A containerized eloquence training method, device and medium with cross-platform consistency

Abstract

The present invention discloses a containerized eloquence training method, device and medium with cross-platform consistency, the method comprising: establishing an image based on application data, and integrating an eloquence dimension indicator algorithm into the image; deploying the image to several applications to obtain an application set; obtaining the user's eloquence training data through the application set, and obtaining an eloquence dimension indicator set based on the features extracted from the data; generating eloquence optimization suggestions based on the eloquence dimension indicator set. The present invention proposes a containerized eloquence training method, device and medium with cross-platform consistency, by integrating the relevant eloquence dimension indicator algorithm of eloquence training into the image and deploying it to multiple applications, ensuring that the training logic and algorithm that the user is exposed to are consistent regardless of the operating system or device, providing an effective basis for generating eloquence optimization suggestions, and being able to solve the problem of difficulty in performing cross-platform consistency eloquence training to generate eloquence optimization suggestions.

Description

Technical Field

[0001] The present invention relates to the field of speech training, and in particular to a containerized eloquence training method, device and medium with cross-platform consistency. Background Art

[0002] Eloquence training is usually a continuous process, requiring users to learn and practice at different times and on different devices. If the training content or progress on different platforms cannot be seamlessly connected, users may face the risk of repeated learning or missing content, which will affect the training effect. Due to differences in user interface, operating habits, compatibility, etc. between different operating systems and devices, if the eloquence training platform does not support cross-platform consistency, users may need to learn different operating methods and rules, which will undoubtedly affect users' willingness to continue to use and participate in training. Therefore, providing a consistent user experience can ensure that users can enjoy the same high-quality training content on any platform, enhancing user stickiness and satisfaction. Existing eloquence optimization suggestions are mainly to use the unified development environment provided by cross-platform development frameworks (such as React Native, Flutter, etc.) to enable developers to use the same code base to deploy applications on different operating systems and achieve cross-platform eloquence training.

[0003] However, in the existing cross-platform eloquence optimization suggestions, different operating systems and devices have significant differences in underlying architecture, API interfaces, hardware characteristics, etc. These differences require developers to use different technologies, libraries and tools when writing software for different platforms, which increases the difficulty of achieving consistency; and cross-platform development requires more time and resources to adapt to different operating systems and devices. For many developers or companies, due to cost and resource constraints, it may not be possible to fully cover all platforms, resulting in poor user experience or limited functions on some platforms. Summary of the invention

[0004] The present invention provides a containerized eloquence training method, device and medium with cross-platform consistency to solve the problem that it is difficult to perform cross-platform consistent eloquence training to generate eloquence optimization suggestions.

[0005] Get application data of several applications;

[0006] An image is established based on the application data, and an eloquence dimension index algorithm is integrated into the image; wherein the eloquence dimension index algorithm is established based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm;

[0007] Deploy the image to the plurality of applications to obtain an application set;

[0008] Acquiring the user's eloquence training data through the application set, and extracting the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain an eloquence dimension indicator set;

[0009] Generate eloquence optimization suggestions based on the eloquence dimension indicator set.

[0010] The present invention ensures the consistency and reusability of the algorithm by integrating the eloquence dimension index algorithm into the image. This integration method makes eloquence training no longer dependent on a single application or device, but can realize the unification of training data in multiple applications, thereby improving the systematicness and scientificity of the training. Deploying the image to multiple applications allows eloquence training to be carried out across different operating systems and devices. Since the eloquence dimension index algorithm is constructed based on a set of eloquence index calculation formulas of different dimensions, it can comprehensively and objectively evaluate the user's eloquence performance. Therefore, this multi-dimensional evaluation method helps to discover the advantages and disadvantages of users in different aspects, and provides a basis for formulating personalized eloquence optimization suggestions, thereby realizing personalized training guidance.

[0011] Compared with the prior art, the present invention integrates the relevant eloquence dimension indicator algorithms of eloquence training into the mirror and deploys them to multiple applications, thereby ensuring that the training logic and algorithm exposed to users are consistent regardless of the operating system or device. This unified deployment method avoids the problem of inconsistent algorithm implementation due to platform differences, thereby ensuring the standardization and uniformity of the training process, and providing an effective basis for generating eloquence optimization suggestions. Therefore, it can solve the problem of difficulty in conducting cross-platform consistent eloquence training to generate eloquence optimization suggestions.

[0012] As a preferred solution, the eloquence index calculation formula set is specifically:

[0013] According to the expression time, pause time and the number of preset unique words, the expression of speech speed, the expression of fluency and the expression of vocabulary richness are established respectively;

[0014] Establishing a pronunciation accuracy expression according to the dynamic time-warping distance between the user pronunciation signal and the standard pronunciation signal;

[0015] According to the number of compound sentences and the average feature score, sentence complexity expressions and expressiveness expressions are established;

[0016] The speaking speed expression, the fluency expression, the vocabulary richness expression, the pronunciation accuracy expression, the sentence complexity expression and the expressiveness expression constitute the eloquence index calculation formula set.

[0017] In this preferred embodiment, since the speed of speech can reflect the speaker's mental agility and information transmission efficiency, fluent oral expression can enhance the coherence and comprehensibility of information and reduce the audience's understanding barriers, rich vocabulary can more accurately express complex thoughts and emotions, and improve the accuracy and appeal of expression, accurate pronunciation is the basis of good oral expression, which can enhance the audience's acceptance and trust, complex sentence structure can carry more information and logical relationships, reflecting the speaker's language organization ability and thinking depth, and expressive language can touch the audience's heart and resonate and identify, so the establishment of speech speed expression, fluency expression, vocabulary richness expression, pronunciation accuracy expression, sentence complexity expression and expressiveness expression can intuitively reflect these eloquence characteristics of the speaker.

[0018] As a preferred solution, the expressive expression is specifically:

[0019]

[0020] Among them, N s is the total number of sentences, V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences and α is the smoothing parameter.

[0021] As a preferred solution, the pronunciation optimization algorithm is specifically as follows:

[0022] Based on the feature values ​​of the speech data, the pronunciation error is calculated by measuring the similarity between the user's pronunciation and the standard pronunciation;

[0023] Combining the weights of different time points, calculating the smoothed speech rate and fluency according to the speech rate and fluency of the user's voice, and identifying the user's pronunciation problems from the smoothed speech rate and fluency;

[0024] The pronunciation optimization algorithm is generated according to the pronunciation error and the solution process of the pronunciation problem.

[0025] This preferred solution can accurately calculate the pronunciation error of the user by measuring the similarity between the user's pronunciation and the standard pronunciation. This error can directly point out which phonemes, words or sentences the user has pronunciation deviations. The user's smoothed speech speed and fluency are calculated in combination with the weights at different time points, that is, the average speech speed and fluency after the fluctuation of the speech speed are considered. This method can identify problems such as too fast or too slow speech speed or insufficient fluency, thereby providing users with targeted improvement suggestions. Pronunciation problems are further identified from the smoothed speech speed and fluency, which not only focuses on the accuracy at the phoneme level, but also considers the overall performance of language rhythm and coherence; therefore, this comprehensive analysis method can more comprehensively reveal the user's pronunciation weaknesses in order to optimize pronunciation.

[0026] As a preferred solution, a mirror is established based on the application data, specifically:

[0027] Removing unused dependencies from the dependency set of the plurality of applications, and establishing a dependency library according to the dependency set after the removal;

[0028] On the basis of the preset basic image, the volume and startup time of the basic image are adjusted according to the image startup time and resource consumption, and the image is generated in combination with the dependent library.

[0029] In this preferred solution, since unused dependencies may take up additional disk space, increase build time and runtime resource consumption, by removing them to build a dependency library, the efficiency and performance of the application can be significantly improved. By clarifying and concentrating the dependencies that the application really needs, the dependency relationship is clearer and easier to manage, which helps reduce problems caused by dependency conflicts or outdated dependencies. Since a smaller image size means faster download and deployment speed, as well as lower storage costs, and a shorter startup time means faster response time and better user experience, the resulting image has higher scalability.

[0030] As a preferred solution, before deploying the image to the plurality of applications, the method further includes:

[0031] Generate unique identifiers for different versions of the image according to the identifier of the base image and the identifier of the list in the dependent library to obtain a unique identifier set;

[0032] Generate tags for different versions of the image according to the function feature sets and metadata sets of different versions of the base image to obtain a tag set;

[0033] Generate a unique identifier of the eloquence dimension index algorithm according to an identifier of the algorithm code used by the eloquence dimension index algorithm and an identifier of the training data set, and obtain a first unique identifier;

[0034] Different versions of the image are tracked based on the set of unique identifiers, the set of tags, and the first unique identifier.

[0035] This preferred solution generates a unique identifier for each different version of the image by combining the identifier of the base image and the identifier of the list in the dependency library, which can ensure that each image version has a unique identity, which is convenient for distinction and identification during storage, distribution and use. According to the functional feature sets and metadata sets of different versions in the base image, corresponding labels are generated for each version. These labels usually contain key information of the version, such as functional features, fixed vulnerabilities, new dependencies, etc., which helps users quickly understand the characteristics and differences of each version. By combining the identifier of the algorithm code and the identifier of the training data set, a unique identifier is generated for each eloquence dimension indicator algorithm. This ensures that the source, version and training data of the algorithm are traceable, thereby enhancing the reliability of the data and the reproducibility of the algorithm.

[0036] As a preferred solution, tracking different versions of the image according to the unique identifier set, the tag set and the first unique identifier is specifically as follows:

[0037] Add, update or delete the parent-child relationship between versions in the image according to the unique identifier set;

[0038] Matching corresponding tags to each version in the image according to the tag set;

[0039] Different versions of the eloquence dimension indicator algorithm are tracked according to the first unique identifier to make the eloquence training content in the several applications consistent.

[0040] The parent-child relationship between each version in the image of this preferred solution usually reflects the inheritance and development relationship between versions. By managing these relationships, the evolution of the image version can be clearly tracked to ensure that it can be traced back to a specific version when needed. By matching the corresponding label for each version in the image according to the label set, the image version that meets specific needs can be directly screened out, thereby improving the efficiency of development and deployment. Tracking different versions of the eloquence dimension indicator algorithm according to the first unique identifier can ensure that the algorithm used in evaluating or optimizing eloquence training content is consistent and traceable. These methods help to avoid differences in training content or deviations in evaluation results due to version inconsistencies, thereby maintaining the consistency and effectiveness of eloquence training content.

[0041] As a preferred solution, before deploying the image to the plurality of applications, the method further includes:

[0042] Deploy a number of content distribution network nodes within a preset range;

[0043] The node with the smallest delay from the plurality of user locations to the plurality of content distribution network nodes is used as the current node to obtain an optimal node set; wherein the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the content distribution network node;

[0044] The mirror image is transmitted to an optimal node set of the plurality of content distribution network nodes, and the data transmission time is reduced according to the weight of the eloquence dimension data.

