A method, device and medium for optimizing eloquence based on instant feedback and tracking

Abstract

The present invention discloses an eloquence optimization method, device and medium based on instant feedback and tracking, the method comprising: calculating the probability distribution of the feedback strategy adapted by the user in a preset situation to obtain a first probability distribution; combining the first probability distribution with the user's current performance data in the speech data to generate a first feedback plan and train the user to obtain real-time training data; generating an eloquence expression optimization plan according to the user's expected skill level value and the real-time training data. The present invention proposes an eloquence optimization method, device and medium based on instant feedback and tracking, which performs preliminary training on the user through the first feedback plan generated by the user's current performance data, can obtain first-hand feedback real-time training data, timely understand the user's eloquence weaknesses, and then generate a more scientific eloquence expression optimization plan to track the user, which can solve the problem that it is difficult to conduct effective eloquence tracking training for the user based on feedback information.

Description

Technical Field

[0001] The present invention relates to the technical field of eloquence training, and in particular to an eloquence optimization method, device and medium based on instant feedback and tracking. Background Art

[0002] Eloquence training is a systematic and complex process. It not only requires individuals to have basic language skills and expression ability, but also requires continuous improvement and improvement in practice. In this process, the feedback mechanism plays a vital role; the feedback mechanism can help users understand their own performance and shortcomings, promote self-awareness, improve training effects, adjust training strategies, stimulate training motivation, and establish an effective communication bridge; the traditional eloquence training program mainly consists of setting goals and selecting materials, generating eloquence training programs, conducting eloquence training for users, and obtaining user feedback. At present, the traditional eloquence training method is that users obtain feedback in a timely manner during the training process, discover their own problems in eloquence expression, and then make adjustments and improvements. This is the user's own unilateral improvement of eloquence based on feedback.

[0003] However, this unilateral method of improving eloquence based on feedback cannot fully utilize the feedback information, making it difficult for users to accurately identify their own shortcomings, resulting in unnecessary training repetition and waste of resources; moreover, after users make adjustments based on feedback, they are unable to conduct timely eloquence training based on the latest adjustment results, making it difficult for users to achieve good training results through scientific eloquence training methods. Summary of the invention

[0004] The present invention provides an eloquence optimization method, device and medium based on instant feedback and tracking, so as to solve the problem that it is difficult to carry out effective eloquence tracking training for users according to feedback information.

[0005] In order to solve the above problems, the present invention provides an eloquence optimization method based on instant feedback and tracking, comprising:

[0006] Get the user's speech data;

[0007] According to the speech data, the probability distribution of the feedback strategy adapted by the user in a preset situation is calculated to obtain a first probability distribution; wherein the first probability distribution is a probability distribution of the most adapted feedback strategy obtained through iterative updating based on the feedback action of the user in a specific situation and a preset learning rate;

[0008] Combining the first probability distribution with the user's current performance data in the speech data to generate a first feedback scheme, and training the user according to the first feedback scheme to obtain real-time training data;

[0009] An eloquence expression optimization plan is generated according to the expected skill level value of the user and the current skill level value of the user in the real-time training data.

[0010] The present invention can understand the possible behavior patterns of users in different situations by performing probability distribution calculations on the feedback strategies of users in preset situations. Therefore, the first probability distribution provides users with quantitative information about behavior patterns, which helps to formulate more effective strategies. Combining the first probability distribution with the user's current performance data, the first feedback plan can be customized for the user more accurately according to the user's current performance and the weak points of eloquence, thereby obtaining the real-time training data fed back by the user after training; for eloquence training, these real-time training data are valuable resources. By analyzing these data, problems and deficiencies in the training process can be discovered at the first time. Therefore, an eloquence expression optimization plan is generated according to the user's expected skill level value and the user's current skill level value in the real-time training data. The user's expected skill level value and the current skill level value can be compared, and then the user's gaps can be clearly understood and further improved. Therefore, the eloquence expression optimization plan can accurately locate the user's weak links, provide targeted improvement suggestions for them, and conduct further tracking training for the user.

[0011] Compared with the prior art, the present invention performs preliminary training on users through a first feedback scheme generated by the user's current performance data, can obtain real-time training data of first-hand feedback, timely understand the user's eloquence weaknesses, and then generate a more scientific eloquence expression optimization scheme to conduct further tracking training on the user, thereby ensuring that the user can achieve good training results. Therefore, it can solve the problem of difficulty in conducting effective eloquence tracking training for users based on feedback information.

[0012] As a preferred solution, based on the speech data, the probability distribution of the feedback strategy adapted by the user in a preset scenario is calculated to obtain a first probability distribution, which is specifically:

[0013] Performing feature extraction on the speech data to obtain an optimized feature set;

[0014] Calculate the probability distribution of the feedback strategy given the optimized feature set and the current situational feature to obtain the initial probability distribution of the feedback strategy adapted by the user; wherein the current situational feature refers to the speech environment feature of the speech data;

[0015] The initial probability distribution is updated by strategy iteration according to the feedback action and learning rate of the user to obtain a first probability distribution of the feedback strategy adapted to the user.

[0016] This preferred solution generates an initial probability distribution by combining the user's optimized feature set and the current situational features. This distribution reflects which feedback strategies are more likely to be suitable for the user in the current situation, and generates a set of strategies by calculating the probability distribution, so that each strategy has its probability of being selected. This diversity enables the system to be more flexible when facing complex and changeable user behaviors and situations. In addition, the introduction of current situational features enables the system to perceive and understand the specific environment or conditions in which the user is located. Since the same user may require different feedback methods in different situations, this is crucial for determining the appropriate feedback strategy. By considering the user's feedback actions, the user's acceptance of the current feedback strategy can be understood in real time, and the introduction of the learning rate can both respond quickly to user changes and prevent instability in the update of the probability distribution.

[0017] As a preferred solution, the first probability distribution is combined with the user's current performance data in the speech data to generate a first feedback solution, specifically:

[0018] Using a personalized feedback model, generating the first feedback scheme according to the first probability distribution, the historical speech data of the user, and the current performance data of the user in the speech data;

[0019] The model parameters of the personalized feedback model are updated according to the learning rate and feedback loss function of the user.

[0020] This preferred solution generates a first feedback solution based on historical data and current performance data, which can more comprehensively evaluate the user's eloquence performance; by comparing the user's historical progress with the current performance, the first feedback solution can accurately fit the user's personal characteristics and current status, and then accurately point out the user's strengths and weaknesses in expression. In addition, the parameters of the model are updated according to the user's learning rate and feedback loss function, so that the personalized feedback model can better adapt to the user's learning characteristics.

[0021] As a preferred solution, an eloquence expression optimization solution is generated according to the expected skill level value of the user and the current skill level value of the user in the real-time training data, specifically:

[0022] quantifying the expressiveness of the real-time training data through a user growth model to obtain expression data;

[0023] Constructing a multi-objective optimization problem based on the expression data, and solving it to obtain personalized development suggestions;

[0024] generating a feedback strategy value according to the expected skill level value of the user and the current skill level value of the user in the real-time training data;

[0025] The eloquence expression optimization plan is generated according to the personalized development suggestions and the feedback strategy value.

