Design of Political Online Teaching Based on Artificial Speech Recognition and Deep Learning

With the emergence of the information age, computers have entered the homes of ordinary people and have become essential daily appliances for people. The integration of people and computers has become more popular and in-depth. Based on this situation, how to make computers and humans communicate intelligently and make human-computer interaction more convenient and practical is the main issue of scientific research and discussion on computers. Language recognition is widely used in industry, the economy, commerce, tourism, style, and sports. The BP neural network based on deep learning has supercomputing functions and is very suitable for speech recognition, which promotes the promotion and application of speech recognition technology in many fields. Different types of applications and diverse scenarios enable users to experience different personal experiences. This paper examines the use of application software for ideological and political theory courses in colleges and universities and the application of application software for ideological and political theory courses in colleges and universities in the management of ideological and political theory courses. Analyzed the use of apps in the management of ideological and political theory courses in colleges and universities, analyzed the possibilities and challenges that this brought to the management of ideological and political theory courses in colleges and universities based on the app, and put forward the necessary countermeasures. Including the development and design of ideological and political theory applications, combining deep learning and CTC algorithms to build acoustic models, using server-client interaction, designing an ideological and political theory course app based on speech recognition and deep learning, and forming an offline speech recognition system software platform, the app provides a certain reference for improving the skills of teaching managers.


Introduction
With the development of Internet technology and the unremitting efforts of manufacturers of mobile devices in various countries in the world, wireless mobile terminal devices have become very popular in people's daily work, production, and lives [1]. With the drastic reduction of communication costs, the rapid increase of network speeds, and the full coverage of wireless networks, all these increasingly perfect conditions have accelerated the rapid promotion and popularization of mobile terminals such as smartphones and tablet computers [2]. e proportion of Internet equipment used by everyone has been rapidly increasing and has gradually become the most important mobile terminal equipment used by people, gradually replacing the use of computers and notebooks in people's lives [3]. erefore, it is no exaggeration to say that smartphones have been closely linked with our work, production, and life and have become an indispensable item in our daily work and life, and even an essential thing in our daily lives. In the current period, with the deepening of research, the exploration of the teaching and management of ideological and political theory courses has gradually deepened. Various applications in smartphones are very convenient and simple, with high efficiency and low cost. ey can fully meet the requirements of students to learn anytime, anywhere, and are favored by students [4]. is is the development trend of learning methods in the future. At this stage, the number of mobile Internet users in my country is very large. Due to the rapid improvement of the educational information industry and the strong growth of user demand, apps as an application are indispensable in education and teaching [5]. It can not only promote the integration of educational theory and information technology but also strengthen the dissemination of educational ideas and scientific theories.
With the emergence of the information age, computers have entered the homes of ordinary people and have become essential daily appliances for people. e integration of people and computers has become more popular and indepth [6]. Based on this situation, how to make computers and humans communicate intelligently and make humancomputer interaction more convenient and practical is the main issue of scientific research and discussion on computers. Language recognition is widely used in industry, economy, commerce, tourism, style, and sports [7]. e BP neural network based on deep learning has super-computing functions and is very suitable for speech recognition, which promotes the promotion and application of speech recognition technology in many fields. After the emergence of new media, while breaking with the original advantages of the old media, the content and methods of teaching and management of ideological and political theory courses have been updated, which stimulated students' interest in learning, creativity, and teachers' enthusiasm [8]Better management efficiency provides new methods for ideological and political theory education. However, the application of apps is extremely rich and complex, which not only provides new methods for ideological and political education management but also brings great challenges [9]. e requirements for teachers and students are higher, and its freedom, diversity, and rapid dissemination also pose challenges and puts forward higher requirements for teachers' abilities. erefore, the need to combine some modern computer science and technology has become the focus of this article.

