvec{H}}^{\dag } } \right)} \right]$$
(10)

Using the Lagrange multiplier method, the optimal power allocation value of the transmitted signal x can be obtained as follows:

$$p_{i} = \left\lceil {\mu - \frac{{\sigma^{2} }}{{\left| {\lambda_{i} } \right|^{2} }}} \right\rceil^{ + } ,\left| {\lambda_{i} } \right| > 0,i = 1, \ldots ,k,k \le N$$
(11)

And the maximum available channel capacity is:

$$\underset{\mu }{max}\sum_{i}{({\mathrm{log}}_{2}(\mu |{\lambda }_{i}|))}^{+}\text{s.t}.\sum_{i}\left(\mu -\frac{1}{|{\lambda }_{i}|}\right)=P$$
(12)

In order to compare with the AWGN channel capacity originally given by Shannon, here it is assumed that the unknown channel state information at the transmitter can only be sent with equal power, and each element of channel H is an independent and identically distributed random variable with a variance of 1, and N = min, {Nt, Nr} is sufficiently large, then use the law of large numbers. This can be approximated by:

$${C}_{\mathrm{MIMO}}({\varvec{H}})\approx B\cdot N{\mathrm{log}}_{2}(1+\mathrm{SNR})$$
(13)

First, the eigen mode wireless transmission method with MIMO channel capacity approximation capability is introduced. For this reason, the diagonal matrix composed of the eigenvalues λi of the channel H subjected to SVD decomposition, namely Λ = diag[λ1 λk, 0 0], and thus the formula is rewritten as:

$$Vy=\Lambda Ux+Vn\text{ or }\widetilde{y}=\Lambda \widetilde{x}+\widetilde{n}$$
(14)

To transform the problem into a solvable convex optimization problem, one of the simpler methods is to use Jensen's inequality to obtain the compact upper bound of its ergodic capacity:

$$\begin{aligned} E_{{\varvec{H}}} \left\{ {C_{{{\text{MIMO}}}} \left( {\varvec{H}} \right)} \right\} & \triangleq B \cdot E_{{\varvec{H}}} \left\{ {{\text{log}}_{2} \left[ {{\text{det}}\left( {{\varvec{I}} + \frac{1}{{\sigma^{2} }}{\varvec{HQ}}_{x} {\varvec{H}}^{\dag } } \right)} \right]} \right\} \\ & \le B \cdot {\text{log}}_{2} \left[ {{\text{det}}\left( {{\varvec{I}} + \frac{1}{{\sigma^{2} }}{\varvec{Q}}_{x} E_{{\varvec{H}}} \left\{ {{\varvec{H}}^{\dag } {\varvec{H}}} \right\}} \right)} \right] \\ \end{aligned}$$
(15)

MIMO channels can have both spatial multiplexing gain and multi-antenna diversity gain, and a compromise between the two can be found. This means that MIMO systems can seek a certain balance between transmission rate and reliability, and increase the importance of improving wireless transmission. Reliability is extremely effective. Furthermore, if Nt antennas are used to repeatedly send the same information, and Nr antennas are used for combined reception, the bit error rate of the resulting MIMO system can be approximated as a measure of the diversity characteristics of the multi-antenna system, and the general diversity of the system is given. The definition of sex is as follows:

$$d=-\underset{\mathrm{SNR}\to \infty }{lim}\frac{{\mathrm{log}}_{2}{P}_{e}(\mathrm{SNR})}{{\mathrm{log}}_{2}\mathrm{SNR}}$$
(16)

This means that the spatial multiplexing gain of the MIMO channel is N. In a general sense, the spatial multiplexing gain of the MIMO channel is measured. The spatial multiplexing gain is defined as follows:

$$r=\underset{\mathrm{SNR}\to \infty }{lim}\frac{R(\mathrm{SNR})}{{\mathrm{log}}_{2}\mathrm{SNR}}$$
(17)

Based on the classic Shannon information theory, MIMO extended form and performance compromise theoretical framework, this section discusses the theoretical approaches and potentials to further improve the core technical indicators of the 6G mobile communication system, including higher spectrum efficiency and power efficiency, and higher Reliability and lower delay, higher frequency band and capacity, etc.