[0026] As a preferred solution, the user growth model is specifically:

[0027] Establish a fluency scoring model by calculating the sentence length and pause ratio of preset data;

[0028] Establishing an emotion expression quantification module by quantifying the emotion expression intensity and diversity of the preset data;

[0029] Establishing a cognitive load analysis module according to the coefficient of the power ratio of each frequency band in the preset data;

[0030] The user growth model is composed of the fluency scoring model, the emotion expression quantification module and the cognitive load analysis module.

[0031] In this preferred solution, fluency is directly related to whether the user can convey his or her meaning clearly and coherently. By calculating the sentence length and pause ratio of the preset data, this indicator can be quantified through the established fluency scoring model, thereby more objectively evaluating the user's level of eloquence fluency. By quantifying the intensity and diversity of emotional expression and establishing an emotional expression quantification module, the user's emotional expression ability can be objectively evaluated, so that the user can better convey his or her intentions and feelings and enhance the effect of the speech. The coefficient of the power ratio of the data in each frequency band reflects the processing and activity of the user's brain on different information frequency bands during the expression process. Therefore, by establishing a cognitive load analysis module, the cognitive load status of the user when processing information and expressing ideas can be revealed, providing a scientific basis for subsequent feedback and guidance.

[0032] As a preferred solution, a multi-objective optimization problem is constructed based on the expression data, and the solution is used to obtain personalized development suggestions, specifically:

[0033] constructing a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load according to the expression data;

[0034] Solving the multi-objective optimization problem to obtain a personalized training path;

[0035] Calculate a phased comprehensive value considering skill growth according to the current state and target state of the user;

[0036] The personalized development suggestions are generated according to the personalized training path and the phased comprehensive value. This preferred solution can simultaneously consider the optimization needs of these three aspects by constructing a multi-objective optimization problem about eloquence expression, emotional expression and cognitive load, avoiding the problem of focusing on only a single indicator in the traditional single-objective optimization method, and ensuring that the established personalized training path can better meet the personalized development of users; by comparing the user's current state and target state, the user's strengths and weaknesses in skills, as well as the aspects that need further improvement, can be clearly identified. Therefore, by generating personalized development suggestions based on personalized training paths and phased comprehensive values, the personalized training path learning methods and strategies can be adjusted in a timely manner to ensure the effectiveness of personalized development suggestions.

[0037] As a preferred solution, the eloquence expression optimization solution is generated according to the personalized development suggestions and the feedback strategy value, specifically:

[0038] Mapping the feedback strategy value to the personalized development suggestion; wherein the feedback strategy value includes proportional gain, integral gain and differential gain;

[0039] According to the numerical value of the feedback strategy value, the load intensity, difficulty and response speed of the personalized development suggestion are adjusted according to the proportional gain, integral gain and differential gain respectively to obtain the eloquence expression optimization plan.

[0040] This preferred solution can provide specific direction for the adjustment of personalized development suggestions by mapping the feedback strategy value to personalized development suggestions; among them, the proportional gain focuses on the immediate adjustment of the current state, the integral gain considers the cumulative effect of the historical state, and the differential gain focuses on the prediction and trend of the future state. Such multi-dimensional adjustment can comprehensively consider the user's current performance, historical progress and future potential, so as to formulate a more comprehensive and effective eloquence expression optimization plan.

[0041] As a preferred solution, the feedback strategy value is specifically:

[0042]

[0043] Among them, K p , K i and K d are the gains of the proportional controller, the integral controller, and the differential controller, respectively. desired is the expected skill level value, S current is the current skill level value.

[0044] The present invention also provides an eloquence optimization device based on instant feedback and tracking, comprising a data module, a probability module, a training module and a solution module;

[0045] Wherein, the data module is used to obtain the user's speech data;

[0046] The probability module is used to calculate the probability distribution of the feedback strategy adapted by the user in a preset situation according to the speech data, and obtain a first probability distribution; wherein the first probability distribution is a probability distribution of the most adapted feedback strategy obtained through iterative updating based on the feedback action of the user in a specific situation and a preset learning rate;

[0047] The training module is used to combine the first probability distribution with the user's current performance data in the speech data to generate a first feedback scheme, and train the user according to the first feedback scheme to obtain real-time training data;

[0048] The solution module is used to generate an eloquence expression optimization solution according to the user's expected skill level value and the user's current skill level value in the real-time training data.

[0049] As a preferred solution, the probability module includes a feature unit, an initial unit and an update unit;

[0050] Wherein, the feature unit is used to extract features from the speech data to obtain an optimized feature set;

[0051] The initial unit is used to calculate the probability distribution of the feedback strategy under the given conditions of the optimized feature set and the current situation feature, and obtain the initial probability distribution of the feedback strategy adapted by the user; wherein the current situation feature refers to the speech environment feature of the speech data;

[0052] The updating unit is used to perform strategy iteration update on the initial probability distribution according to the feedback action and learning rate of the user to obtain a first probability distribution of the feedback strategy adapted to the user.

[0053] As a preferred solution, the training module includes a feedback unit;

[0054] The feedback unit is used to generate the first feedback scheme using a personalized feedback model according to the first probability distribution, the historical speech data of the user and the current performance data of the user in the speech data;

[0055] The model parameters of the personalized feedback model are updated according to the learning rate and feedback loss function of the user.

[0056] As a preferred solution, the solution module includes an expression unit, a suggestion unit, a skill unit and a comprehensive unit;

[0057] The expression unit is used to quantify the expression ability of the real-time training data through the user growth model to obtain expression data;

[0058] The suggestion unit is used to construct a multi-objective optimization problem based on the expression data and solve it to obtain personalized development suggestions;

[0059] The skill unit is used to generate a feedback strategy value according to the expected skill level value of the user and the current skill level value of the user in the real-time training data;

[0060] The integration unit is used to generate the eloquence expression optimization plan according to the personalized development suggestions and the feedback strategy value.

[0061] As a preferred solution, the user growth model is specifically:

[0062] Establish a fluency scoring model by calculating the sentence length and pause ratio of preset data;

[0063] Establishing an emotion expression quantification module by quantifying the emotion expression intensity and diversity of the preset data;

[0064] Establishing a cognitive load analysis module according to the coefficient of the power ratio of each frequency band in the preset data;

[0065] The user growth model is composed of the fluency scoring model, the emotion expression quantification module and the cognitive load analysis module.

[0066] As a preferred solution, the suggestion unit includes a target subunit, a path subunit, a stage subunit, and a development subunit;

[0067] Wherein, the target subunit is used to construct a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load according to the expression data;

[0068] The path subunit is used to solve the multi-objective optimization problem and obtain a personalized training path;

[0069] The stage subunit is used to calculate a stage-by-stage comprehensive value considering skill growth according to the current state and target state of the user;

[0070] The development subunit is used to generate the personalized development suggestion according to the personalized training path and the stage-by-stage comprehensive value.