Related Work
For a long time, speech recognition has mainly focused on speech machines and natural language processing, as well as the sound signal itself. People obtain a lot of outside information through human-computer interaction. About 76% of people's daily communication is conducted through speech. Automatic language recognition technology is also called speech recognition technology, which is the technology responsible for converting speech into text [10]. is includes technology and knowledge of multiple disciplines. Automatic speech recognition technology performs speech recognition in human-computer interaction. Automatic speech recognition technology is committed to completing the smooth communication between humans and machines. e world's first speech recognition system was first researched and developed by American scientists. At first, it could only recognize simple numbers, but it provided new ideas for the development and research of automatic speech recognition technology. Compared with the original text interaction method, the method of human-computer interaction using speech recognition technology is more suitable for people's habit of communicating through language [11]. In addition, voice interaction will be smoother, more efficient, and have fewer errors. Voice recognition technology is mainly used in voice management systems, intelligent dialogue inquiry systems, dictation, transcription systems, and situational dialogue systems. In the beginning, researchers used pattern engineering methods to build a speech recognition system, but the results were minimal.
After the emergence of deep learning, speech recognition technology has gradually reached a modern level on a large scale [12]. is research study uses a deep neural network, and the research proves that, compared with previous research results, the recognition error rate has been significantly reduced by 40%. e literature records a deep-faith network that pre-trains deep neural networks layer by layer through a limited Boltzmann machine, which greatly reduces the difficulty of deep neural networks and significantly improves the modeling level of neural networks. e study of neural networks has triggered a wave of deep learning. In depth neural network and hidden Markov mixture model, depth neural network is used to replace Gaussian mixture model, and speech signal model is established. Great progress has been made in speech recognition effect. e reference records a sequence discrimination technique that makes deep neural network training directly determine the level of recognition frequency, greatly reducing the huge error in recognition and becoming a standard part of subsequent deep neural network training. Literature established a highly consistent speech-recognition neural network [13]. e literature accurately simulates the correlation between the different dimensions of the speech frequency spectrum function, shows a high level of antinoise, and greatly improves the noise reduction ability of the deep neural network. Researchers used long-term and shortterm memory models in speech recognition. People use long-term historical framework information to make more precise decisions and achieve better results than prospective neural networks. Researchers used short-term classification termination technology in speech recognition, which severely weakened the role of HMM (a technology that has been used for nearly half a century) and significantly improved the performance of continuous numbering and other items [14]. A codec-based speech recognition scheme is used. e input of this end-to-end model is the speech spectrum function, and the output is direct words. e original acoustic model and language model are interwoven into a deep neural network for common modeling, almost completely destroying the traditional speech recognition framework.
Digital recognition technology is one of the key technologies for mobile application development. e term "mobile learning" refers to a type of learning that can be performed at any time and any place using mobile computer equipment. Mobile computer equipment used for mobile learning requires a certain ability to display learning content and provide teachers with communication between students [15]. Technical features of the new system include intelligence and digitization, digital transformation and coding, and new functions beyond the media. e literature explains the important role of mobile media devices. From the birth of human beings to the birth of mobile phones, these two functions have been separated [16]. e development of science and technology has merged these two mutually independent parts. e record pointed out that the emergence and use of the Internet has improved the level of cultural thought and educational technology, and the idea of educational management has been added to cultural and technical learning to guide and inspire students. Media literacy education is no longer a violation of students. e education of media experience is no longer regarded as a means of discrimination or an observation of ideology, and pointed out that the future of media education will be different [17]. It has the characteristics of socialization. It can improve its own quality through traditional school education. Participate in media life to achieve social media education and improve students' social participation. e positive and negative effects of the Internet era on college education are used as a basis to study on how to improve and perfect college ideological and political theory education to cope with the changes of the times.

Deep Learning Model and Algorithm
Application. e original neural network is a decision model, and the deepfaith network is a hybrid model, which combines a generative model and a decision model, as shown in Figure 1. erefore, the network can realize the common probability arrangement of observation data and tags, which is convenient for estimating the former probability and the latter probability. In the deep trust network model, the middle hidden layer is mainly composed of limited Boltzmann machines. e neurons in the speech recognition system present a Gaussian distribution, so neuron V is a hidden neuron h. Under this condition, the energy function is (1) In the energy function, the conditional probability that can be achieved is In order for the contrast divergence algorithm to be quickly applied, the input data of visual data is usually normalized into multiple islands with an average value of 0 and a variance of 1. e energy function and conditional probability are changed to In the actual situation, the main considerations are the operating environment of the management system and the network environment.
is requires that the extracted function value must have good robustness, and tries to obtain the absorbed sound in different situations and environments corresponding to the function value. It is necessary to improve the original auto-encoder structure and add noise to generate noise reduction codes, as shown in Figure 2.
is will create a new hidden layer code for input data and information. When decoding, the decoder can estimate the initial information that has been interfered from the information without noise interference, and send most of the initial information to reduce the impact of these noises. After multiple in-depth training, the function value improves the robustness.
For noise reduction coding calculation, the general algorithm is: consider the initial input data x, add noise to the input data, and obtain new input data. When coding, the characteristic formula obtained is e output obtained during decoding is e final loss function of Ender is