The multi-user MIMO (MU-MIMO) used in the 5G mobile communication system can be further extended to multi-base station and multi-user joint processing situations, thereby forming a multi-point-to-multipoint (MP-2-MP) form of cell-free Mobile communication system, and thereby further improve the spectrum utilization rate of the entire system. Figure 3 shows a comparison between a traditional cellular system and a non-cellular system. The following description is given as an example of the above uplink. For non-cellular systems, due to the introduction of multi-cell joint processing, multiple users form MP-2-MP distributed MU-MIMO in the coverage of multiple cells, and all users and base stations can work at the same frequency at the same time; MIMO channel capacity model is still applicable, the difference is that the multiple antennas on the base station side come from different cells. The difference is that traditional cellular systems do not have multi-cell joint processing capabilities, and therefore cannot obtain MU-MIMO spatial multiplexing gains.

Fig. 3
figure 3

Traditional cellular system architecture (a)

Full size image

Because each user occupies different frequency resources, the total spectrum utilization of M users is equivalent to a single user. For MU-MIMO, all M users are superimposed on the same frequency band and communicate with a base station at the same time; and under ideal conditions, the communication rate of each user is statistically the same as the FDMA user; therefore, the MU of M users -The total spectrum utilization of MIMO is M times that of FDMA. The above examples illustrate that for MU-MIMO, the spectrum utilization rate needs to be comprehensively evaluated at the multi-user cell level (rather than the single user level). (As shown in Fig. 4).

Fig. 4
figure 4

Comparison of non-cellular system architecture

Full size image

Building a low-energy, environment-friendly mobile communication network will be an important research direction for future 6G development. Due to space limitations, this article only discusses the effective use of the total transmit power of one side of the base station without a cellular system. Assuming that the total number of transmitting antennas on the base station side is N (infinitely many), the downlink M users are all configured with a single antenna, and the positions are in accordance with random independent and identical distribution:

$${P}_{\mathrm{BS}}\approx \frac{({2}^{\eta /M}-1)}{\overline{\mathrm{SNR}}}\approx \frac{\eta }{\overline{\mathrm{SNR}}\cdot M},N>M$$
(18)

For the sake of simplicity, the following discussion takes a linear uniform array (ULA) with equally spaced array elements as an example. Assuming that the transmitter and receiver of a point-to-point wireless link are equipped with ULA with antenna array elements of Nr and Nt, and the steering vector is used at the transmitter and receiver respectively, the Nt × Nr-dimensional equivalent channel formed is:

$${\varvec{H}} = \sqrt {N_{r} N_{t} } \beta {\varvec{u}}_{r} \left( {\theta_{r} } \right){\varvec{u}}_{t}^{\dag } \left( {\theta_{t} } \right)$$
(19)

When there are i = 1, 2,…, L reflection paths in the channel, Eq. (19) evolves into the following beam-domain channel model or finite-dimensional MIMO channel model:

$${\varvec{H}} = \sqrt {\frac{{N_{r} N_{t} }}{L}} \mathop \sum \limits_{i = 1}^{L} \beta_{i} {\varvec{u}}_{r} \left( {\theta_{r,i} } \right){\varvec{u}}_{t}^{\dag } \left( {\theta_{t,i} } \right) \triangleq {\varvec{U}}_{r} {{\varvec{\Lambda}}}{\varvec{U}}_{t}^{\dag } ,$$
(20)

Assuming equal power distribution for the transmitting end antenna, the tight upper bound of the ergodic capacity of formula (20) can be easily obtained as follows:

$$E\left\{ {C_{{{\text{MIMO}}}} \left( {\varvec{H}} \right)} \right\} \le B \cdot L \cdot {\text{log}}_{2} \left( {1 + {\text{SNR}}\frac{{N_{r} E\{ |\beta |^{2} \} }}{L}} \right)$$
(21)

3.2 Voice network analysis

Since the correlation of the coding rate sequence of the speech frame is relatively complicated, if the method of manually extracting features is adopted, it may lead to incomplete features or high feature dimensions. In view of the fact that deep learning can automatically extract the deep features of the data, this chapter uses deep learning to learn the two correlations we propose, and then, the learned features are spliced and sent to the connection layer for final calculation and classification. The entire deep learning architecture Contains 3 modules, namely LSTM module, CNN module and classification module.

LSTM is a special network unit, LSTM has the function of mapping all the previously input information to the output of each time step. Each LSTM unit contains input gates, output gates and forget gates.