[0071] As a preferred solution, the integration unit includes a mapping subunit and an adjustment subunit;

[0072] Wherein, the mapping subunit is used to map the feedback strategy value to the personalized development suggestion; wherein the feedback strategy value includes proportional gain, integral gain and differential gain;

[0073] The adjustment subunit is used to adjust the load intensity, difficulty and response speed of the personalized development suggestions according to the numerical value of the feedback strategy value, the proportional gain, the integral gain and the differential gain, so as to obtain the eloquence expression optimization plan.

[0074] As a preferred solution, the feedback strategy value is specifically:

[0075]

[0076] Among them, K p , K i and K d are the gains of the proportional controller, the integral controller, and the differential controller, respectively. desired is the expected skill level value, S current is the current skill level value.

[0077] 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 method for optimizing eloquence based on instant feedback and tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a flowchart of an eloquence optimization method based on instant feedback and tracking provided in an embodiment of the present application;

[0079] Figure 2 It is a structural diagram of an eloquence optimization device based on instant feedback and tracking provided in an embodiment of the present application. DETAILED DESCRIPTION

[0080] 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.

[0081] 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.

[0082] The embodiment of the present application provides an eloquence optimization method based on instant feedback and tracking, which is mainly used in situations where it is necessary to analyze the user's eloquence training feedback data and regenerate an eloquence expression optimization plan to track the user.

[0083] Embodiment 1:

[0084] See also Figure 1 The embodiment of the present invention provides an eloquence optimization method based on instant feedback and tracking, including S1 to S4, and the specific implementation steps are as follows:

[0085] S1. Obtain the user's speech data.

[0086] Step S1 of the embodiment of the present application is specifically as follows:

[0087] When the user is speaking, the user's voice is captured and recorded through a microphone or a dedicated voice recognition device, the user's facial expression is captured through a 3D facial expression capture device or a camera, and the user's physiological data is collected through a device containing a biosensor to obtain speech data.

[0088] S2. Calculate the probability distribution of the user's adapted feedback strategy in a preset situation based on the speech data to obtain a first probability distribution; wherein the first probability distribution is the probability distribution of the most adapted feedback strategy obtained through iterative updating based on the user's feedback action in a specific situation and a preset learning rate.

[0089] Step S2 of the embodiment of the present application includes S2.1 to S2.2; wherein S2.1 is a process of extracting an optimized feature set, and S2.2 is a process of generating a first probability distribution according to the optimized feature set, specifically:

[0090] S2.1. Extracting real-time feature vector X from speech data t ; Among them, X t =E(D t ), E(.) represents the feature extraction function; D t is a real-time data stream, representing the multimodal information contained at each time point t, such as speech feature S t 、Facial expression features F t and physiological characteristics t ;

[0091] Using online feature selection algorithms, such as online LASSO or stochastic gradient descent (SGD), according to the real-time feature vector X t Calculate the optimized feature set

[0092] Among them, the optimized feature set is:

[0093]

[0094] Among them, L(X t ,Y t ) is the loss function used to measure the real-time feature vector X t Predicted label Y t (e.g., instant speech performance scoring); λ is a regularization parameter used to control feature sparsity; ‖X t ‖1 is the L1 norm, which is used to facilitate feature selection.

[0095] This embodiment S2.1 can adapt to the user's state and environmental changes by using an online feature selection algorithm, making the extracted features more representative.

[0096] S2.2, through the preset situational awareness model H(.|C t ), using the probabilistic graphical model (PGM) or conditional random field (CRF) to calculate the given optimized feature set The most appropriate feedback strategy A under the conditions of the current situation characteristics t The probability distribution of the user's most suitable feedback strategy is obtained Among them, the current situation feature refers to the speech environment feature of the speech data;

[0097] Using the policy iteration method in reinforcement learning, such as the Actor-Critic algorithm, the initial probability distribution is adjusted according to the user's feedback action and the learning rate α Perform strategy iteration and update to obtain the first probability distribution π of the most appropriate feedback strategy for the user in a given situation t+1 (a|s);

[0098] Among them, the initial probability distribution is:

[0099]

[0100] The updating process of probability distribution can be expressed as:

[0101]

[0102] Among them, C t Indicates current situational characteristics, such as audience reaction, speech topic, etc.;

[0103] s is the state vector (containing and C t ), a is the feedback action, α is the learning rate; A t is the advantage function based on the current strategy, reflecting the additional benefit of taking action a relative to the current strategy, that is, A t =Q t (s,a)-V t (s), and Q t (s,a) is the value of the state-action pair, V t (s) is the value of state s; π t (a|s) means at time t, facing state s (including and C t ), the probability of taking action a (i.e., strategy), and π t+1 (a|s) is the updated strategy (i.e., the first probability distribution); is the policy gradient term, which represents the derivative of the policy with respect to its parameters and is used to guide how the policy should be adjusted in the direction of gain;

[0104] Among them, feedback action a refers to the reaction or behavior change made by the user after receiving the feedback provided by the system. It can be the user's adoption, modification or rejection of a feedback plan, or the improvement or regression shown by the user during the training process. For example, through voice recognition technology to monitor changes in speech speed, facial expression recognition to determine the richness of body language, physiological signal analysis to infer the degree of tension, etc., it can be judged whether the user has actually increased the speaking speed after receiving the suggestion to increase the speaking speed, or whether the user has reduced repeated expressions after receiving the prompt to reduce redundant information, and the degree of adoption of the suggestion. These feedback actions are an important basis for the system to evaluate the effectiveness of feedback strategies;

[0105] In addition, feedback actions a can be divided into several categories, mainly including but not limited to:

[0106] 1. Active adoption: Users implement feedback suggestions, such as improving speaking speed, adding body language, adjusting volume, etc.

[0107] 2. Passive resistance: The user ignores the feedback and does not make any changes.

[0108] 3. Partial adoption: The user only partially follows the suggestions, such as improving the speaking speed without changing the volume.

[0109] 4. Proactive Adjustment: Users reflect on themselves based on feedback and make improvements beyond the suggested scope.

[0110] Secondly, the learning rate α is the speed at which the system adjusts the feedback strategy, which determines the system's sensitivity to user feedback actions, that is, the speed at which the system updates the feedback strategy. If the learning rate α is too high, the system will be overly dependent on the most recent feedback actions, ignoring long-term trends, resulting in an unstable training process; if it is too low, the system may be unresponsive and unable to adjust the strategy in time to adapt to changes in user needs. Therefore, the setting of the learning rate α should take into account the user's learning ability and the stability of the system. It can be initially set through experiments or empirical formulas, and then dynamically adjusted according to actual conditions;

[0111] In addition, the dynamic adjustment of the learning rate α should follow the following principles:

[0112] 1. Adaptive adjustment: The learning rate should be adjusted dynamically according to the user's progress speed and feedback effect. If the user makes significant progress, the learning rate can be appropriately reduced, otherwise it can be increased.

[0113] 2. User characteristic matching: Consider the user's learning ability and preferences, and provide personalized learning rate settings for different users.