[0045] This preferred solution deploys several content distribution network nodes within a preset range, and these nodes are scattered in different geographical locations so as to cover the user group more widely. Since the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the node, and multiple key factors of network transmission are comprehensively considered, the delay of data transmission can be further reduced by selecting the optimal node. Transmitting the mirror image related to eloquence training to the optimal node set can ensure that users can obtain the required data with the shortest path and the fastest speed.

[0046] As a preferred solution, the data transmission time is reduced according to the weight of the eloquence dimension data, specifically:

[0047] A weighted distance set is calculated based on a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and an equivalent distance from the eloquence dimension data to the number of content distribution network nodes;

[0048] The content distribution network node corresponding to the minimum weighted distance in the weighted distance set is used as the optimal path, and the eloquence dimension data in the mirror is controlled to be transmitted to the optimal node set.

[0049] This preferred solution selects the content distribution network node corresponding to the minimum weighted distance from the weighted distance set as the optimal path. This node is considered to be the node with the lowest data transmission cost and the highest efficiency after comprehensively considering the data weight and transmission distance. By selecting the optimal path, it can ensure that the eloquence dimension data can be transmitted to the target content distribution network node in the shortest time and with the lowest delay, thereby improving the real-time performance of data processing. The eloquence dimension data in the control mirror is transmitted to the optimal node set according to the optimal path, which means that the data will be sent to the content distribution network nodes selected as the optimal path, rather than randomly or evenly distributed to all nodes. This targeted data transmission method can significantly improve the efficiency and accuracy of data transmission, while reducing unnecessary network bandwidth consumption and transmission delays.

[0050] The present invention also provides a containerized eloquence training device with cross-platform consistency, including a data module, a mirror module, a deployment module, a training module and a solution module;

[0051] Wherein, the data module is used to obtain application data of several application programs;

[0052] The mirror module is used to establish a mirror based on the application data and integrate the eloquence dimension index algorithm into the mirror; wherein the eloquence dimension index algorithm is established based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm;

[0053] The deployment module is used to deploy the image to the plurality of applications to obtain an application set;

[0054] The training module is used to obtain the user's eloquence training data through the application set, and extract the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set;

[0055] The solution module is used to generate eloquence optimization suggestions based on the eloquence dimension indicator set.

[0056] As a preferred solution, the eloquence index calculation formula set is specifically:

[0057] According to the expression time, pause time and the number of preset unique words, the expression of speech speed, the expression of fluency and the expression of vocabulary richness are established respectively;

[0058] Establishing a pronunciation accuracy expression according to the dynamic time-warping distance between the user pronunciation signal and the standard pronunciation signal;

[0059] According to the number of compound sentences and the average feature score, sentence complexity expressions and expressiveness expressions are established;

[0060] The speaking speed expression, the fluency expression, the vocabulary richness expression, the pronunciation accuracy expression, the sentence complexity expression and the expressiveness expression constitute the eloquence index calculation formula set.

[0061] As a preferred solution, the expressive expression is specifically:

[0062]

[0063] Among them, N s is the total number of sentences, V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences and α is the smoothing parameter.

[0064] As a preferred solution, the pronunciation optimization algorithm is specifically as follows:

[0065] Based on the feature values ​​of the speech data, the pronunciation error is calculated by measuring the similarity between the user's pronunciation and the standard pronunciation;

[0066] Combining the weights of different time points, calculating the smoothed speech rate and fluency according to the speech rate and fluency of the user's voice, and identifying the user's pronunciation problems from the smoothed speech rate and fluency;

[0067] The pronunciation optimization algorithm is generated according to the pronunciation error and the solution process of the pronunciation problem.

[0068] As a preferred solution, the mirror module includes a dependency unit and a regulation unit;

[0069] The dependency unit is used to remove unused dependencies from the dependency set of the plurality of applications, and to establish a dependency library according to the removed dependency set;

[0070] The adjustment unit is used to adjust the volume and startup time of the base image based on the preset base image according to the image startup time and resource consumption, and generate the image in combination with the dependent library.

[0071] As a preferred solution, before deploying the image to the plurality of applications, it further includes an identification unit, a label unit, a code unit and a version unit;

[0072] The identification unit is used to generate unique identifiers of different versions in the image according to the identifier of the base image and the identifier of the list in the dependent library to obtain a unique identifier set;

[0073] A label unit, configured to generate labels for different versions of the image according to function feature sets and metadata sets of different versions of the base image, to obtain a label set;

[0074] A code unit, used to generate a unique identifier of the eloquence dimension index algorithm according to an identifier of an algorithm code used by the eloquence dimension index algorithm and an identifier of a training data set, to obtain a first unique identifier;

[0075] A version unit is used to track different versions of the image according to the unique identifier set, the tag set and the first unique identifier.

[0076] As a preferred solution, the version unit further includes a management subunit, a matching subunit and a tracking subunit;

[0077] Wherein, the management subunit is used to add, update or delete the parent-child relationship between the versions in the image according to the unique identifier set;

[0078] A matching subunit, used to match corresponding tags to each version in the image according to the tag set;

[0079] The tracking subunit is used to track different versions of the eloquence dimension indicator algorithm according to the first unique identifier, so as to make the eloquence training content in the several applications consistent.

[0080] As a preferred solution, before deploying the image to the plurality of applications, it further includes a distribution unit, a node unit and a transmission unit;

[0081] Wherein, the distribution unit is used to deploy a number of content distribution network nodes within a preset range;

[0082] A node unit, used to take a node with the smallest delay from a plurality of user locations to the plurality of content distribution network nodes as a current node, to obtain an optimal node set; wherein the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the content distribution network node;

[0083] The transmission unit is used to transmit the image to the optimal node set of the plurality of content distribution network nodes, and reduce the data transmission time according to the weight of the eloquence dimension data.

[0084] As a preferred solution, the transmission unit further includes a distance subunit and a path subunit;

[0085] The distance subunit is used to calculate a weighted distance set according to a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and an equivalent distance from the eloquence dimension data to the number of content distribution network nodes;

[0086] The path subunit is used to take the content distribution network node corresponding to the minimum weighted distance in the weighted distance set as the optimal path, and control the transmission of the eloquence dimension data in the mirror to the optimal node set.

[0087] The present application also provides a storage medium, on which a computer program is stored. The computer program is called and executed by a computer to implement the above-mentioned containerized eloquence training method with cross-platform consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 It is a flowchart of a containerized eloquence training method with cross-platform consistency provided by an embodiment of the present application;

[0089] Figure 2 It is a structural diagram of a containerized eloquence training device with cross-platform consistency provided in an embodiment of the present application. DETAILED DESCRIPTION

[0090] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0091] In the description of this application, it should be understood that the term "first" is used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "several" means two or more.

[0092] The embodiment of the present application provides a cross-platform consistent containerized eloquence training method, which is mainly used in situations where the eloquence training applications on existing mobile devices have poor correlation, it is difficult to ensure the consistency of eloquence training on different devices and operating systems, and it is necessary to obtain training effect feedback and optimization suggestions in a timely manner.

[0093] Embodiment 1:

[0094] See also Figure 1 The embodiment of the present application provides a containerized eloquence training method with cross-platform consistency, including S1 to S5, and the specific implementation steps are as follows:

[0095] S1. Obtain application data of several applications.

[0096] Step S1 of the embodiment of the present application is specifically as follows:

[0097] Get application data for several applications (such as "ELSA Speak", "Speeko", etc.).

[0098] S2. Building a mirror based on application data, and integrating the eloquence dimension index algorithm into the mirror; wherein the eloquence dimension index algorithm is built based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm.

[0099] Step S2 of the embodiment of the present application includes S2.1 to S2.4; wherein S2.1 is the process of constructing an eloquence index calculation formula set, S2.2 is the process of generating a pronunciation optimization algorithm, S2.3 is the process of generating an eloquence dimension index algorithm based on the eloquence index calculation formula set and the pronunciation optimization algorithm, and S2.4 is the process of generating a mirror image, specifically:

[0100] S2.1. According to the expression time, pause time, and the number of preset unique words, establish a speech rate expression SR, a fluency expression F, and a vocabulary richness expression VR respectively;

[0101] According to the dynamic time-warping distance between the user's pronunciation signal and the standard pronunciation signal, a pronunciation accuracy expression PA is established;

[0102] According to the number of compound sentences and the average feature score, the sentence complexity expression SC and the expressiveness expression E are established;

[0103] The eloquence index calculation formula set is composed of a speech speed expression, a fluency expression, a vocabulary richness expression, a pronunciation accuracy expression, a sentence complexity expression and an expressiveness expression; and the eloquence index calculation formula set also includes a content relevance expression CR;

[0104] Among them, each expression is specifically:

[0105] The expression of speaking speed is:

[0106]

[0107] The expression of fluency is:

[0108]

[0109] The expression of vocabulary richness is:

[0110]

[0111] The pronunciation accuracy expression is:

[0112]

[0113] The complexity expression is:

[0114]

[0115] The expressive expression is:

[0116]

[0117] The content relevance expression is:

[0118]

[0119] Among them, N w is the total number of words, T is the total time (seconds); N p is the number of pause words, N f is the number of filler words; N uis the number of unique words; DTW(X,Y) is the dynamic time warping distance between the user pronunciation signal X and the standard pronunciation signal Y; N cs is the number of compound sentences, N s is the total number of sentences; V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences, α is the smoothing parameter; N k is the number of keywords, θ i is the angle between the ith keyword and the topic vector.

[0120] In this embodiment S2.1, since the speed of speech can reflect the speaker's mental agility and information transmission efficiency, fluent oral expression can enhance the coherence and comprehensibility of information and reduce the audience's understanding barriers, rich vocabulary can more accurately express complex thoughts and emotions, and improve the accuracy and appeal of expression. Accurate pronunciation is the basis of good oral expression and can enhance the audience's acceptance and trust. Complex sentence structure can carry more information and logical relationships, reflecting the speaker's language organization ability and thinking depth. Expressive language can touch the hearts of the audience and resonate and identify. Therefore, the establishment of speech speed expression, fluency expression, vocabulary richness expression, pronunciation accuracy expression, sentence complexity expression and expressiveness expression can intuitively reflect these eloquence characteristics of the speaker.

[0121] S2.2, using Short-Time Fourier Transform (STFT) to perform frequency domain analysis on the user's eloquence data to obtain the characteristic value of the speech data;

[0122] Based on the feature values ​​of the speech data, the pronunciation error is calculated by measuring the similarity between the user's pronunciation and the standard pronunciation according to the fast DTW (Dynamic Time Warping) algorithm. Based on the pronunciation error, specific optimization suggestions are provided, for example, which syllables or frequency components need to be improved to help users improve their pronunciation accuracy.