Construction and Training of Acoustic Model Based on
Deep Learning. When the spectrogram function is decomposed, the Fourier transform is only performed on each frame signal within a short time after the frame is divided, and the spectrum is obtained through the window. en square the modulus spectrum to get the energy spectrum, and then take the logarithm to get the logarithmic energy spectrum of each frame. e logarithmic conversion makes the low-amplitude component higher than the high-amplitude component, so you can see the signal period hidden in the low-amplitude noise. Finally, the logarithmic spectrum calculated for each frame is rotated and decomposed in time to obtain the final characteristic sequence spectrum. Figure 3 is a calculation diagram of the parameters of the spectrogram function.
In the Fourier transformation, the transformation of a nonperiodic continuous time signal x(t) is Given a discrete signal of finite length x(n), n � 0,1, . . ., n − 1, the discrete Fourier transform is In the basic acoustic model of this article, the middle layer 3 adopts the pool layer module. Each convolutional pool layer module is two layers with a folding layer + 1 layer pool layer. e fully connected layer adopts two hidden layers, and the number of modules constructed is 1,425, and the size of the rear probability matrix of 200 × 1425 is transmitted via Softmax. e specific structural parameters are shown in Table 1.

Optimized Design of Deep Learning Speech Enhancement
Baseline System. e deep neural network has strong nonlinear adaptability and can obtain the nonlinear mapping relationship between noisy speech and speech. e deep neural network improvement algorithm includes two processes. e specific process is shown in Figure 4.
Before subjective evaluation, testers must first understand the scoring criteria and then give them more audit training so that the testers can evaluate the sound quality more accurately. Subjective and objective scoring uses the average sentence core scale method. e MOS score is mainly divided into five levels to evaluate the quality of the sound. e score reflects the process of voice quality. See Table 2 for details. e normalization method in this article is BN. It can converge quickly and improve the versatility of the network. Because the use of BN may interfere with the test and training data, the batch test data is different, which helps to improve the accuracy of the model. BN needs to learn these parameters when normalizing the c and β parameters, and transform and construct the normalized data to break the learned function distribution. Assuming that the input of the small batch data is B and the parameters to be learned are c, β, then the calculation of BN is: the average calculation formula: Variance calculation formula: Standardized calculation formula: Modification formula: Preweight, frame, add window

Voice input
The fast Fourier transform Take the absolute or square value (linear amplitude spectrum) Take the logarithm (logarithmic amplitude spectrum) Eigenvector Mosaic Output eigenvector e entrance of the previous layer is directly transmitted to the next layer. e expression output from each remaining module is

Experimental Results and Analysis.
As the speed and volume of the training set body change, the signal length of the training body changes accordingly. e method described in this paper is applied to the regeneration of the adjusted GMM-HMM system of the velocity distortion data and the adjusted GMM-HMM system of the volume adjustment data, for DFSMN acoustic model training based on the speech enhancement data. First, the experiment researched the influence of the speed distortion-based speech data amplification method on the acoustic DFSMN model recognition. e experimental conditions and results are shown in Table 3: Secondly, the experiment researched the effect of speech enhancement data on DFSMN acoustic model recognition; see Table 4 for details.
Based on the abovementioned experimental results, the discriminant training method based on the sMBR criterion, the treble adaptive method based on the discriminant vector function, and the voice data amplification method based on speed distortion are further combined. e experimental results of these three methods are shown in Table 5. A speech recognition system using DFSMN and TDNN acoustic models can recognize and transcribe experimental datasets without manual annotations. Among them, DFSMN chose the best model so far, and its vocabulary on the test set is 20.34%. e word error rate on the TDNN test set is 29.88%, so the DFSMN recognition result is regarded as a text label, and the word error rate of each voice recognition result in the dataset is calculated and determined by TDNN, as shown in Table 6. Table 6 shows that the distribution of word error rate in each time interval is more balanced. Because of the insertion problem in speech recognition, the vocabulary error rate of a   single voice may exceed 100%. erefore, we selected audio tapes with vocabulary frequencies of 0-10%, 10%-20%, and 20%-30% as the extended dataset, used the recognition results of these audio tapes on DFSMN as text annotations, and added them to the existing speech training package, as shown in Table 7.