The input gate of LSTM determines how much information of the input Xt of the current time step is stored in the current storage unit Ct, which can be expressed as:

$${I}_{t}=\sigma ({W}_{i}\cdot [{h}_{t-1},{x}_{t}]+{b}_{i})$$
(22)

The forget gate of LSTM determines how much of the previous state Ct−1 and the previous information to be saved in the current storage unit Ct can be expressed as:

$${F}_{t}=\sigma ({W}_{f}\cdot [{h}_{t-1},{x}_{t}]+{b}_{f})$$
(23)

The output gate controls Ot output state and ht hidden state, expressed as:

$${O}_{t}=\sigma \left({W}_{o}\cdot \left[{h}_{t-1},{x}_{t}\right]+{b}_{o}\right)$$
(24)

Among them, Wo, bo are the weight and bias of the output gate. The storage unit is used to store and control long-term information, and its calculation formula is:

$${C}_{t}={F}_{t}\cdot {C}_{t-1}+{I}_{t}\cdot \mathrm{tanh}({W}_{c}\cdot [{h}_{t-1},{x}_{t}]+{b}_{c})$$
(25)

The hidden layer output of the LSTM unit at time t is:

$${h}_{i}={O}_{i}\cdot \mathrm{tanh}({C}_{i})$$
(26)

In order to achieve better training and detection results, we use batch standardization to process the same batch of data. Define the activation function used by the CNN module as a rectified linear unit, expressed as:

$$f(x)=max(0,x)$$
(27)

Finally, we will reduce the feature dimension through the global pooling layer through the output of the three convolutional layers. Before inputting the learned correlation into the classification module, it is necessary to connect the learning first, and then use it as the input connection layer. In the first two layers, the ReLU activation function. In the last layer, we apply the softmax function to predict whether the carrier contains secret information. The softmax expression is:

$${y}_{i}=\frac{{e}^{{x}_{i}}}{\sum_{j=1}^{2}{e}^{{x}_{i}}}$$
(28)

In order to comprehensively evaluate the performance of different steganography detection schemes, we collected 6000 speech data with a length of 10 s, including four types of speech, namely male Chinese, male English, female Chinese, and female English speech. The ITU G.723.1 codec is used to encode all speech. The encoder has two encoding rates. In all experiments, the high rate (6.3 kbps) is expressed as binary number 1, and the low rate (5.3 kbps) is expressed as binary number 0. In addition, the initial encoding rate defaults to a high rate. In the experiment, three latest rate adaptation algorithms (represented as RAAI, RAAII, RAIII) are applied to generate simulated original coding rate samples. Therefore, the experiment is divided into three categories, where BS, ABSI, and ABS II are used to generate steganographic samples with packet lengths of 16, 32, 64, 128, and 256, respectively. (As shown in Fig. 5).

Fig. 5
figure 5

Based on the RAA I method to generate the detection results of the original carrier

Full size image

From Fig. 6 we can draw the following conclusions: (1) The detection results of the steganalysis scheme method in different situations, we can find that the original rate steganalysis performance generated for RAAI is the best, and the original rate steganalysis generated for RAAIII has the best performance. The write analysis performance is the worst, because the original carrier rate distribution of different generation algorithms is different; (2) In the case of using RAAI to generate the original rate, the two steganalysis methods both achieve the best detection performance.

Fig. 6
figure 6

Based on the RAAII method to generate the detection result of the original carrier

Full size image

In the case of using RAA II and RAA III to generate the original rate, we can find that the RCN method is better than the BIP method. It is easy to find that for the length of each group, the detection performance of the two methods decreases as the group length increases, because the embedding capacity decreases with the increase of the speech frame group length. However, the RCN method is better than the BIP method at any packet length. Under 10 s speech length, when the correct rate of the BIP method drops to 90% when the packet length is 32, the RCN method can still maintain a correct rate of more than 90% when the packet length is 64. (As shown in Fig. 7).

Fig. 7
figure 7

Based on the RAA III method to generate the detection results of the original carrier

Full size image

4 Voice network analysis for 6G wireless transmission technology and practical application of oral English teaching

4.1 Demand analysis of oral English teaching system

Business requirements represent the high-level goals of the organization or the client. According to repeated communication with managers and obtained copywriting requirements, the business requirements of the system are defined as follows:

An Educational Technology Co., Ltd. has a large number of foreign teachers. The company hopes to integrate overseas teaching resources by establishing an online teaching platform. After investigating similar websites in the market, I decided to increase investment in IELTS education, and won some market shares with stronger teachers and more considerate services.