[0114] 3. Environmental considerations: The learning rate may need to be adjusted in different situations. For example, in a high-stress environment, the learning rate should be appropriately reduced to prevent users from feeling overly burdened.

[0115] In order to ensure the rationality of the iterative update of the strategy, it is also necessary to set up an evaluation mechanism to regularly check whether the update of the feedback strategy has effectively promoted the improvement of the user's eloquence. This may include regularly collecting users' satisfaction evaluations on the feedback plan and measuring the improvement of users' eloquence level through objective tests.

[0116] Specifically include:

[0117] 1. Quantitative analysis: Quantify the changes in users’ expressive abilities through the user growth model, such as data on fluency, emotional expression, and cognitive load.

[0118] 2. User satisfaction survey: Collect users’ subjective evaluation of feedback strategies and understand users’ acceptance and satisfaction with feedback.

[0119] 3. Skill level comparison: Compare the user’s desired skill level with their current skill level to evaluate the contribution of feedback strategies to skill improvement.

[0120] 4. Long-term tracking: Observe the user’s skill growth curve over a long period of time to verify the long-term effect of strategy iteration.

[0121] This embodiment S2.2 generates an initial probability distribution by combining the user's optimized feature set and the current situation features. This distribution reflects which feedback strategies are more likely to be suitable for the user in the current situation, and generates a set of strategies by calculating the probability distribution, so that each strategy has its probability of being selected. This diversity enables the system to be more flexible when facing complex and changeable user behaviors and situations. In addition, the introduction of current situation features enables the system to perceive and understand the specific environment or conditions in which the user is located. Since the same user may require different feedback methods in different situations, this is crucial for determining the appropriate feedback strategy. By considering the user's feedback actions, the user's acceptance of the current feedback strategy can be understood in real time, and the introduction of the learning rate can both respond quickly to user changes and prevent instability in the update of the probability distribution;

[0122] Moreover, the first probability distribution describes how to adjust the probability of taking the same action in the future according to the advantages generated by the current strategy, so that the strategy gradually tends to those behaviors that can bring higher returns or advantages. In this way, the system can learn better decision-making strategies under different states, which is helpful for the subsequent solution generation. In addition, reinforcement learning is integrated into the process of generating feedback strategies, so that the strategy can be continuously optimized according to the feedback effect, pursuing long-term performance improvement.

[0123] In addition, by defining feedback actions, setting learning rates, and evaluating the effect of feedback strategy updates, we can ensure that the strategy iteration updates are data-driven and meet the actual needs of users, thereby achieving more personalized and effective eloquence optimization training. These detailed operation methods enhance the operability and theoretical depth of the document, providing more solid support for practical applications.

[0124] S3. Combine the first probability distribution with the user's current performance data in the speech data to generate a first feedback plan, and train the user according to the first feedback plan to obtain real-time training data.

[0125] Step S3 of the embodiment of the present application is specifically as follows:

[0126] Using the personalized feedback model PFM, based on the user's historical speech data H hist and the user's current performance data in the speech data Through the first probability distribution π t+1 (a|s) performs continuous policy iteration to generate the first feedback solution R t ; For example, if a solution is particularly effective for a user in a specific situation, through learning, the system will be more inclined to recommend this solution in similar situations;

[0127] According to the first feedback scheme R tTrain users and obtain real-time training data;

[0128] Among them, the model parameter Θ of the personalized feedback model PFM t+1 It is updated according to the user's learning rate and feedback loss function, and the updating process can be expressed as:

[0129]

[0130] The first feedback plan is:

[0131]

[0132] Among them, Θ t is the original model parameter of the personalized feedback model PFM, η is the learning rate, is the feedback R t With ideal feedback The loss function between .

[0133] It should be noted that in the process of updating the parameters of the personalized feedback model PFM, designing a reasonable loss function is the key to ensuring the model training effect and guidance. The design of the loss function must take into account the goal of personalized feedback, that is, to maximize the efficiency and effect of improving the user's eloquence. An effective loss function should be able to accurately measure the gap between feedback and ideal feedback, while reflecting the characteristics and needs of user learning.

[0134] In the personalized feedback model PFM, the design of the loss function needs to meet the following principles:

[0135] 1. Accuracy: The loss function should be able to accurately reflect the quality of the feedback. That is, the closer the feedback is to the ideal feedback, the smaller the value of the loss function should be.

[0136] 2. Personalization: Considering the learning ability and preferences of different users, the loss function should be able to adapt to different users, that is, adjust the weight or form of the loss function according to the user's learning rate.

[0137] 3. Dynamicity: As the user's skill level improves, the loss function should be able to adjust dynamically to adapt to the user's current ability level and learning stage.

[0138] 4. Interpretability: The loss function should be designed to be easy to understand and interpret to facilitate debugging and optimizing the model.

[0139] Moreover, in actual design, the loss function may contain multiple components, such as prediction error, user satisfaction, skill improvement rate, etc. For example, a composite loss function can be designed that simultaneously considers the deviation between feedback and expected results, user acceptance of feedback, and the impact of feedback on user skill improvement.

[0140] In the document, a paragraph detailing the loss function design principles and specific implementation should be added after describing the personalized feedback model parameter update formula.

[0141] The following is an example of the process of designing a loss function:

[0142] In order to ensure the correct direction of model parameter updates, the design of the loss function must take into account both accuracy and personalization; a good loss function should be able to accurately measure the gap between feedback and ideal feedback while taking into account the user's learning characteristics. When designing a loss function, the following points should be considered:

[0143] 1. Ensure that the loss function accurately reflects the quality of the feedback, that is, when the feedback is closer to the ideal state, the value of the loss function should decrease accordingly.

[0144] 2. The loss function needs to be personalized and can be adjusted according to the user's learning rate and preferences to meet the needs of different users.

[0145] 3. As the user's skills improve, the loss function should be adjusted dynamically to ensure that the feedback strategy is always targeted at the user's current ability level and learning stage.

[0146] 4. The design of the loss function should be transparent and explainable to facilitate model debugging and optimization.

[0147] In this embodiment S3, the first feedback scheme is generated based on historical data and current performance data, which can more comprehensively evaluate the user's eloquence performance; by comparing the user's historical progress with the current performance, the first feedback scheme can accurately fit the user's personal characteristics and current status, and then accurately point out the user's strengths and weaknesses in expression. In addition, the parameters of the model are updated according to the user's learning rate and feedback loss function, so that the personalized feedback model can better adapt to the user's learning characteristics, ensuring the personalization and effectiveness of the feedback scheme;

[0148] In addition, by continuously receiving new data and updating model parameters online, the system can quickly adapt to the speaker's real-time performance and environmental changes.

[0149] S4. Generate an eloquence expression optimization plan based on the user's expected skill level value and the user's current skill level value in real-time training data.