Fine-Tuning and Optimization of Deep Learning Speech
System. When the RBM model is completed, a classifier is added on top of the pattern classification. Compared with logistic regression, only two types of nonlinear classification can be performed. e Softmax method is used to extend logistic regression to complete multiple classifications. ey are mutually exclusive, that is, two classifications cannot be occupied by a sample at the same time to meet the experimental requirements. e specific method is given below. In order to minimize the difference between the actual output and the expected output of the model, the fault uses cross-human correction.
Train network parameters i by minimizing failures.
Finally find the child thread: e partial derivatives of W and b are e process of using the gradient reduction method to update the weights is

(18)
Taking the time-distorted Mel cepstral coefficient and the first-order differential mixing parameter of Mel cepstral coefficient as input data, one can construct classification in the output for identification. If the amount of sample data is small and the number of hidden layers is large, the experimental effect is better, but the relative time spent increases sharply. e experiment process improves the network model according to the set rules and improves the learning efficiency of the model. RBM reconstruction performance is show in Figure 5

e Principle of Scalability.
With the rapid development of Internet technology and the advancement of society, students' requirements for learning platforms will also change accordingly. erefore, when developing and     designing a platform, not only must it meet the requirements of primary school students, but it must also leave opportunities for future development and improvement. is leaves room for development in subsequent content and function expansion.

Principle of Feasibility.
is principle needs to fundamentally complete the research and development of the main functions under the actual conditions of the current software and hardware and the operation of student mobile terminals.

Safety Principles.
First, when developing the platform, it is necessary to fully consider the information security situation, suspend user permissions, and protect user privacy and platform security. Second, the platform should try to avoid abusing users, and the interface should be concise and clear to reduce the number of abused users; the information prompt function of the operation should be set so that the user clearly knows the result of this operation.

Principle of Good
Interactivity. Good interactivity is very important for the development and design of this application. In this process, the student must be the center. First of all, we must meet the needs of students. We must use the simplest and most practical learning results to achieve the most optimized goal. e ideological and political education app combines the characteristics of the mobile learning environment, the design concept, and development principles of the app and is a new device for carrying out ideological and political education activities and a new way of mobile learning. e details are shown in Figure 6.

Database Design of Ideological and Political Education
App. Table 8 is a detailed description of various learning resource modules.
Sound performance, complete functions, and fast operation are closely related to the database. erefore, comprehensive consideration must be given to database design.
rough the analysis of the application of the ideological and political education app, this article points out that the units of the database design mainly include elementary school students, news information, topic discussions, etc. Table. e structure of the specific user information table is shown in Table 9. Table. e detailed content of the specific news information table is shown in Table 10.

Performance Test and Guarantee Design of Ideological and
Political Education App. When the network is transplanted to the last layer, the neural network compares the predicted situation with the actual situation and calculates the deviation inside. is calculation function is also called the loss function.
e commonly used loss function is the cross entropy loss function, and the calculation formula is e concept of the gradient formed by the jth neuron in the lth layer is (20) C is the value of the loss function to calculate the deviation of the last layer of the neural network en continue to calculate the deviation of the previous layer And the slope of the previous layer zc Assuming that the learning speed is set, the learning speed must be multiplied by the weight w and the deviation b of the variable update. Using the software testing method, package the application in the form of an apk and install it on a suitable machine, and test the main functions of the application, as shown in Table 11.

Analysis of App Communication Strategies in Ideological
and Political Education. Usually, people generally pay attention to the information they are interested in. erefore, we must first take the ideological and political training app as the focus of party building and focus on the timely change and system optimization of the app's content. erefore, when disseminating and promoting ideological and political Computational Intelligence and Neuroscience education apps, we should focus on the needs of the public, pay attention to customer feedback, choose effective content from the platform, and promote interactive circular information. e application of ideological and political education is very important for the rapid dissemination of information. In the Internet era, the public has access to a huge amount of information, and attention has become less important in the media. e speed and area of information transmission will have the greatest effect on the success or failure of evaluation and management.
Although the content of the ideological and political education application is different from other applications that people know, it is also a mobile terminal application that should pay great attention to user experience, system innovation, and optimization of functions and uses. Because the application is suitable for mobile terminal devices, the environment used by the application is mainly a mobile environment. erefore, the convenience of product operation and user comfort have become the main criteria for evaluating the experience of application users. erefore, the  primary goal of optimizing the process of ideological and political education app user experience is to be direct, simple, fast, comfortable, and easy to understand and give people a user-friendly and easy-to-use experience.

Conclusion
Different types of applications and diverse scenarios enable users to experience different personal experiences. is paper examines the use of application software for ideological and political theory courses in colleges and universities and the application of application software for ideological courses and analyzes the use of apps in the management of ideological and political theory courses in colleges and universities. Including the development and design of ideological and political theory applications, combining deep learning and CTC algorithms to build acoustic models, using server-client interaction, designing an ideological and political theory course app based on speech recognition and deep learning, and forming an offline speech recognition system software platform, the app provides a certain reference for improving the skills of teaching managers.   Data Availability e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e author(s) declare(s) that they have no conflicts of interest.