With the development and comprehensive modernization of China's economy, the number of people studying abroad has increased year by year, and the IELTS testing institutions have grown simultaneously. However, many institutions are small in scale and cannot meet market demand. The emergence of online video teaching provides a solution to this problem. However, the online IELTS teaching platform on the market is very low, and there are huge profits to be tapped in the IELTS teaching field.

Business goal: Within one year after the website is launched, the number of subscribers will reach 6000, and the number of paying users will reach 1500, and more than 80% of users are satisfied. Success criteria: occupy 50% of the market share.

The existing online English teaching websites in the market provide courses with insufficient pertinence, and a large number of users who want to learn the IELTS test cannot find a suitable method for learning. In this market, the market needs a video for teaching English for the IELTS test system.

Operation means risk. The website needs to accurately grasp and analyze various relevant data in real time and present it in the form of reports for management decision-making.

4.2 Functional module design

The system is divided into three modules, namely student module, teacher module and background management module. Each module has its own division.

The student module is divided into seven functional modules, as shown in Fig. 8. The login module functions include registration, login and password recovery. The online test module can be tested online. The course management module can book, view and cancel courses. The video teaching module can be used for classroom input, evaluation and viewing. The account management module can view and change personal information. The consumption module can be used for shopping and ordering courses. The service module can be used for customer service and resource download.

Fig. 8
figure 8

Student module design drawing

Full size image

The teacher module is divided into five functional modules as shown in Fig. 9. The login module can log in and password reply. The teacher teaching module can view courses and student information. The history record module can view the history record. The account management module can manage personal information. The service module can be used for customer service and resource download.

Fig. 9
figure 9

Teacher module design drawing

Full size image

4.3 Database design

The database used by the system is non-relational MongoDB, and a total of 10 database tables (management panel, advertisement table, activity view table, course table, system log, package table, report card, message board, student bulletin board and teacher bulletin board are set up) Course schedule is the database table of the main operations of the system, and the main system operations and business operations are executed according to the schedule of the program.

The processing program table stores the information of all processing programs. When the administrator uses a password to log in, they are divided into different roles and have different permissions at the same time. Its structure is shown in Table 1.

Table 1 Administrator table
Full size table

The advertisement information is stored in the advertisement table, and the advertisement is divided into two states: published and unpublished. Its structure is shown in Table 2.

Table 2 Advertising sheet
Full size table

The active view table stores information about certain active images on the page. The image name is stored in the database, while the actual image is stored in a fixed directory on the server. Its structure is shown in Table 3.

Table 3 Active view table
Full size table

The course table is the most important of all tables, it stores the information of each class. Initially, the school administrator sets the date and time for each teacher. When the student makes an appointment, the student enters the course type, course name, syllabus and notes. At the end of the course, teachers and students will conduct after-class assessments separately, and teachers can also leave feedback and approvals for students’ homework. Its structure is shown in Table 4.

Table 4 Class schedule
Full size table

The operations performed by each ID on the system are stored in the system log table. Its structure is shown in Table 5.

Table 5 System log table
Full size table

The student table is used for data storage of website user information. The student table identifies the user's identity through the user's email. The nickname is the name of the student that the teacher sees in class, usually in English. The student table not only stores basic student information, but also stores student account information. When a student purchases a course on the website, the available information about the course will be stored in the student table. Its structure is shown in Table 6.

Table 6 Student table
Full size table

The teacher table is used to save the account information of the teacher. The teacher account is designated and activated by the administrator. The number of classrooms refers to the number of online virtual classrooms that teachers have in the classroom. At the same time, the teacher's name, age, interest, Chinese level and other information are also stored in the teacher table for students to refer to and choose. Its structure is shown in Table 7.

Table 7 Classroom table
Full size table

The system architecture design shows that it is designed according to the MTV architecture pattern, that is, the presentation layer, logic layer and data layer of the system are controlled through the Model-Template-View pattern. The database design explains the database structure of the system, and at the same time lists 7 database tables related to the main functions of the system.

4.4 System test

From the perspective of the application system, testing represents functional testing and performance testing. Functional testing is a test of every function required by the developed application system; Web application testing has its own characteristics, in addition to the difference in performance and functionality of the test, it also comes from aspects such as user interface and security. The test content of the Web application usually includes: user interface, function, interface, compatibility, power and security test.