[0123] In the container environment associated with the image, voice processing and indicator calculation are accelerated through parallel computing, and different computing tasks are distributed to multiple CPU cores for execution using multi-threading or multi-process technology;

[0124] Continuously collect speech speed and fluency data over a period of time, and assign weights to the values ​​at each time point; combine the weights at different time points, calculate the smoothed speech speed and fluency based on the speech speed and fluency of the user's voice, and identify the user's pronunciation problems from the smoothed speech speed and fluency; the weights are assigned in the following way: the data at the recent time point has a larger weight, and the data at the distant time point has a smaller weight;

[0125] A pronunciation optimization algorithm is generated based on the solution process of pronunciation errors and pronunciation problems.

[0126] Among them, the formula for frequency domain analysis of user eloquence data is:

[0127]

[0128] Where x(t) is the input time series signal, which represents the signal value at time t; x[m] is the sampled value of the discrete time signal x(t) at the mth moment; w[tm] is the window function, which is used to intercept a part of the signal in the short-time Fourier transform so that the transformation is concentrated in a short time window, and commonly used window functions include Hamming window, Hanning window, etc. is a complex exponential function, where j is an imaginary unit, j 2 = -1, used to convert the signal from the time domain to the frequency domain; k is the frequency index, indicating the position of a specific frequency component in the transform; m is the time index, indicating the sampling point of the signal in time; N is the window length, indicating the number of sampling points contained in each window in the short-time Fourier transform;

[0129] Moreover, in the short-time Fourier transform (STFT), the signal x(t) is divided into several short time periods, each of which is weighted by a window function w[tm], and then each weighted segment is subjected to a Fourier transform to obtain the frequency component of the signal in each time period, thereby analyzing the signal in both time and frequency dimensions.

[0130] The formula for calculating pronunciation error is:

[0131]

[0132] The calculation formula of smoothed speech rate is:

[0133]

[0134] The calculation formula for fluency is:

[0135]

[0136] If FastDTW(X,Y,R) is less than the preset distance value, it means that the user's pronunciation is highly similar to the standard pronunciation and the pronunciation accuracy is high; if FastDTW(X,Y,R) is greater than or equal to the preset distance value, it means that there is a large difference between the user's pronunciation and the standard pronunciation; i and i are the i-th user characteristic pronunciation and the i-th standard pronunciation feature respectively;

[0137] SR(t) is the speech rate at time t; w i is the weight of the i-th time point; n is a preset parameter for the size of the time window used for smoothing;

[0138] F(t) is the fluency at time t.

[0139] This embodiment S2.2 can accurately calculate the pronunciation error of the user by measuring the similarity between the user's pronunciation and the standard pronunciation, and the error can directly point out which phonemes, words or sentences the user has pronunciation deviations. The smoothed speech rate and fluency of the user are calculated in combination with the weights at different time points, that is, the average speech rate and fluency after the fluctuation of the speech rate are considered. This method can identify problems such as too fast, too slow or insufficient fluency, thereby providing users with targeted improvement suggestions. Pronunciation problems are further identified from the smoothed speech rate and fluency, which not only focuses on the accuracy at the phoneme level, but also takes into account the overall performance of language rhythm and coherence; therefore, this comprehensive analysis method can more comprehensively reveal the user's pronunciation weaknesses so as to optimize the pronunciation.

[0140] S2.3, establish an eloquence dimension index algorithm based on a set of calculation formulas for eloquence indexes of different dimensions and a pronunciation optimization algorithm;

[0141] Use an efficient database to store and manage the relevant training data of the eloquence dimension indicator algorithm; the efficient database can use a relational database or a NoSQL database;

[0142] Clean and preprocess the training data to remove noise and outliers to ensure data accuracy and reliability;

[0143] Use a version control system to manage updates and version control of training data to ensure data traceability and consistency.

[0144] S2.4. Define the image building process, including base image selection, dependency installation, and application configuration; optimize the image building process by reducing the image size through multi-stage building;

[0145] According to the optimized build process, several images are generated, and a small-sized and high-performance image is selected from the images as the base image;

[0146] The availability of the image is guaranteed by testing the base image. Specifically, it includes: integrating unit testing in the image building process to ensure the correctness of the application function; configuring and running integration tests to verify the overall operation of the application in the container; automated performance testing to detect the startup time and resource consumption of the image;

[0147] Remove unused dependencies from the dependency set of several applications, and create a dependency library based on the removed dependency set;

[0148] Based on the basic image, adjust the size and startup time of the basic image according to the image startup time and resource consumption, generate the image in combination with the dependent library, and integrate the eloquence dimension indicator algorithm into the image;

[0149] Optimize the operating environment of the image and improve the performance of the application in the container.

[0150] In this embodiment S2.4, since unused dependencies may take up additional disk space, increase build time and runtime resource consumption, by removing them to build a dependency library, the efficiency and performance of the application can be significantly improved. By clarifying and concentrating the dependencies that the application really needs, the dependency relationship is clearer and easier to manage, which helps to reduce problems caused by dependency conflicts or outdated dependencies. Since a smaller image size means faster download and deployment speed, as well as lower storage costs, and a shorter startup time means faster response time and better user experience, the image generated thereby has higher scalability;

[0151] In addition, integrating algorithms for eloquence dimension indicators such as speaking speed, fluency, and pronunciation accuracy into container images can ensure the consistency of training applications on different devices.

[0152] S3. Deploy the image to several applications to obtain an application set.

[0153] Step S3 of the embodiment of the present application includes S3.1 to S3.6; wherein S3.1 is a process of tracking and managing data, S3.2 is a process of optimizing the algorithm, S3.3 is a process of accelerating data transmission through CDN, S3.4 is a process of encrypting and storing data, S3.5 is a process of performing CI / CD integration to obtain an application set, and S3.6 is a process of monitoring data, specifically:

[0154] S3.1. Generate unique identifiers for different versions of the image based on the identifier of the base image and the identifier of the list in the dependency library to obtain a unique identifier set for the image version;

[0155] According to the functional feature sets and metadata sets of different versions in the basic image, labels of different versions in the image are generated to obtain a label set;

[0156] Generate a unique identifier of the eloquence dimension indicator algorithm according to an identifier of the algorithm code used by the eloquence dimension indicator algorithm and an identifier of the training data set, and obtain a first unique identifier;

[0157] Add, update, or delete the parent-child relationship between versions in the image based on a unique identifier set to facilitate tracing and rollback;

[0158] Match each version in the image with a corresponding tag according to the tag set, thereby identifying the functional characteristics of the version for easy retrieval and management;

[0159] Manage and track different versions of the eloquence dimension indicator algorithm according to the first unique identifier to make the eloquence training content in several applications consistent;

[0160] Stores the version number, generation time, and version description of each image, as well as image metadata such as image size, dependency library, and update log, and records the eloquence dimension indicator version corresponding to each image version;

[0161] The formula for generating a unique identifier set is:

[0162] VersionID=hash(BaseImage+Dependencies+Timestamp)

[0163] The formula for generating the label set is:

[0164] Tags i = hash(FeatureSet i +Metadata i )

[0165] The formula for generating the first unique identifier is:

[0166] AlgorithmVersionID=hash(AlgorithmCode+TrainingDataSet+Timestamp)

[0167] Among them, VersionID is the unique identifier of the image version, BaseImage is the identifier of the base image, Dependencies is the identifier of the dependency library list, and Timestamp is the timestamp of the version creation;

[0168] Tags i is the label of the i-th image version, FeatureSet i Metadata is the set of functional features of the i-th image version. i is the metadata set of the i-th image version;

[0169] AlgorithmVersionID is the unique identifier of the Eloquence Dimension algorithm version, AlgorithmCode is the identifier of the algorithm code, TrainingDataSet is the identifier of the training dataset, and Timestamp is the timestamp when the algorithm version was created.

[0170] This embodiment S3.1 generates a unique identifier for each different version of the image by combining the identifier of the base image and the identifier of the list in the dependency library, which can ensure that each image version has a unique identity, which is convenient for distinction and identification during storage, distribution and use. According to the functional feature sets and metadata sets of different versions in the base image, corresponding labels are generated for each version. These labels usually contain key information of the version, such as functional features, fixed vulnerabilities, new dependencies, etc., which helps users quickly understand the characteristics and differences of each version. By combining the identifier of the algorithm code and the identifier of the training data set, a unique identifier is generated for each eloquence dimension indicator algorithm. This ensures that the source, version and training data of the algorithm are traceable, thereby enhancing the reliability of the data and the reproducibility of the algorithm;

[0171] In addition, the parent-child relationship between versions in the image usually reflects the inheritance and development relationship between versions. By managing these relationships, the evolution of the image version can be clearly tracked to ensure that it can be traced back to a specific version when needed. By matching the corresponding label for each version in the image according to the label set, the image version that meets specific needs can be directly filtered out, thereby improving the efficiency of development and deployment. Tracking different versions of the eloquence dimension indicator algorithm according to the first unique identifier can ensure that the algorithm used when evaluating or optimizing eloquence training content is consistent and traceable. These methods help avoid differences in training content or deviations in evaluation results due to inconsistent versions, thereby maintaining the consistency and effectiveness of eloquence training content.

[0172] S3.2. Use parallel computing to accelerate data processing and feature extraction, and use multi-threading or multi-process technology to improve data processing efficiency;

[0173] Use weighted moving average (WMA) to smooth version data and reduce instantaneous fluctuations in data;

[0174] Use efficient database design and index structure to optimize data storage and retrieval performance;

[0175] Manage the eloquence dimension algorithm version to ensure the consistency of algorithms used by different image versions;

[0176] The formula for data processing is:

[0177]

[0178] The formula for feature extraction is:

[0179] FeatureExtraction(x(t))=STFT(x(t))

[0180] The formula for the smoothed version of the data is:

[0181]

[0182] Where D is the entire data set; N is the number of parts into which the data set is divided, that is, the number of tasks to be processed in parallel; Process(D i ) is to process the i-th data block D i process or function, x(t) represents the original signal or data that changes with time; STFT(x(t)) represents the function of performing short-time Fourier transform (STFT) on the signal x(t) to extract its frequency domain features, and Version(t) represents the mirror version corresponding to time t.

[0183] In this embodiment S3.2, based on the values ​​over a period of time, the weighted moving average effectively eliminates the instantaneous fluctuations caused by accidental factors, and the smoothed values ​​can more accurately reflect the overall trend of speech speed and fluency, rather than being disturbed by short-term fluctuations, thereby helping to provide more accurate training feedback and evaluation.

[0184] S3.3. Deploy several content delivery network (CDN) nodes around the world;

[0185] The node with the minimum transmission delay from several user locations to several content delivery network nodes (CDN nodes) is taken as the current node to obtain the optimal node set; wherein the transmission delay is calculated based on the bandwidth of the content delivery network node and the distance from the user location to the content delivery network node;

[0186] By transferring the image to the optimal node set of several content distribution network nodes, the image download speed is accelerated and the transmission delay is reduced; and the data transmission time is reduced through dynamic routing optimization.