[0150] Step S4 of the embodiment of the present application includes S4.1 to S4.7, wherein S4.1 is a process of constructing a user growth model, S4.2 is a process of obtaining expression data, S4.3 is a process of generating personalized development suggestions, S4.4 is a process of generating an eloquence expression optimization plan, S4.5 is a process of training users in eloquence expression and optimizing the plan according to the eloquence expression optimization plan, S4.6 is a process of realizing data visualization, and S4.7 is a process of performing privacy protection and personalized configuration, specifically:

[0151] S4.1. Establish a fluency scoring model by calculating the sentence length and pause ratio of the preset data;

[0152] By quantifying the intensity and diversity of emotional expression of preset data, an emotional expression quantification module is established;

[0153] According to the coefficient of the power ratio of each frequency band in the preset data, a cognitive load analysis module is established;

[0154] The user growth model is composed of a fluency scoring model, an emotion expression quantification module, and a cognitive load analysis module.

[0155] S4.2. Based on the fluency scoring model in the user growth model, natural language processing technology is used to calculate the number of words per minute (speech speed), average sentence length, pause duration and other indicators in the real-time training data to obtain the fluency value S fluency ;

[0156] According to the emotional expression quantification module in the user growth model, the emotional analysis algorithm is used to quantify the intensity and diversity of emotional expression in the real-time training data, such as the emotional expression value S obtained by the frequency of emotional words and the range of tone changes. emotion ;

[0157] According to the cognitive load analysis module in the user growth model, the power ratio of α, β, θ and δ bands is extracted through spectrum analysis based on EEG signals, and the cognitive load value Y is calculated for real-time training data. cogLoad ;

[0158] By the fluency value S fluency , emotional expression value S emotion and cognitive load value Y cogLoad constitutes expression data;

[0159] Among them, the fluency value is:

[0160]

[0161] The emotional expression values ​​are:

[0162]

[0163] The cognitive load values ​​are:

[0164]

[0165] Among them, w1, w2 and w3 are weighting factors; σ SentenceLength and σ PauseRatio are the standard deviation of sentence length and the standard deviation of pause ratio, which are used for normalization; Words represents vocabulary, Time represents the time occupied by vocabulary, SentenceLengthMean represents the average sentence length, and PauseRatio represents the pause ratio;

[0166] f i is the frequency of emotion category i, EmotionIntensity i It corresponds to the intensity of the emotion;

[0167] β j is the coefficient of the power ratio of each frequency band, ∈ is the error term, PowerRatio j It is the power ratio of each frequency band, and β0, δ and α refer to specific EEG frequency bands of brain electrical activity.

[0168] In the present embodiment S4.2, fluency is directly related to whether the user can clearly and coherently convey his or her meaning. By calculating the sentence length and pause ratio of the real-time training data, this indicator can be quantified through the established fluency scoring model, thereby more objectively evaluating the user's eloquence fluency level. By quantifying the intensity and diversity of emotional expression of real-time training data through the emotional expression quantification model, the user's emotional expression ability can be objectively evaluated, so that the user can better convey his or her intentions and feelings and enhance the effect of the speech. The coefficient of the power ratio of the data in each frequency band reflects the processing and activity of the user's brain on different information frequency bands during the expression process. Therefore, through the cognitive load analysis module, the cognitive load status of the user when processing information and expressing ideas can be revealed, providing a scientific basis for subsequent feedback and guidance.

[0169] S4.3, construct a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load based on expression data;

[0170] Solve the multi-objective optimization problem and obtain the personalized training path P * ;

[0171] According to the user's current status and target status, a dynamic programming method is used to plan long-term development goals and milestones, so as to establish a phased comprehensive value G that takes into account skill growth. longTerm ;

[0172] According to the personalized training path P * and the phased comprehensive value G longTermGenerate personalized development recommendations;

[0173] Among them, the personalized training path is:

[0174]

[0175] The comprehensive value of the phase is:

[0176]

[0177] Among them, U fluency and U emotion is the measure of the improvement of eloquence fluency and emotional expression by the training path, U cogLoad is the expected increase in cognitive load; W is the weight used to balance cognitive load and skill improvement; P represents the training path;

[0178] C is the cost function, S t is the state of stage t, A t is the action taken and T is the time period required to achieve the final goal.

[0179] This embodiment S4.3 constructs a multi-objective optimization problem about eloquence expression, emotional expression and cognitive load, and can simultaneously consider the optimization needs of these three aspects, avoiding the problem of focusing on only a single indicator in the traditional single-objective optimization method, and ensuring that the established personalized training path can better meet the personalized development of users; by comparing the user's current status and target status, the user's strengths and weaknesses in skills, as well as areas that need further improvement, can be clearly identified, so personalized development suggestions are generated based on the personalized training path and the staged comprehensive value, and the personalized training path learning methods and strategies can be adjusted in time to ensure the effectiveness of the personalized development suggestions.

[0180] S4.4. According to the user's expected skill level value S desired and the user's current skill level value S in the real-time training data current , generate feedback strategy value Δπ;

[0181] The feedback strategy value Δπ is mapped to the personalized development suggestion according to the mapping rule; for example, if the feedback strategy value Δπ is positive, it may mean that the current personalized development suggestion is too conservative and the user needs more challenges or higher intensity exercises; on the contrary, if the feedback strategy value Δπ is negative, it may be necessary to reduce the difficulty of the exercise or provide more supportive feedback; the feedback strategy value Δπ includes the gain K of the proportional controller p , the gain K of the integral controller i and controls the gain K of the derivative controller d ;

[0182] According to the numerical value and sign of the feedback strategy value, the difficulty coefficient of the training task, the detail of the feedback information, the incentive intensity of the reward system, etc. are adjusted according to the adjustment algorithm to obtain the eloquence expression optimization plan; specifically: if the gain K of the proportional controller p If the deviation is large, the direct intensity of the personalized development suggestion is adjusted according to the deviation; through the gain K of the integral controller i Accumulate past deviations to determine whether longer-term strategy adjustments are needed, such as adjusting the difficulty of the course plan; control the gain K of the derivative controller d Respond quickly to changes in user skill levels, such as increasing difficulty to maintain challenge as users progress rapidly.

[0183] Among them, the generation feedback strategy value is:

[0184]

[0185] Among them, K p , K i and K d are the gain of the proportional controller, the gain of the integral controller and the gain of the differential controller respectively; S desired is the expected skill level, S current is the current skill level value, and S is reduced by adjusting the feedback strategy π desired and S current The difference between them is used to obtain the final feedback strategy value Δπ.

[0186] It should be noted that the above mapping rules and adjustment algorithms are specifically as follows:

[0187] (I) Mapping rules

[0188] The proportional gain (P) directly affects the instant adjustment strength of the personalized development suggestions, that is, the strength of the suggestions. If the feedback strategy value is positive, it means that the user's current training intensity is not enough to achieve the expected skill level. At this time, the load intensity of the personalized development suggestions should be increased to encourage the user to accept more challenging tasks. On the contrary, if the feedback strategy value is negative, it means that the user may face too much pressure, and the training difficulty should be reduced and more supportive feedback should be given.

[0189] Integral gain (I) focuses on historical performance and is used to adjust the long-term strategy of personalized development suggestions. Integral gain helps the system understand the user's learning curve by accumulating past deviations and decides whether the difficulty of the course plan needs to be adjusted. For example, if the user fails to reach the goal in the past period of time, the integral gain will gradually increase, prompting the system to provide more coaching and support.