  1. (1)

    User interface test: Referred to as UI test, it tests whether customers are satisfied with the interface style, whether the content is perfect, whether the internal text is incomplete, whether the combination of text and pictures is perfect, and so on.

  2. (2)

    Functional test: To test whether the function of the software system is operating normally, the most important factor of the software is its correctness and stability, so functional testing is indispensable.

  3. (3)

    Database test: In the Web application technology, the database plays an important role, and the database provides space for the management, operation, query and realization of the data storage requirements of the application system. Two types of errors usually occur in Web application systems that use databases, namely, data consistency errors and output errors. Data consistency errors are mainly caused by users submitting incorrect form information, while output errors are mainly caused by network speed. For two kinds of errors, you can test separately.

  4. (4)

    Safety test: Check whether the system can be operated correctly, reliably and safely.

  5. (5)

    Robustness test: Whether the software system can be restored normally in the case of system errors.

  6. (6)

    Performance test: Use automated testing tools to simulate various normal, peak and irregular load conditions to check system performance indicators to determine the system performance in the workload, check when the load increases, and observe changes in system performance indicators.

The results of the system test are as follows:

  1. (1)

    The site map, navigation bar, content, color/background, image, table, etc. meet customer requirements.

  2. (2)

    All links are valid.

  3. (3)

    All forms can be submitted correctly and processed correctly by the background processing program.

  4. (4)

    The system can correctly prompt the input of non-compliant data.

  5. (5)

    The data related to the database can be saved and read correctly.

  6. (6)

    Different roles log in to the system. The system only presents pages related to role permissions. Users whose usernames and passwords do not match are not allowed to log in. The system records all critical behaviors.

In summary, the system meets the expectations of the test and can meet the customer's requirements for interface, function, and performance. The system has passed the test.

This part analyzes the test results and analyzes the test process, and the product resource information provides a reference for the designation of future test plans. Realize whether the software quality evaluation complies with the detailed test performance evaluation and test plan to analyze the defects in the system and propose suggestions for correcting and stopping the failure.

Testing includes functional testing, robustness testing, performance testing and user interface testing. The function test mainly checks whether the system achieves the expected result. This test is a basic software test, and the main reference objects are owners, developers, testers, etc. Durability testing refers to the feasibility of testing procedures under abnormal and hazardous conditions. It refers to the characteristic that the control system maintains certain performance in certain parameter settings (structure, size). Performance testing mainly tests strict indicators, such as system response time, CPU utilization, memory usage and system responsiveness, and the simplicity and aesthetics of UI testing.

After the above tests, many problems were found. Through active modifications, the performance of all aspects of the system has been strengthened, and it has become a product that meets the requirements and can be released.

5 Conclusion

With the development of Internet technology, the technology used to hide information has attracted widespread attention in the academic community. At the same time, criminals can also use information hiding technology in cybercrime activities. Therefore, it is urgent to restrain the development of reverse technology of hidden technology in information technology analysis. Nowadays, online voice streaming has the characteristics of strong hiding ability and strong real-time performance, and has gradually become a search hotspot for hiding confidential information or communications. However, the current cryptanalysis technology used for network voice streaming is still in its infancy, and there are many problems that need to be resolved. This article starts with the problems and difficulties of the existing steganalysis technology, discusses the steganalysis methods of network voice streams, and recommends corresponding steganalysis methods from multiple angles. In this article, the steganalysis method will analyze the carrier format and masking capabilities of the network voice stream, and modify the chi-square detection of the classic analysis method in the figure to perform hidden analysis on the network voice stream. When performing steganalysis, the suspicious bits in the dialog box will be extracted according to the suspicious level, and then random numbers will be used to connect and expand to construct a detection parameter sequence similar to the gray value, and then use traditional chi-square to calculate the hidden information It can improve the detection judgment within an appropriate window length. The multi-speed adaptive speech coding analysis method combines statistical features and feature selection, making the research results applicable to a variety of different coding formats. In addition, the fixed codebook format of AMR encoding is different from conventional compressed speech encoding. Therefore, based on the research on the fixed encryption structure and encryption method of AMR encryption, this article proposes a new encryption function. At the same time, in order to reduce the computational consumption in the discovery process and prevent the classification program from over-matching, this paper uses the AdaBoost method to select the extracted features for training and testing. The test results show that, compared with the existing methods, this method has higher detection efficiency in various situations.