[0187] The specific method of dynamic routing optimization is as follows:

[0188] According to a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and the equivalent distances from the eloquence dimension data to a number of content distribution network nodes, a weighted distance set is calculated;

[0189] The content distribution network node corresponding to the minimum weighted distance in the weighted distance set is taken as the optimal path, and the eloquence dimension data in the control mirror is transmitted to the optimal node set;

[0190] The formula for calculating the transmission delay is:

[0191]

[0192] The weighted distance in the weighted distance set can be expressed as:

[0193]

[0194] Among them, Latency(L,N i ) is the distance from user location L to the i-th node N i The delay, L is the user location; N is the CDN node set, including multiple nodes {N 1 ,N 2 ,…,N n}; D(L,N i ) is the user location L and the i-th node N i The distance between i ) is the i-th node N i bandwidth;

[0195] d i is the weighted distance to the jth CDN node, w j is the weight of the jth CDN node, d i,j Yes i To CDN node n j The distance is , and m is the total number of CDN nodes.

[0196] The specific process of transmitting the image is as follows:

[0197] Split the large image file in the image into multiple small blocks for transmission to improve transmission efficiency and reliability and avoid transmitting too much data at one time; for each uploaded block, if the upload is successful, continue to upload the next block, if the upload fails, return the failed block index i; for the upload result, after all blocks are successfully uploaded, return the total number of blocks k, if the upload fails, return the failed block index i;

[0198] If an interruption occurs during the transmission process, the transmission will continue from the interruption point to reduce resource waste;

[0199] The breakpoint resume function ensures the stability and continuity of the transmission process;

[0200] Among them, the file segmentation process can be expressed as:

[0201] Bi =F[(i-1)B:iB]

[0202] Where F is the file, B is the block size, and F = {B 1 ,B 2 ,…,B k}, each block B i The size of is B, and the size of the last block may be smaller than B; i is the sequence number of each block;

[0203] The breakpoint resume function is as follows:

[0204] Continue reading the file from the breakpoint position R and calculate the block index j corresponding to the breakpoint position;

[0205] For each block B i , from B j Start uploading operation; B j Indicates the block corresponding to block index j;

[0206] If the upload is successful, continue uploading the next block; if the upload fails, return to the new breakpoint position R';

[0207] After all blocks are uploaded successfully, "\text{None}" is returned to indicate that all are successful; if the upload fails, the new breakpoint position R' is returned.

[0208] Among them, the block index corresponding to the breakpoint position can be expressed as:

[0209]

[0210] The new breakpoint position can be expressed as:

[0211] R'=(i-1)B+(current position)

[0212] Among them, B is the block size and i is the sequence number of each block.

[0213] In this embodiment S3.3, several content distribution network nodes are deployed within a preset range, and these nodes are scattered in different geographical locations so as to cover a wider range of user groups. Since the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the node, multiple key factors of network transmission are comprehensively considered. Therefore, by selecting the optimal node, the delay of data transmission can be further reduced to ensure that users can get the fastest response time. Transmitting the mirror image related to eloquence training to the optimal node set can ensure that users can obtain the required data with the shortest path and the fastest speed;

[0214] In addition, the content distribution network node corresponding to the minimum weighted distance is selected from the weighted distance set as the optimal path. This node is considered to be the node with the lowest data transmission cost and the highest efficiency after comprehensively considering the data weight and transmission distance. By selecting the optimal path, it can be ensured that the eloquence dimension data can be transmitted to the target content distribution network node in the shortest time and with the lowest delay, thereby improving the real-time performance of data processing. The eloquence dimension data in the control mirror is transmitted to the optimal node set according to the optimal path, which means that the data will be sent to the content distribution network nodes selected as the optimal path, rather than randomly or evenly distributed to all nodes. This targeted data transmission method can significantly improve the efficiency and accuracy of data transmission, while reducing unnecessary network bandwidth consumption and transmission delays.

[0215] In addition, splitting large image files into multiple small blocks for transmission can improve transmission efficiency and reliability. CDN acceleration uses the content distribution network to accelerate image distribution, which can increase download speed and reduce latency. The block transmission and breakpoint resume mode supports block transmission and breakpoint resume of large image files, which improves the transmission speed and reliability of training data and ensures data consistency and real-time performance across regions and networks.

[0216] S3.4, encrypt and store the image and related data;

[0217] The role-based access control (RBAC) mechanism ensures that only authorized users can access and operate images.

[0218] Stores dimensional indicator data of eloquence training, such as speaking speed, fluency, pronunciation accuracy, etc.

[0219] Manage and track different versions of eloquence dimension indicators to ensure the timeliness and consistency of training content updates;

[0220] By optimizing the transmission path, the distribution of eloquence dimension indicator data is accelerated and the response speed of training applications is improved.

[0221] In this embodiment S3.4, the image and data are encrypted and stored, and a role-based access control (RBAC) mechanism is adopted so that only authorized users can access and operate, thereby ensuring the security and privacy of the data.

[0222] S3.5. When a developer submits the relevant code of the image to the version control system (such as Git), the CI / CD process is automatically triggered; and the code changes are automatically merged into the main branch or development branch;

[0223] After the code is merged, unit testing, integration testing, and performance testing are automatically triggered to ensure the correctness of the code changes;

[0224] After all tests pass, build a new container image based on the image and the test results, push the built container image to the image repository, and deploy the container image pulled from the image repository to several applications to obtain an application set. Tools such as Kubernetes can be used to deploy container images.

[0225] Automatically install the dependencies and libraries required by the application to ensure that the application set can run normally in the new environment; and configure the environment variables required by the application to ensure configuration consistency at runtime.

[0226] Automatically manage application dependencies and configurations through scripts to reduce manual configuration errors and time; manage and configure application dependencies and environment variables to ensure application operation consistency in different environments;

[0227] The process of configuring the environment variables required by the application can be expressed as:

[0228] Configure Env =Install Dependencies +Set Env Vars

[0229] Among them, Install Dependencies Indicates installation of application dependencies, Set Env Vars It is the variable that sets the environment;

[0230] This embodiment S3.5 can ensure that each code change is tested and built to ensure code quality by automating code testing and building new images; and configure the operating environment on different devices to ensure the correct operation of the container image.

[0231] S3.6. Monitor the deployment process and container operation status, and promptly detect and handle exceptions; record deployment and operation logs for troubleshooting and performance analysis;

[0232] Define and manage rollback strategies in different situations to ensure rapid recovery of versions when deployment fails; quickly roll back to the previous stable version when problems occur to reduce business impact;

[0233] Monitor the indicator feedback of the post-deployment training platform in real time to ensure consistent training effects of the application.

[0234] By real-time monitoring of the deployment process and container operation status, S3.6 of this embodiment can promptly discover and handle abnormal situations in deployment and operation, thereby reducing operational risks; by managing the rollback strategy, it can ensure that data can be quickly restored when deployment fails.

[0235] S4. Acquire the user's eloquence training data through the application set, and extract the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set.

[0236] Step S4 of the embodiment of the present application includes S4.1 to S4.3; wherein S4.1 is the process of training the user's eloquence and intelligently scheduling containers, S4.2 is the process of health check, and S4.3 is the process of obtaining the eloquence dimension indicator set, specifically:

[0237] S4.1, use the application set to train users in eloquence;

[0238] When the resource usage of the container exceeds or falls below the set threshold, resource scheduling is triggered. Based on the real-time needs of the container, the resource scheduling algorithm is used to dynamically allocate and adjust the resources (such as CPU, memory, etc.) of the computing application set to ensure that the container can maintain good operating performance under high load conditions.

[0239] Monitor the resource usage of containers (such as CPU, memory, etc.), and use time series prediction models to predict future resource requirements, adjust resource allocation in advance, and ensure stable operation of the system;

[0240] Monitor the traffic and load of each container instance; based on the monitoring results, use the load balancing algorithm to evenly distribute the traffic to different container instances to avoid single point of failure;

[0241] Regularly check the running status of container instances (such as CPU usage, memory usage, response time, etc.); isolate faulty container instances in a timely manner based on the inspection results to ensure that only healthy instances receive traffic and isolate faulty instances in a timely manner;

[0242] Among them, the formula of the resource allocation algorithm can be expressed as:

[0243]

[0244] The time series forecasting model can be expressed as:

[0245]

[0246] The formula of the load balancing algorithm can be expressed as:

[0247]

[0248] The process of checking the running status of a container instance can be expressed as:

[0249]

[0250] Among them, R d (t) is the resource demand of the container at time t, Rt (t) is the total resources available to the system at time t;

[0251] R i (t) is the resource usage in the i-th time period, w i is the weight of the i-th time period;

[0252] L i is the load of the ith container, C i is the capacity of the i-th container.

[0253] Health(i) is the health status of the i-th container, and Threshold is the threshold of the health status.

[0254] This embodiment S4.1 dynamically allocates computing resources according to the actual needs of the container and optimizes the operating performance, which can ensure the efficiency and flexibility of the container in resource use, improve resource utilization, reduce resource waste, and reduce operating costs; by balancing the load of different containers, the high availability and stability of the system are ensured.

[0255] S4.2. Provide a visual monitoring interface through the monitor to monitor the health of the system in real time and display key performance indicators (KPIs) such as CPU, memory and network usage through the dial;

[0256] Monitor system status in real time, discover and handle abnormal situations in a timely manner through alarm rules and notification channels, and configure to ensure that abnormal situations can be handled in a timely manner;

[0257] The health check mechanism monitors the running status of the container in real time, automatically detects faults during operation, and promptly discovers problems in container operation;

[0258] When a failure is detected, recovery measures such as automatically restarting the failed container and reallocating resources are automatically taken to ensure the continuous availability of the system.

[0259] The visual monitoring interface provided by S4.2 of this embodiment allows operation and maintenance personnel to understand the system status in real time; through real-time alarms, it can effectively ensure that problems in system operation can be responded to and resolved quickly, and automatically detect faults in container operation, ensuring the continuous availability of the system.

[0260] S4.3. In the application set, the dimension indicators of eloquence training (such as speech speed, fluency and pronunciation accuracy, etc.) are monitored in real time, and the user's voice data is collected in real time through a microphone or other audio input device to obtain the user's eloquence training data;

[0261] The eloquence training data is preprocessed by noise reduction, silent segment removal, and volume normalization, and then the features of the eloquence training data are extracted according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set and its related training data; wherein the eloquence dimension indicator set includes speech speed, fluency, vocabulary richness, pronunciation accuracy, sentence complexity, expressiveness, and content relevance;

[0262] The eloquence dimension indicator set is fed back to the user in real time through the user interface.

[0263] In this embodiment S4.3, the quality of voice input can be ensured through appropriate data collection methods, and the data is preprocessed to improve the quality of voice data, laying a foundation for feature extraction and index calculation.