[0190] Differential gain (D) focuses on the prediction of future states, ensuring that personalized development recommendations can quickly respond to changes in the user's skills. Differential gain is adjusted according to the speed at which the user's skill level changes. If the user progresses rapidly, the differential gain will increase so that the task difficulty can be increased in time to maintain the challenge; conversely, if the progress is slow, the differential gain will be reduced to prevent the user from losing confidence due to excessive task difficulty.

[0191] (II) Adjustment algorithm

[0192] The proportional gain (P) directly affects the immediate adjustment strength of the personalized development suggestions, the integral gain (I) focuses on historical performance and is used to adjust the long-term strategy of personalized development suggestions, and the differential gain (D) focuses on the prediction of future states to ensure that personalized development suggestions can quickly respond to user skill changes;

[0193] Furthermore, the algorithm needs to be adjusted to take into account the gap between the user’s current skill level and the desired skill level, as well as the user’s past performance. The core of the algorithm is to calculate the feedback strategy value, which is determined by the difference between the desired skill level value and the current skill level value.

[0194] In this embodiment S4.4, by mapping the feedback strategy value to the personalized development suggestion, a specific direction can be provided for adjusting the personalized development suggestion; wherein, the gain of the proportional controller focuses on the immediate adjustment of the current state, the gain of the integral controller considers the cumulative effect of the historical state, and the gain of the differential controller focuses on the prediction and trend of the future state. Such a multi-dimensional adjustment can comprehensively consider the user's current performance, historical progress and future potential, thereby formulating a more comprehensive and effective eloquence expression optimization plan;

[0195] In addition, the mapping and adjustment of feedback strategy values ​​are key links in ensuring the accuracy and practicality of the eloquence expression optimization plan; the proportional gain, integral gain and differential gain correspond to the load intensity, difficulty and response speed of personalized development suggestions respectively; through the formulation of clear rules and algorithms, effective mapping and adjustment can be achieved, and the feedback strategy value can be ensured to accurately reflect the user's eloquence level and training needs.

[0196] S4.5. Provide eloquence training to users according to the eloquence optimization plan, and continuously collect user feedback and performance data;

[0197] The eloquence expression optimization scheme is further refined based on user feedback and performance data, and the calculation parameters (i.e., K p , K i and K d) to make fine adjustments so that the eloquence optimization plan can be dynamically optimized as the user progresses and the environment changes, and an optimized eloquence optimization plan can be obtained; for example, if the user responds well to positive feedback and makes significant progress, the proportion of positive incentives can be increased; if the user shows an increased cognitive load when encountering a specific type of challenge, the plan can be adjusted to reduce the frequency of such tasks or provide auxiliary tools to reduce the burden;

[0198] By analyzing users' responsiveness and adoption rates to different feedback strategies, and through methods such as A / B testing, we evaluate the effectiveness of the strategies and adjust the strategy generation logic accordingly to ensure continuous improvement and personalized matching of the feedback mechanism and eloquence expression optimization solutions.

[0199] S4.6. Build a multi-dimensional achievement indicator system based on user feedback during training, including but not limited to skill scores, cognitive load improvement rate, and training duration, and use scatter plots, line graphs, etc. to visualize the multi-dimensional achievement indicator system.

[0200] S4.7. Protect data security, specifically: encrypt user data using the Advanced Encryption Standard (AES) or other encryption technologies to ensure data security during transmission and storage; and implement strict access control policies to ensure that only authorized users can access their personal data;

[0201] Set up data management, specifically: set personal preferences for the first feedback plan and eloquence expression optimization plan in the plan according to user preferences, such as the style, difficulty and content of eloquence training; and allow users to manage their own privacy settings through the user interface, including data sharing and usage preferences.

[0202] Track data, specifically: use logs to record data usage, provide transparency, and allow users to track how their data is used; provide clear privacy policy explanations for data to ensure that users understand how their data is processed and protected.

[0203] Overall, this embodiment has the following beneficial effects:

[0204] The present invention can understand the possible behavior patterns of users in different situations by performing probability distribution calculations on the feedback strategies of users in preset situations. Therefore, the first probability distribution provides users with quantitative information about behavior patterns, which helps to formulate more effective strategies. Combining the first probability distribution with the user's current performance data, the first feedback plan can be customized for the user more accurately based on the user's current performance and targeting the weak points of eloquence, thereby obtaining the real-time training data fed back by the user after training; for eloquence training, these real-time training data are valuable resources. By analyzing these data, problems and deficiencies in the training process can be discovered at the first time. Therefore, an eloquence expression optimization plan is generated based on the user's expected skill level value and the user's current skill level value in the real-time training data. The user's expected skill level value and the current skill level value can be compared, and then a clear understanding of the user's gaps and the areas in which further improvement is needed can be obtained. Therefore, the eloquence expression optimization plan can accurately locate the user's weak links, provide targeted improvement suggestions for the user, and conduct further tracking training for the user;

[0205] In addition, by combining the user's historical performance and real-time training data, the training content and difficulty are dynamically adjusted to provide users with a learning path that best suits their current ability level and progress rate. This adaptability ensures the continued challenge of training, avoids overload or inefficient learning, and promotes long-term stable development. When processing user data, the system uses advanced encryption technology and strict data access control mechanisms to ensure the security and privacy of user information. At the same time, users are allowed to personalize data sharing permissions, which increases users' trust in the system and willingness to use it.

[0206] Embodiment 2:

[0207] See also Figure 2 , an embodiment of the present invention provides an eloquence optimization device based on instant feedback and tracking, including a data module 10, a probability module 20, a training module 30 and a solution module 40;

[0208] Wherein, the data module 10 is used to obtain the user's speech data;

[0209] The probability module 20 is used to calculate the probability distribution of the feedback strategy adapted by the user in a preset situation according to the speech data, and obtain a first probability distribution; wherein the first probability distribution is a probability distribution of the most adapted feedback strategy obtained through iterative updating based on the feedback action of the user in a specific situation and a preset learning rate;

[0210] A training module 30, for combining the first probability distribution with the user's current performance data in the speech data to generate a first feedback scheme, and training the user according to the first feedback scheme to obtain real-time training data;

[0211] The solution module 40 is used to generate an eloquence expression optimization solution according to the user's expected skill level value and the user's current skill level value in the real-time training data.

[0212] In one embodiment, the data module 10 is specifically:

[0213] When the user is speaking, the user's voice is captured and recorded through a microphone or a dedicated voice recognition device, the user's facial expression is captured through a 3D facial expression capture device or a camera, and the user's physiological data is collected through a device containing a biosensor to obtain speech data.