[0264] S5. Generate eloquence optimization suggestions based on the eloquence dimension indicator set.

[0265] Step S5 of the embodiment of the present application includes S5.1 to S5.2; wherein S5.1 is the process of generating optimization suggestions, and S5.2 is the process of motivating users to continuously improve, specifically:

[0266] S5.1. Use a machine learning algorithm to perform modeling based on training data to obtain a machine learning model;

[0267] Use the user's historical eloquence data to train the machine learning model and optimize the model parameters;

[0268] Based on real-time monitoring data and the output of machine learning models, identify the user's eloquence strengths and weaknesses and generate personalized optimization suggestions;

[0269] The optimization suggestions are presented to users through the user interface, and detailed improvement plans are provided to guide users to conduct targeted exercises.

[0270] S5.2. Provide a secure user login mechanism to ensure the legitimacy of user identity; assign permissions based on user roles to ensure that different users can only access and operate the functions for which they have permission;

[0271] Provides a simple and easy-to-use application launch portal, allowing users to quickly enter the training platform where the application set is located; provides application running status and management functions, allowing users to easily view and control the running status of training applications;

[0272] By providing interactive interfaces that support text, audio, video and other forms to display training content, and based on real-time interactive feedback, help users understand the training effect and make corresponding adjustments;

[0273] The indicator display interface displays the user's eloquence training dimension indicators, such as speaking speed, fluency, pronunciation accuracy, and content relevance; and helps users improve their eloquence skills based on optimization suggestions;

[0274] Record the user's training progress, display the training process and results; set achievement goals and reward mechanisms to motivate users to continuously improve.

[0275] This embodiment S5.2 provides a rich training content display and interactive feedback interface, which displays eloquence dimension indicators and optimization suggestions in real time, helping users understand the training effect and make adjustments; through progress tracking and achievement reward mechanism, it can motivate users to continuously improve and enhance user experience;

[0276] Overall, this embodiment has the following beneficial effects:

[0277] The present invention ensures the consistency and reusability of the algorithm by integrating the eloquence dimension index algorithm into the image. This integration method makes eloquence training no longer dependent on a single application or device, but can unify training data in multiple applications, thereby improving the systematicness and scientificity of the training. Deploying the image to multiple applications allows eloquence training to be conducted across different operating systems and devices. Since the eloquence dimension index algorithm is constructed based on a set of eloquence index calculation formulas of different dimensions, it can comprehensively and objectively evaluate the user's eloquence performance. Therefore, this multi-dimensional evaluation method helps to discover the user's strengths and weaknesses in different aspects, and provides a basis for formulating personalized eloquence optimization suggestions, thereby enabling personalized training guidance;

[0278] In addition, through standardized container images, the training platform can run stably on Windows, Linux, macOS and other systems, ensuring the uniformity of user experience; by using edge computing and localized data processing methods, network transmission delays are reduced, and real-time data collection, analysis and feedback can be effectively realized.

[0279] Embodiment 2:

[0280] See also Figure 2 , the embodiment of the present application provides a containerized eloquence training device with cross-platform consistency, including a data module 10, a mirror module 20, a deployment module 30, a training module 40 and a solution module 50;

[0281] Wherein, the data module 10 is used to obtain application data of several application programs;

[0282] A mirror module 20, used to establish a mirror based on application data, and integrate an eloquence dimension index algorithm into the mirror; wherein the eloquence dimension index algorithm is established based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm;

[0283] A deployment module 30, used to deploy the image to a number of applications to obtain an application set;

[0284] The training module 40 is used to obtain the user's eloquence training data through the application program set, and extract the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set;

[0285] The solution module 50 is used to generate eloquence optimization suggestions based on the eloquence dimension indicator set.

[0286] In one embodiment, the data module 10 is specifically:

[0287] Get application data for several applications (such as "ELSA Speak", "Speeko", etc.).

[0288] In one embodiment, the mirror module 20 includes a formula unit, an algorithm unit, a building unit, a dependency unit and an adjustment unit; wherein the formula unit is a process of constructing an eloquence index calculation formula set, the algorithm unit is a process of generating a pronunciation optimization algorithm, the building unit is a process of generating an eloquence dimension index algorithm based on the eloquence index calculation formula set and the pronunciation optimization algorithm, and the dependency unit and the adjustment unit are processes of generating mirror images, specifically:

[0289] A formula unit, for establishing a speech rate expression SR, a fluency expression F, and a vocabulary richness expression VR according to the expression duration, pause time, and the number of preset unique words;

[0290] The formula unit is also used to establish a pronunciation accuracy expression PA according to the dynamic time warping distance between the user pronunciation signal and the standard pronunciation signal;

[0291] The formula unit is also used to establish sentence complexity expression SC and expressiveness expression E according to the number of compound sentences and the average feature score of the sentence;

[0292] The formula unit is further used to form an eloquence index calculation formula set from a speech speed expression, a fluency expression, a vocabulary richness expression, a pronunciation accuracy expression, a sentence complexity expression and an expressiveness expression; and the eloquence index calculation formula set also includes a content relevance expression CR;

[0293] Among them, each expression is specifically:

[0294] The expression of speaking speed is:

[0295]

[0296] The expression of fluency is:

[0297]

[0298] The expression of vocabulary richness is:

[0299]

[0300] The pronunciation accuracy expression is:

[0301]

[0302] The complexity expression is:

[0303]

[0304] The expressive expression is:

[0305]

[0306] The content relevance expression is:

[0307]

[0308] Among them, N w is the total number of words, T is the total time (seconds); N p is the number of pause words, N f is the number of filler words; N u is the number of unique words; DTW(X,Y) is the dynamic time warping distance between the user pronunciation signal X and the standard pronunciation signal Y; N cs is the number of compound sentences, N s is the total number of sentences; V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences, α is the smoothing parameter; N k is the number of keywords, θ i is the angle between the ith keyword and the topic vector.

[0309] In the formula unit of this embodiment, since the speed of speech can reflect the speaker's mental agility and information transmission efficiency, fluent oral expression can enhance the coherence and comprehensibility of information and reduce the audience's understanding barriers, rich vocabulary can more accurately express complex thoughts and emotions, and improve the accuracy and appeal of expression. Accurate pronunciation is the basis of good oral expression and can enhance the audience's acceptance and trust. Complex sentence structure can carry more information and logical relationships, reflecting the speaker's language organization ability and thinking depth. Expressive language can touch the hearts of the audience and resonate and identify. Therefore, the establishment of speech speed expressions, fluency expressions, vocabulary richness expressions, pronunciation accuracy expressions, sentence complexity expressions and expressive expressions can intuitively reflect these eloquence characteristics of the speaker.

[0310] An algorithm unit, used to perform frequency domain analysis on the user's eloquence data using a short-time Fourier transform (STFT) to obtain a characteristic value of the speech data;

[0311] The algorithm unit is also used to calculate the pronunciation error based on the feature value of the speech data by measuring the similarity between the user's pronunciation and the standard pronunciation according to the fast DTW (Dynamic Time Warping) algorithm; and provide specific optimization suggestions based on the pronunciation error; for example, point out which syllables or frequency components need to be improved to help users improve pronunciation accuracy;

[0312] The algorithm unit is also used to accelerate speech processing and indicator calculation through parallel computing in the container environment associated with the image, and use multi-threading or multi-process technology to distribute different computing tasks to multiple CPU cores for execution;

[0313] The algorithm unit is also used to continuously collect speech speed and fluency data over a period of time, assign weights to the values ​​at each time point, combine the weights at different time points, calculate the smoothed speech speed and fluency according to the speech speed and fluency of the user's voice, and identify the user's pronunciation problems from the smoothed speech speed and fluency; wherein the weights are assigned in such a way that the data at the nearest time point has a larger weight, and the data at the farther time point has a smaller weight;

[0314] The algorithm unit is also used to generate a pronunciation optimization algorithm based on the pronunciation error and pronunciation problem solution process.

[0315] Among them, the formula for frequency domain analysis of user eloquence data is:

[0316]

[0317] Where x(t) is the input time series signal, which represents the signal value at time t; x[m] is the sampled value of the discrete time signal x(t) at the mth moment; w[tm] is the window function, which is used to intercept a part of the signal in the short-time Fourier transform so that the transformation is concentrated in a short time window, and commonly used window functions include Hamming window, Hanning window, etc. is a complex exponential function, where j is an imaginary unit, j 2 = -1, used to convert the signal from the time domain to the frequency domain; k is the frequency index, indicating the position of a specific frequency component in the transform; m is the time index, indicating the sampling point of the signal in time; N is the window length, indicating the number of sampling points contained in each window in the short-time Fourier transform;

[0318] Moreover, in the short-time Fourier transform (STFT), the signal x(t) is divided into several short time periods, each of which is weighted by a window function w[tm], and then each weighted segment is subjected to a Fourier transform to obtain the frequency component of the signal in each time period, thereby analyzing the signal in both time and frequency dimensions.

[0319] The formula for calculating pronunciation error is:

[0320]

[0321] The calculation formula of smoothed speech rate is:

[0322]

[0323] The calculation formula for fluency is:

[0324]

[0325] If FastDTW(X,Y,R) is less than the preset distance value, it means that the user's pronunciation is highly similar to the standard pronunciation and the pronunciation accuracy is high; if FastDTW(X,Y,R) is greater than or equal to the preset distance value, it means that there is a large difference between the user's pronunciation and the standard pronunciation; i and i are the i-th user characteristic pronunciation and the i-th standard pronunciation feature respectively;

[0326] SR(t) is the speech rate at time t; w i is the weight of the i-th time point; n is a preset parameter for the size of the time window used for smoothing;

[0327] F(t) is the fluency at time t.

[0328] The algorithm unit of this embodiment can accurately calculate the pronunciation error of the user by measuring the similarity between the user's pronunciation and the standard pronunciation. The error can directly point out which phonemes, words or sentences the user has pronunciation deviations. The smoothed speech speed and fluency of the user are calculated in combination with the weights at different time points, that is, the average speech speed and fluency after the fluctuation of the speech speed are considered. This method can identify problems such as too fast, too slow or insufficient fluency, thereby providing users with targeted improvement suggestions. Pronunciation problems are further identified from the smoothed speech speed and fluency, which not only focuses on the accuracy at the phoneme level, but also takes into account the overall performance of language rhythm and coherence; therefore, this comprehensive analysis method can more comprehensively reveal the user's pronunciation weaknesses in order to optimize the pronunciation.

[0329] Establishing a unit for establishing an eloquence dimension indicator algorithm based on a set of calculation formulas for eloquence indicators of different dimensions and a pronunciation optimization algorithm;

[0330] The establishment unit is also used to use an efficient database to store and manage relevant training data of the eloquence dimension indicator algorithm; wherein the efficient database can use a relational database or a NoSQL database;

[0331] The establishment unit is also used to ensure the accuracy and reliability of the data by cleaning and preprocessing the training data to remove noise and outliers;

[0332] The establishment unit is also used to manage the update and version control of training data using a version control system to ensure the traceability and consistency of the data.

[0333] Dependency units are used to define the image building process, including base image selection, dependency installation, and application configuration. Multi-stage building can be used to reduce the image size to optimize the image building process.

[0334] The dependency unit is also used to generate several images according to the optimized build process, and select a small-sized and high-performance image from the several images as the base image;

[0335] Dependency units are also used to ensure the availability of images by testing the base images. Specifically, they include: ensuring the correctness of application functions by integrating unit tests in the image building process; configuring and running integration tests to verify the overall operation of the application in the container; and automated performance tests to detect the startup time and resource consumption of the image.

[0336] The dependency unit is also used to remove unused dependencies from the dependency set of several applications and to establish a dependency library based on the removed dependency set;

[0337] The adjustment unit is used to adjust the volume and startup time of the basic image based on the basic image according to the image startup time and resource consumption, generate the image in combination with the dependent library, and integrate the eloquence dimension indicator algorithm into the image;

[0338] The adjustment unit is also used to optimize the operating environment of the image and improve the performance of the application in the container.

[0339] In the dependency unit and adjustment unit of this embodiment, since unused dependencies may occupy additional disk space, increase build time and runtime resource consumption, by removing them to build a dependency library, the efficiency and performance of the application can be significantly improved. By clarifying and concentrating the dependencies that the application really needs, the dependency relationship is clearer and easier to manage, which helps to reduce problems caused by dependency conflicts or outdated dependencies. Since a smaller image size means faster download and deployment speed, as well as lower storage costs, and a shorter startup time means faster response time and better user experience, the image generated thereby has higher scalability;

[0340] In addition, integrating algorithms for eloquence dimension indicators such as speaking speed, fluency, and pronunciation accuracy into container images can ensure the consistency of training applications on different devices.

[0341] In one embodiment, the deployment module 30 includes an identification unit, a label unit, a code unit, a management subunit, a matching subunit, a tracking subunit, an optimization unit, a distribution unit, a node unit, a transmission unit, a storage unit, an integration unit, and a monitoring unit;

[0342] Among them, the identification unit, label unit, code unit, management subunit, matching subunit and tracking subunit are the processes of tracking and managing data, the optimization unit is the process of optimizing the algorithm, the distribution unit, node unit and transmission unit are the processes of accelerating data transmission through CDN, the storage unit is the process of encrypting and storing data, the integration unit is the process of performing CI / CD integration to obtain an application set, and the monitoring unit is the process of monitoring data, specifically:

[0343] An identification unit, used to generate unique identifiers of different versions in the image according to the identifier of the base image and the identifier of the list in the dependency library, and obtain a unique identifier set of the image version;

[0344] A label unit, used to generate labels for different versions of the image according to the function feature sets and metadata sets of different versions in the basic image, thereby obtaining a label set;

[0345] A code unit, used to generate a unique identifier of the eloquence dimension indicator algorithm according to an identifier of the algorithm code used by the eloquence dimension indicator algorithm and an identifier of the training data set, to obtain a first unique identifier;

[0346] The management subunit is used to add, update or delete the parent-child relationship between versions in the image based on the unique identifier set, which is convenient for tracing and rollback;

[0347] The matching subunit is used to match the corresponding tags of each version in the image according to the tag set, so as to identify the functional characteristics of the version for easy retrieval and management;

[0348] A tracking subunit, used for managing and tracking different versions of the eloquence dimension indicator algorithm according to the first unique identifier, so as to make the eloquence training content in several applications consistent;

[0349] The tracking subunit is also used to store the version number, generation time, and version description of each image, store the metadata of the image, such as image size, dependent libraries, and update logs, and record the eloquence dimension indicator version corresponding to each image version;

[0350] The formula for generating a unique identifier set is:

[0351] VersionID=hash(BaseImage+Dependencies+Timestamp)

[0352] The formula for generating the label set is:

[0353] Tags i = hash(FeatureSet i +Metadata i )

[0354] The formula for generating the first unique identifier is:

[0355] AlgorithmVersionID=hash(AlgorithmCode+TrainingDataSet+Timestamp)

[0356] Among them, VersionID is the unique identifier of the image version, BaseImage is the identifier of the base image, Dependencies is the identifier of the dependency library list, and Timestamp is the timestamp of the version creation;

[0357] Tags i is the label of the i-th image version, FeatureSet i Metadata is the set of functional features of the i-th image version. i is the metadata set of the i-th image version;

[0358] AlgorithmVersionID is the unique identifier of the Eloquence Dimension algorithm version, AlgorithmCode is the identifier of the algorithm code, TrainingDataSet is the identifier of the training dataset, and Timestamp is the timestamp when the algorithm version was created.

[0359] The identification unit, label unit, code unit, management subunit, matching subunit and tracking subunit of this embodiment generate a unique identifier for each different version of the image by combining the identifier of the base image and the identifier of the list in the dependency library, which can ensure that each image version has a unique identity, which is convenient for distinction and identification during storage, distribution and use. According to the functional feature sets and metadata sets of different versions in the base image, corresponding labels are generated for each version. These labels usually contain key information of the version, such as functional features, fixed vulnerabilities, new dependencies, etc., which helps users quickly understand the characteristics and differences of each version. By combining the identifier of the algorithm code and the identifier of the training data set, a unique identifier is generated for each eloquence dimension indicator algorithm. This ensures that the source, version and training data of the algorithm are traceable, thereby enhancing the reliability of the data and the reproducibility of the algorithm;

[0360] In addition, the parent-child relationship between versions in the image usually reflects the inheritance and development relationship between versions. By managing these relationships, the evolution of the image version can be clearly tracked to ensure that it can be traced back to a specific version when needed. By matching the corresponding label for each version in the image according to the label set, the image version that meets specific needs can be directly filtered out, thereby improving the efficiency of development and deployment. Tracking different versions of the eloquence dimension indicator algorithm according to the first unique identifier can ensure that the algorithm used when evaluating or optimizing eloquence training content is consistent and traceable. These methods help avoid differences in training content or deviations in evaluation results due to inconsistent versions, thereby maintaining the consistency and effectiveness of eloquence training content.

[0361] An optimization unit for accelerating data processing and feature extraction by using parallel computing, and using multi-threading or multi-process technology to improve data processing efficiency;

[0362] The optimization unit is also used to smooth the version data through weighted moving average (WMA) to reduce the instantaneous fluctuation of the data;

[0363] The optimization unit is also used to optimize data storage and retrieval performance by using efficient database design and index structure;

[0364] The optimization unit is also used to manage the version of the Eloquent Dimension algorithm to ensure the consistency of the algorithms used by different image versions;

[0365] The formula for data processing is:

[0366]

[0367] The formula for feature extraction is:

[0368] FeatureExtraction(x(t))=STFT(x(t))

[0369] The formula for the smoothed version of the data is:

[0370]

[0371] Where D is the entire data set; N is the number of parts into which the data set is divided, that is, the number of tasks to be processed in parallel; Process(D i ) is to process the i-th data block D i process or function, x(t) represents the original signal or data that changes with time; STFT(x(t)) represents the function of performing short-time Fourier transform (STFT) on the signal x(t) to extract its frequency domain features, and Version(t) represents the mirror version corresponding to time t.

[0372] The optimization unit of this embodiment is based on numerical values ​​over a period of time, and effectively eliminates instantaneous fluctuations caused by accidental factors through weighted moving average. The smoothed numerical values ​​can more accurately reflect the overall trend of speech speed and fluency, rather than being disturbed by short-term fluctuations, thereby helping to provide more accurate training feedback and evaluation.

[0373] A distribution unit, used to deploy a number of content distribution network (CDN) nodes around the world;

[0374] A node unit is used to take a node with the minimum transmission delay from a plurality of user locations to a plurality of content distribution network nodes (CDN nodes) as a current node to obtain an optimal node set; wherein the transmission delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the content distribution network node;

[0375] The transmission unit is used to accelerate the download speed of the image and reduce the transmission delay by transmitting the image to the optimal node set of several content distribution network nodes; and to reduce the data transmission time by means of dynamic route optimization.

[0376] The specific method of dynamic routing optimization is as follows:

[0377] According to a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and the equivalent distances from the eloquence dimension data to a number of content distribution network nodes, a weighted distance set is calculated;

[0378] The content distribution network node corresponding to the minimum weighted distance in the weighted distance set is taken as the optimal path, and the eloquence dimension data in the control mirror is transmitted to the optimal node set;

[0379] The formula for calculating the transmission delay is:

[0380]

[0381] The weighted distance in the weighted distance set can be expressed as:

[0382]

[0383] Among them, Latency(L,N i ) is the distance from user location L to the i-th node N i The delay, L is the user location; N is the CDN node set, including multiple nodes {N 1 ,N 2 ,…,N n}; D(L,N i ) is the user location L and the i-th node N i The distance between i ) is the i-th node N i bandwidth;

[0384] d i is the weighted distance to the jth CDN node, w j is the weight of the jth CDN node, d i,j Yes i To CDN node n j The distance is , and m is the total number of CDN nodes.

[0385] The specific process of transmitting the image is as follows:

[0386] Split the large image file in the image into multiple small blocks for transmission to improve transmission efficiency and reliability and avoid transmitting too much data at one time; for each uploaded block, if the upload is successful, continue to upload the next block, if the upload fails, return the failed block index i; for the upload result, after all blocks are successfully uploaded, return the total number of blocks k, if the upload fails, return the failed block index i;

[0387] If an interruption occurs during the transmission process, the transmission will continue from the interruption point to reduce resource waste;

[0388] The breakpoint resume function ensures the stability and continuity of the transmission process;

[0389] Among them, the file segmentation process can be expressed as:

[0390] B i =F[(i-1)B:iB]

[0391] Where F is the file, B is the block size, and F = {B 1 ,B2 ,…,B k}, each block B i The size of the block is B, and the size of the last block may be smaller than B; i is the ordinal number of the block;

[0392] The breakpoint resume function is as follows:

[0393] Continue reading the file from the breakpoint position R and calculate the block index j corresponding to the breakpoint position;

[0394] For each block B i , from B j Start uploading operation; B j Indicates the block corresponding to block index j;

[0395] If the upload is successful, continue uploading the next block; if the upload fails, return to the new breakpoint position R';

[0396] After all blocks are uploaded successfully, "\text{None}" is returned to indicate that all are successful; if the upload fails, the new breakpoint position R' is returned.

[0397] Among them, the block index corresponding to the breakpoint position can be expressed as:

[0398]

[0399] The new breakpoint position can be expressed as:

[0400] R'=(i-1)B+(current position)

[0401] Among them, B is the block size and i is the sequence number of each block.