[0214] In one embodiment, the probability module 20 includes a feature unit, an initial unit, and an update unit; wherein the feature unit is a process of extracting an optimized feature set, and the initial unit and the update unit are processes of generating a first probability distribution according to the optimized feature set, specifically:

[0215] Feature unit, used to extract real-time feature vector X from speech data t ; Among them, X t =E(D t ), E(.) represents the feature extraction function; D t is a real-time data stream, representing the multimodal information contained at each time point t, such as speech feature S t 、Facial expression features F t and physiological characteristics t ;

[0216] The feature unit is also used to adopt online feature selection algorithms, such as online LASSO or stochastic gradient descent (SGD), according to the real-time feature vector X t Calculate the optimized feature set

[0217] Among them, the optimized feature set is:

[0218]

[0219] Among them, L(X t ,Y t ) is the loss function used to measure the real-time feature vector X t Predicted label Y t (e.g., instant speech performance scoring); λ is a regularization parameter used to control feature sparsity; ‖X t ‖1 is the L1 norm, which is used to facilitate feature selection.

[0220] The feature unit of this embodiment can adapt to the user's state and environmental changes by using an online feature selection algorithm, making the extracted features more representative.

[0221] Initial unit, used to detect the situation through the preset situation perception model H(.|C t ), using the probabilistic graphical model (PGM) or conditional random field (CRF) to calculate the given optimized feature set The most appropriate feedback strategy A under the conditions of the current situation characteristics t The probability distribution of the user's most suitable feedback strategy is obtained Among them, the current situation feature refers to the speech environment feature of the speech data;

[0222] The update unit is used to use the policy iteration method in reinforcement learning, such as the Actor-Critic algorithm, to update the initial probability distribution according to the user's feedback action and the learning rate α Perform strategy iteration and update to obtain the first probability distribution π of the most appropriate feedback strategy for the user in a given situation t+1 (a|s);

[0223] Among them, the initial probability distribution is:

[0224]

[0225] The updating process of probability distribution can be expressed as:

[0226]

[0227] Among them, C t Indicates current situational characteristics, such as audience reaction, speech topic, etc.;

[0228] s is the state vector (containing and C t ), a is the feedback action, α is the learning rate; A t is the advantage function based on the current strategy, reflecting the additional benefit of taking action a relative to the current strategy, that is, A t =Q t (s,a)-V t (s), and Q t (s,a) is the value of the state-action pair, V t (s) is the value of state s; π t (a|s) means at time t, facing state s (including and C t ), the probability of taking action a (i.e., strategy), and π t+1 (a|s) is the updated strategy (i.e., the first probability distribution); is the policy gradient term, which represents the derivative of the policy with respect to its parameters and is used to guide how the policy should be adjusted in the direction of gain;

[0229] Among them, feedback action a refers to the reaction or behavior change made by the user after receiving the feedback provided by the system. It can be the user's adoption, modification or rejection of a feedback plan, or the improvement or regression shown by the user during the training process. For example, through voice recognition technology to monitor changes in speech speed, facial expression recognition to determine the richness of body language, physiological signal analysis to infer the degree of tension, etc., it can be judged whether the user has actually increased the speaking speed after receiving the suggestion to increase the speaking speed, or whether the user has reduced repeated expressions after receiving the prompt to reduce redundant information, and the degree of adoption of the suggestion. These feedback actions are an important basis for the system to evaluate the effectiveness of feedback strategies;

[0230] In addition, feedback actions a can be divided into several categories, mainly including but not limited to:

[0231] 1. Active adoption: Users implement feedback suggestions, such as improving speaking speed, adding body language, adjusting volume, etc.

[0232] 2. Passive resistance: The user ignores the feedback and does not make any changes.

[0233] 3. Partial adoption: The user only partially follows the suggestions, such as improving the speaking speed without changing the volume.

[0234] 4. Proactive Adjustment: Users reflect on themselves based on feedback and make improvements beyond the suggested scope.

[0235] Secondly, the learning rate α is the speed at which the system adjusts the feedback strategy, which determines the system's sensitivity to user feedback actions, that is, the speed at which the system updates the feedback strategy. If the learning rate α is too high, the system will be overly dependent on the most recent feedback actions, ignoring long-term trends, resulting in an unstable training process; if it is too low, the system may be unresponsive and unable to adjust the strategy in time to adapt to changes in user needs. Therefore, the setting of the learning rate α should take into account the user's learning ability and the stability of the system. It can be initially set through experiments or empirical formulas, and then dynamically adjusted according to actual conditions;

[0236] In addition, the dynamic adjustment of the learning rate α should follow the following principles:

[0237] 1. Adaptive adjustment: The learning rate should be adjusted dynamically according to the user's progress speed and feedback effect. If the user makes significant progress, the learning rate can be appropriately reduced, otherwise it can be increased.

[0238] 2. User characteristic matching: Consider the user's learning ability and preferences, and provide personalized learning rate settings for different users.

[0239] 3. Environmental considerations: The learning rate may need to be adjusted in different situations. For example, in a high-stress environment, the learning rate should be appropriately reduced to prevent users from feeling overly burdened.

[0240] In order to ensure the rationality of the iterative update of the strategy, it is also necessary to set up an evaluation mechanism to regularly check whether the update of the feedback strategy has effectively promoted the improvement of the user's eloquence. This may include regularly collecting users' satisfaction evaluations on the feedback plan and measuring the improvement of users' eloquence level through objective tests.

[0241] Specifically include:

[0242] 1. Quantitative analysis: Quantify the changes in users’ expressive abilities through the user growth model, such as data on fluency, emotional expression, and cognitive load.

[0243] 2. User satisfaction survey: Collect users’ subjective evaluation of feedback strategies and understand users’ acceptance and satisfaction with feedback.

[0244] 3. Skill level comparison: Compare the user’s desired skill level with their current skill level to evaluate the contribution of feedback strategies to skill improvement.

[0245] 4. Long-term tracking: Observe the user’s skill growth curve over a long period of time to verify the long-term effect of strategy iteration.

[0246] The initial unit and update unit of this embodiment generate an initial probability distribution by combining the user's optimized feature set and the current situation features. This distribution reflects which feedback strategies are more likely to be suitable for the user in the current situation, and generates a set of strategies by calculating the probability distribution, so that each strategy has its probability of being selected. This diversity enables the system to be more flexible when facing complex and changeable user behaviors and situations. In addition, the introduction of current situation features enables the system to perceive and understand the specific environment or conditions in which the user is located. Since the same user may require different feedback methods in different situations, this is crucial to determining the appropriate feedback strategy. By considering the user's feedback actions, the user's acceptance of the current feedback strategy can be understood in real time, and the introduction of the learning rate can both respond quickly to user changes and prevent instability in the update of the probability distribution;

[0247] Moreover, the first probability distribution describes how to adjust the probability of taking the same action in the future according to the advantages generated by the current strategy, so that the strategy gradually tends to those behaviors that can bring higher returns or advantages. In this way, the system can learn better decision-making strategies under different states, which is helpful for the subsequent solution generation. In addition, reinforcement learning is integrated into the process of generating feedback strategies, so that the strategy can be continuously optimized according to the feedback effect, pursuing long-term performance improvement.

[0248] In addition, by defining feedback actions, setting learning rates, and evaluating the effect of feedback strategy updates, we can ensure that the strategy iteration updates are data-driven and meet the actual needs of users, thereby achieving more personalized and effective eloquence optimization training. These detailed operation methods enhance the operability and theoretical depth of the document, providing more solid support for practical applications.