[0402] In this embodiment, the distribution unit, node unit and transmission unit deploy several content distribution network nodes within a preset range. These nodes are scattered in different geographical locations so as to cover the user group more widely. Since the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the node, multiple key factors of network transmission are comprehensively considered. Therefore, by selecting the optimal node, the delay of data transmission can be further reduced to ensure that users can get the fastest response time. Transmitting the mirror image related to eloquence training to the optimal node set can ensure that users can obtain the required data with the shortest path and the fastest speed;

[0403] In addition, the content distribution network node corresponding to the minimum weighted distance is selected from the weighted distance set as the optimal path. This node is considered to be the node with the lowest data transmission cost and the highest efficiency after comprehensively considering the data weight and transmission distance. By selecting the optimal path, it can be ensured that the eloquence dimension data can be transmitted to the target content distribution network node in the shortest time and with the lowest delay, thereby improving the real-time performance of data processing. The eloquence dimension data in the control mirror is transmitted to the optimal node set according to the optimal path, which means that the data will be sent to the content distribution network nodes selected as the optimal path, rather than randomly or evenly distributed to all nodes. This targeted data transmission method can significantly improve the efficiency and accuracy of data transmission, while reducing unnecessary network bandwidth consumption and transmission delays.

[0404] In addition, splitting large image files into multiple small blocks for transmission can improve transmission efficiency and reliability. CDN acceleration uses the content distribution network to accelerate image distribution, which can increase download speed and reduce latency. The block transmission and breakpoint resume mode supports block transmission and breakpoint resume of large image files, which improves the transmission speed and reliability of training data and ensures data consistency and real-time performance across regions and networks.

[0405] A storage unit, used for encrypting and storing the image and related data;

[0406] The storage unit is also used for the role-based access control (RBAC) mechanism to ensure that only authorized users can access and operate the image.

[0407] The storage unit is also used to store dimensional indicator data of eloquence training, such as speaking speed, fluency, pronunciation accuracy, etc.;

[0408] The storage unit is also used to manage and track different versions of eloquence dimension indicators to ensure the timeliness and consistency of training content updates;

[0409] The storage unit is also used to accelerate the distribution of eloquence dimension indicator data by optimizing the transmission path and improve the response speed of the training application.

[0410] The storage unit of this embodiment encrypts and stores the image and data, and adopts a role-based access control (RBAC) mechanism so that only authorized users can access and operate, thereby ensuring the security and privacy of the data.

[0411] Integration unit, which is used to automatically trigger the CI / CD process when the developer submits the relevant code of the image to the version control system (such as Git); and automatically merge the code changes to the main branch or development branch;

[0412] Integration unit, which is used to automatically trigger unit testing, integration testing, and performance testing after the code is merged to ensure the correctness of the code changes;

[0413] The integration unit is used to build a new container image based on the image and the test results after all tests have passed, and push the built container image to the image warehouse, and deploy the container image pulled from the image warehouse to several applications to obtain an application set; tools such as Kubernetes can be used to deploy container images;

[0414] The integration unit is used to automatically install the dependencies and libraries required by the application to ensure that the application set can run normally in the new environment; and configure the environment variables required by the application to ensure configuration consistency at runtime.

[0415] Integration unit, which is used to automatically manage application dependencies and configurations through scripts, reducing manual configuration errors and time; managing and configuring application dependencies and environment variables to ensure application operation consistency in different environments;

[0416] The process of configuring the environment variables required by the application can be expressed as:

[0417] Configure Env =Install Dependencies +Set Env Vars

[0418] Among them, Install Dependencies Indicates installation of application dependencies, Set Env Vars It is the variable that sets the environment;

[0419] The integration unit of this embodiment can ensure that each code change is tested and built to ensure code quality by automating code testing and building new images; and configure the operating environment on different devices to ensure the correct operation of the container image.

[0420] The monitoring unit is used to monitor the deployment process and container operation status, detect and handle exceptions in a timely manner, and record deployment and operation logs for troubleshooting and performance analysis.

[0421] The monitoring unit is also used to define and manage rollback strategies in different situations to ensure rapid version recovery when deployment fails; when problems occur, quickly roll back to the previous stable version to reduce business impact;

[0422] The monitoring unit is also used to monitor the indicator feedback of the training platform after deployment in real time to ensure consistent training effects of the application.

[0423] The monitoring unit of this embodiment can timely discover and handle abnormal situations in deployment and operation by real-time monitoring of the deployment process and container operation status, thereby reducing operation risks; by managing the rollback strategy, it can ensure that data can be quickly restored when deployment fails.

[0424] In one embodiment, the training module 40 includes a training unit, an inspection unit, and a collection unit; wherein the training unit is a process of training the user's eloquence and intelligently scheduling containers, the inspection unit is a process of health inspection, and the collection unit is a process of obtaining an eloquence dimension indicator set, specifically:

[0425] A training unit for training users on eloquence using the application set;

[0426] The training unit is also used to trigger resource scheduling when the resource usage of the container exceeds or falls below the set threshold. According to the real-time needs of the container, the resource scheduling algorithm is used to dynamically allocate and adjust the resources (such as CPU, memory, etc.) of the computing application set to ensure that the container can still maintain good operating performance under high load conditions.

[0427] The training unit is also used to monitor the resource usage of the container (such as CPU, memory, etc.), and use the time series prediction model to predict future resource requirements, adjust resource allocation in advance, and ensure the stable operation of the system;

[0428] The training unit is also used to monitor the traffic and load of each container instance. Based on the monitoring results, a load balancing algorithm is used to evenly distribute the traffic to different container instances to avoid single point failures.

[0429] The training unit is also used to regularly check the running status of the container instance (such as CPU usage, memory usage, response time, etc.); isolate the faulty container instance in a timely manner based on the inspection results to ensure that only healthy instances receive traffic and isolate the faulty instance in a timely manner;

[0430] Among them, the formula of the resource allocation algorithm can be expressed as:

[0431]

[0432] The time series forecasting model can be expressed as:

[0433]

[0434] The formula of the load balancing algorithm can be expressed as:

[0435]

[0436] The process of checking the running status of a container instance can be expressed as:

[0437]

[0438] Among them, R d (t) is the resource demand of the container at time t, R t (t) is the total resources available to the system at time t;

[0439] R i (t) is the resource usage in the i-th time period, w i is the weight of the i-th time period;

[0440] L i is the load of the ith container, C i is the capacity of the i-th container.

[0441] Health(i) is the health status of the i-th container, and Threshold is the threshold of the health status.

[0442] The training unit of this embodiment dynamically allocates computing resources according to the actual needs of the container and optimizes the operating performance, which can ensure the efficiency and flexibility of the container in resource use, improve resource utilization, reduce resource waste, and reduce operating costs; by balancing the load of different containers, the high availability and stability of the system are ensured.

[0443] The inspection unit is used to provide a visual monitoring interface through the monitor, monitor the health of the system in real time, and display key performance indicators (KPIs) such as CPU, memory, and network usage through the dial;

[0444] The inspection unit is also used to monitor the system status in real time, promptly discover and handle abnormal situations through alarm rules and notification channels, and configure to ensure that abnormal situations can be handled in a timely manner;

[0445] The inspection unit is also used to monitor the running status of the container in real time through the health check mechanism, automatically detect faults during operation, and promptly discover problems in the container operation;

[0446] The inspection unit is also used to automatically take recovery measures such as automatically restarting the faulty container and reallocating resources when a fault is detected to ensure the continuous availability of the system.

[0447] The visual monitoring interface provided by the inspection unit of this embodiment allows operation and maintenance personnel to understand the system status in real time; through real-time alarms, it can effectively ensure that problems in system operation can be responded to and resolved quickly, and automatically detect faults in container operation, ensuring the continuous availability of the system.

[0448] The collection unit is used to monitor the dimension indicators of eloquence training (such as speaking speed, fluency and pronunciation accuracy, etc.) in real time in the application program set, collect the user's voice data in real time through a microphone or other audio input device, and obtain the user's eloquence training data;

[0449] The acquisition unit is also used to perform pre-processing operations such as noise reduction, removal of silent segments and volume normalization on the eloquence training data, and then extract the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set and its related training data; wherein the eloquence dimension indicator set includes speech speed, fluency, vocabulary richness, pronunciation accuracy, sentence complexity, expressiveness and content relevance;

[0450] The collection unit is also used to feed back the eloquence dimension indicator set to the user in real time through the user interface.

[0451] The acquisition unit of this embodiment can ensure the quality of voice input through appropriate data acquisition methods; and pre-process the data to improve the quality of voice data, laying a foundation for feature extraction and indicator calculation.

[0452] In one embodiment, the solution module 50 includes a suggestion unit and an incentive unit; wherein the suggestion unit is a process of generating optimization suggestions, and the incentive unit is a process of encouraging users to continuously improve, specifically:

[0453] A suggestion unit is used to use a machine learning algorithm to perform modeling according to the training data to obtain a machine learning model;

[0454] The suggestion unit is also used to train the machine learning model using the user's eloquence history data and optimize the model parameters;

[0455] The suggestion unit is also used to identify the user's eloquence strengths and weaknesses based on real-time monitoring data and the output of the machine learning model, and generate personalized optimization suggestions;

[0456] The suggestion unit is also used to display optimization suggestions to users through the user interface and provide detailed improvement plans to guide users to perform targeted exercises.

[0457] The incentive unit is used to provide a secure user login mechanism to ensure the legitimacy of the user's identity; assign permissions based on user roles to ensure that different users can only access and operate the functions for which they have permissions;

[0458] The incentive unit is also used to provide a simple and easy-to-use application startup portal, allowing users to quickly enter the training platform where the application set is located; it provides application running status and management functions, making it convenient for users to view and control the running status of the training application;

[0459] The motivation unit is also used to display training content by providing interactive interfaces that support text, audio, video and other forms, and to help users understand the training effect and make corresponding adjustments based on real-time interactive feedback;

[0460] The incentive unit is also used to display the user's eloquence training dimension indicators, such as speaking speed, fluency, pronunciation accuracy and content relevance, through the indicator display interface; and help users improve their eloquence skills according to optimization suggestions;

[0461] The incentive unit is also used to record the user's training progress, display the training process and results; set achievement goals and reward mechanisms to motivate users to continue to improve.

[0462] The incentive unit of this embodiment provides a rich training content display and interactive feedback interface, and displays eloquence dimension indicators and optimization suggestions in real time, helping users understand the training effect and make adjustments; through progress tracking and achievement reward mechanism, it can motivate users to continuously improve and enhance user experience;

[0463] Overall, this embodiment has the following beneficial effects:

[0464] The present invention ensures the consistency and reusability of the algorithm by integrating the eloquence dimension index algorithm into the image. This integration method makes eloquence training no longer dependent on a single application or device, but can unify training data in multiple applications, thereby improving the systematicness and scientificity of the training. Deploying the image to multiple applications allows eloquence training to be conducted across different operating systems and devices. Since the eloquence dimension index algorithm is constructed based on a set of eloquence index calculation formulas of different dimensions, it can comprehensively and objectively evaluate the user's eloquence performance. Therefore, this multi-dimensional evaluation method helps to discover the user's strengths and weaknesses in different aspects, and provides a basis for formulating personalized eloquence optimization suggestions, thereby enabling personalized training guidance;

[0465] In addition, through standardized container images, the training platform can run stably on Windows, Linux, macOS and other systems, ensuring the uniformity of user experience; by using edge computing and localized data processing methods, network transmission delays are reduced, and real-time data collection, analysis and feedback can be effectively realized.

[0466] Embodiment three:

[0467] The embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the containerized eloquence training method with cross-platform consistency;

[0468] Among them, the containerized eloquence training method with cross-platform consistency, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0469] The above are preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A containerized eloquence training method with cross-platform consistency, characterized in that: include: Get application data of several applications; An image is established based on the application data, and an eloquence dimension index algorithm is integrated into the image; wherein the eloquence dimension index algorithm is established based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm, wherein the set of eloquence index calculation formulas is composed of a speech rate expression, a fluency expression, a vocabulary richness expression, a pronunciation accuracy expression, a sentence complexity expression and an expressiveness expression, wherein the sentence complexity expression and the expressiveness expression are established based on the number of compound sentences and the average feature score of the sentence; Deploy the image to the plurality of applications to obtain an application set; Acquiring the user's eloquence training data through the application set, and extracting the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain an eloquence dimension indicator set; Generate eloquence optimization suggestions based on the eloquence dimension indicator set; The expressive expression is specifically: Among them, N s is the total number of sentences, V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences and α is the smoothing parameter.

2. A containerized eloquence training method with cross-platform consistency as claimed in claim 1, characterized in that: The eloquence index calculation formula set is specifically: According to the expression time, pause time and the number of preset unique words, the expression of speech speed, the expression of fluency and the expression of vocabulary richness are established respectively; Establishing a pronunciation accuracy expression according to the dynamic time warping distance between the user pronunciation signal and the standard pronunciation signal; According to the number of compound sentences and the average feature score, sentence complexity expressions and expressiveness expressions are established; The speaking speed expression, the fluency expression, the vocabulary richness expression, the pronunciation accuracy expression, the sentence complexity expression and the expressiveness expression constitute the eloquence index calculation formula set.

3. A containerized eloquence training method with cross-platform consistency as claimed in claim 1, characterized in that: The pronunciation optimization algorithm is specifically: Based on the feature values ​​of the speech data, the pronunciation error is calculated by measuring the similarity between the user's pronunciation and the standard pronunciation; Combining the weights of different time points, calculating the smoothed speech rate and fluency according to the speech rate and fluency of the user's voice, and identifying the user's pronunciation problems from the smoothed speech rate and fluency; The pronunciation optimization algorithm is generated according to the pronunciation error and the solution process of the pronunciation problem.

4. A containerized eloquence training method with cross-platform consistency as claimed in claim 1, characterized in that: The mirror image is established based on the application data, specifically: Removing unused dependencies from the dependency set of the plurality of applications, and establishing a dependency library according to the dependency set after the removal; On the basis of the preset basic image, the volume and startup time of the basic image are adjusted according to the image startup time and resource consumption, and the image is generated in combination with the dependent library.

5. A containerized eloquence training method with cross-platform consistency as claimed in claim 4, characterized in that: Before deploying the image to the plurality of applications, the method further includes: Generate unique identifiers for different versions of the image according to the identifier of the base image and the identifier of the list in the dependent library to obtain a unique identifier set; Generate tags for different versions of the image according to the function feature sets and metadata sets of different versions of the base image to obtain a tag set; Generate a unique identifier of the eloquence dimension index algorithm according to an identifier of the algorithm code used by the eloquence dimension index algorithm and an identifier of the training data set, and obtain a first unique identifier; Different versions of the image are tracked based on the set of unique identifiers, the set of tags, and the first unique identifier.

6. A containerized eloquence training method with cross-platform consistency as claimed in claim 5, characterized in that: Tracking different versions of the image according to the unique identifier set, the tag set, and the first unique identifier is specifically as follows: Add, update or delete the parent-child relationship between versions in the image according to the unique identifier set; Matching corresponding tags to each version in the image according to the tag set; Different versions of the eloquence dimension indicator algorithm are tracked according to the first unique identifier to make the eloquence training content in the several applications consistent.

7. A containerized eloquence training method with cross-platform consistency as claimed in claim 1, characterized in that: Before deploying the image to the plurality of applications, the method further includes: Deploy a number of content distribution network nodes within a preset range; The node with the smallest delay from the plurality of user locations to the plurality of content distribution network nodes is used as the current node to obtain an optimal node set; wherein the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the content distribution network node; The mirror image is transmitted to an optimal node set of the plurality of content distribution network nodes, and the data transmission time is reduced according to the weight of the eloquence dimension data.

8. A containerized eloquence training method with cross-platform consistency as claimed in claim 7, characterized in that: The data transmission time is reduced according to the weight of the eloquence dimension data, specifically: A weighted distance set is calculated based on a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and an equivalent distance from the eloquence dimension data to the number of content distribution network nodes; The content distribution network node corresponding to the minimum weighted distance in the weighted distance set is used as the optimal path, and the eloquence dimension data in the mirror is controlled to be transmitted to the optimal node set.

9. A containerized eloquence training device with cross-platform consistency, characterized in that: Includes data module, image module, deployment module, training module and solution module; Wherein, the data module is used to obtain application data of several application programs; The mirror module is used to establish a mirror based on the application data and integrate the eloquence dimension index algorithm into the mirror; wherein the eloquence dimension index algorithm is established based on a set of eloquence index calculation formulas of different dimensions and a pronunciation optimization algorithm, the set of eloquence index calculation formulas is composed of a speech rate expression, a fluency expression, a vocabulary richness expression, a pronunciation accuracy expression, a sentence complexity expression and an expressiveness expression, and the sentence complexity expression and the expressiveness expression are established according to the number of compound sentences and the average feature score of the sentence; The deployment module is used to deploy the image to the plurality of applications to obtain an application set; The training module is used to obtain the user's eloquence training data through the application set, and extract the features of the eloquence training data according to the eloquence dimension indicator algorithm to obtain the eloquence dimension indicator set; The solution module is used to generate eloquence optimization suggestions based on the eloquence dimension indicator set; The expressive expression is specifically: Among them, N s is the total number of sentences, V i is the comprehensive score of the ith sentence on several features, V avg is the average feature score of all sentences and α is the smoothing parameter.

10. A containerized eloquence training device with cross-platform consistency as claimed in claim 9, characterized in that: The formula set for calculating the eloquence index is specifically: According to the expression time, pause time and the number of preset unique words, establish the expression of speech speed, the expression of fluency and the expression of vocabulary richness respectively; Establishing a pronunciation accuracy expression according to the dynamic time warping distance between the user pronunciation signal and the standard pronunciation signal; According to the number of compound sentences and the average feature score, sentence complexity expressions and expressiveness expressions are established; The speaking speed expression, the fluency expression, the vocabulary richness expression, the pronunciation accuracy expression, the sentence complexity expression and the expressiveness expression constitute the eloquence index calculation formula set.

11. The cross-platform consistent containerized eloquence training device according to claim 9, characterized in that: The pronunciation optimization algorithm is specifically: Based on the feature values ​​of the speech data, the pronunciation error is calculated by measuring the similarity between the user's pronunciation and the standard pronunciation; Combining the weights of different time points, calculating the smoothed speech rate and fluency according to the speech rate and fluency of the user's voice, and identifying the user's pronunciation problems from the smoothed speech rate and fluency; The pronunciation optimization algorithm is generated according to the pronunciation error and the solution process of the pronunciation problem.

12. The cross-platform consistent containerized eloquence training device according to claim 9, characterized in that: The mirror module includes a dependency unit and a regulation unit; The dependency unit is used to remove unused dependencies from the dependency set of the plurality of applications, and to establish a dependency library according to the removed dependency set; The adjustment unit is used to adjust the volume and startup time of the base image based on the preset base image according to the image startup time and resource consumption, and generate the image in combination with the dependent library.

13. A containerized eloquence training device with cross-platform consistency as claimed in claim 12, characterized in that: Before deploying the image to the plurality of applications, it also includes an identification unit, a label unit, a code unit and a version unit; The identification unit is used to generate unique identifiers of different versions in the image according to the identifier of the base image and the identifier of the list in the dependent library to obtain a unique identifier set; A label unit, configured to generate labels for different versions of the image according to function feature sets and metadata sets of different versions of the base image, to obtain a label set; A code unit, used to generate a unique identifier of the eloquence dimension index algorithm according to an identifier of an algorithm code used by the eloquence dimension index algorithm and an identifier of a training data set, to obtain a first unique identifier; A version unit is used to track different versions of the image according to the unique identifier set, the tag set and the first unique identifier.

14. A containerized eloquence training device with cross-platform consistency as claimed in claim 13, characterized in that: The version unit also includes a management subunit, a matching subunit and a tracking subunit; Wherein, the management subunit is used to add, update or delete the parent-child relationship between the versions in the image according to the unique identifier set; A matching subunit, used to match corresponding tags to each version in the image according to the tag set; The tracking subunit is used to track different versions of the eloquence dimension indicator algorithm according to the first unique identifier, so as to make the eloquence training content in the several applications consistent.

15. The cross-platform consistent containerized eloquence training device according to claim 9, characterized in that: Before said deploying said image to said several applications, it also includes a distribution unit, a node unit and a transmission unit; Wherein, the distribution unit is used to deploy a number of content distribution network nodes within a preset range; A node unit, used to take a node with the smallest delay from a plurality of user locations to the plurality of content distribution network nodes as a current node, to obtain an optimal node set; wherein the delay is calculated based on the bandwidth of the content distribution network node and the distance from the user location to the content distribution network node; The transmission unit is used to transmit the image to the optimal node set of the plurality of content distribution network nodes, and reduce the data transmission time according to the weight of the eloquence dimension data.

16. A containerized eloquence training device with cross-platform consistency as claimed in claim 15, characterized in that: The transmission unit also includes a distance subunit and a path subunit; The distance subunit is used to calculate a weighted distance set according to a number of weights of the eloquence dimension data in the eloquence dimension index algorithm and an equivalent distance from the eloquence dimension data to the number of content distribution network nodes; The path subunit is used to take the content distribution network node corresponding to the minimum weighted distance in the weighted distance set as the optimal path, and control the transmission of the eloquence dimension data in the mirror to the optimal node set.

17. A storage medium, characterized in that: The storage medium stores a computer program, which is called and executed by a computer to implement a containerized eloquence training method with cross-platform consistency as described in any one of claims 1 to 8 above.

Images