[0249] In one embodiment, the training module 30 includes a feedback unit and a timely training unit;

[0250] The feedback unit is used to use the personalized feedback model PFM to calculate the user's historical speech data H hist and the user's current performance data in the speech data Through the first probability distribution π t+1 (a|s) performs continuous policy iteration to generate the first feedback solution R t ; For example, if a solution is particularly effective for a user in a specific situation, through learning, the system will be more inclined to recommend this solution in similar situations;

[0251] A timely training unit is used to train the first feedback scheme R t Train users and obtain real-time training data;

[0252] Among them, the model parameter Θ of the personalized feedback model PFM t+1 It is updated according to the user's learning rate and feedback loss function, and the updating process can be expressed as:

[0253]

[0254] The first feedback plan is:

[0255]

[0256] Among them, Θ t is the original model parameter of the personalized feedback model PFM, η is the learning rate, is the feedback R t With ideal feedback The loss function between .

[0257] It should be noted that in the process of updating the parameters of the personalized feedback model PFM, designing a reasonable loss function is the key to ensuring the model training effect and guidance. The design of the loss function must take into account the goal of personalized feedback, that is, to maximize the efficiency and effect of improving the user's eloquence. An effective loss function should be able to accurately measure the gap between feedback and ideal feedback, while reflecting the characteristics and needs of user learning.

[0258] In the personalized feedback model PFM, the design of the loss function needs to meet the following principles:

[0259] 1. Accuracy: The loss function should be able to accurately reflect the quality of the feedback. That is, the closer the feedback is to the ideal feedback, the smaller the value of the loss function should be.

[0260] 2. Personalization: Considering the learning ability and preferences of different users, the loss function should be able to adapt to different users, that is, adjust the weight or form of the loss function according to the user's learning rate.

[0261] 3. Dynamicity: As the user's skill level improves, the loss function should be able to adjust dynamically to adapt to the user's current ability level and learning stage.

[0262] 4. Interpretability: The loss function should be designed to be easy to understand and interpret to facilitate debugging and optimizing the model.

[0263] Moreover, in actual design, the loss function may contain multiple components, such as prediction error, user satisfaction, skill improvement rate, etc. For example, a composite loss function can be designed that simultaneously considers the deviation between feedback and expected results, user acceptance of feedback, and the impact of feedback on user skill improvement.

[0264] In the document, a paragraph detailing the loss function design principles and specific implementation should be added after describing the personalized feedback model parameter update formula.

[0265] The following is an example of the process of designing a loss function:

[0266] In order to ensure the correct direction of model parameter updates, the design of the loss function must take into account both accuracy and personalization; a good loss function should be able to accurately measure the gap between feedback and ideal feedback while taking into account the user's learning characteristics. When designing a loss function, the following points should be considered:

[0267] 1. Ensure that the loss function accurately reflects the quality of the feedback, that is, when the feedback is closer to the ideal state, the value of the loss function should decrease accordingly.

[0268] 2. The loss function needs to be personalized and can be adjusted according to the user's learning rate and preferences to meet the needs of different users.

[0269] 3. As the user's skills improve, the loss function should be adjusted dynamically to ensure that the feedback strategy is always targeted at the user's current ability level and learning stage.

[0270] 4. The design of the loss function should be transparent and explainable to facilitate model debugging and optimization.

[0271] The training module 30 of this embodiment generates a first feedback scheme based on historical data and current performance data, which can more comprehensively evaluate the user's eloquence performance; by comparing the user's historical progress with the current performance, the first feedback scheme can accurately fit the user's personal characteristics and current status, and then accurately point out the user's strengths and weaknesses in expression. In addition, the parameters of the model are updated according to the user's learning rate and feedback loss function, so that the personalized feedback model can better adapt to the user's learning characteristics, ensuring the personalization and effectiveness of the feedback scheme;

[0272] In addition, by continuously receiving new data and updating model parameters online, the system can quickly adapt to the speaker's real-time performance and environmental changes.

[0273] In one embodiment, the solution module 40 includes a model unit, an expression unit, a target subunit, a path subunit, a stage subunit, a development subunit, a skill unit, a mapping subunit, an adjustment subunit, an optimization unit, a visualization unit, and a configuration unit; wherein the model unit is a process of constructing a user growth model, the expression unit is a process of obtaining expression data, the target subunit, the path subunit, the stage subunit, and the development subunit are processes of generating personalized development suggestions, the skill unit, the mapping subunit, and the adjustment subunit are processes of generating an eloquence expression optimization solution, the optimization unit is a process of training users in eloquence expression and optimizing the solution according to the eloquence expression optimization solution, the visualization unit is a process of realizing data visualization, and the configuration unit is a process of performing privacy protection and personalized configuration, specifically:

[0274] A model unit, used to establish a fluency scoring model by calculating the sentence length and pause ratio of preset data;

[0275] The model unit is also used to establish an emotion expression quantification module by quantifying the emotion expression intensity and diversity of preset data;

[0276] The model unit is also used to establish a cognitive load analysis module according to the coefficient of the power ratio of each frequency band in the preset data;

[0277] The model unit is also used to form a user growth model consisting of a fluency scoring model, an emotional expression quantification module, and a cognitive load analysis module.

[0278] The expression unit is used to calculate the fluency score model in the user growth model using natural language processing technology to calculate the number of words per minute (speech speed), average sentence length, pause duration and other indicators in the real-time training data to obtain the fluency value S fluency ;

[0279] The expression unit is also used to quantify the intensity and diversity of emotional expression in real-time training data using the emotional expression quantification module in the user growth model, such as obtaining the emotional expression value S through the frequency of emotional words and the range of tone changes. emotion ;

[0280] The expression unit is also used to calculate the cognitive load value Y according to the cognitive load analysis module in the user growth model by extracting the power ratio of α, β, θ and δ bands based on the spectrum analysis of EEG signals, and calculating the real-time training data. cogLoad ;

[0281] Expression unit, also used by the fluency value S fluency , emotional expression value S emotion and cognitive load value Y cogLoad constitutes expression data;

[0282] Among them, the fluency value is:

[0283]

[0284] The emotional expression values ​​are:

[0285]

[0286] The cognitive load values ​​are:

[0287]

[0288] Among them, w1, w2 and w3 are weighting factors; σ SentenceLength and σ PauseRatio are the standard deviation of sentence length and the standard deviation of pause ratio, which are used for normalization; Words represents vocabulary, Time represents the time occupied by vocabulary, SentenceLengthMean represents the average sentence length, and PauseRatio represents the pause ratio;

[0289] f i is the frequency of emotion category i, EmotionIntensity i It corresponds to the intensity of the emotion;

[0290] β j is the coefficient of the power ratio of each frequency band, ∈ is the error term, PowerRatio j is the power ratio of each frequency band.

[0291] In the expression unit of this embodiment, fluency is directly related to whether the user can convey his or her meaning clearly and coherently. By calculating the sentence length and pause ratio of the real-time training data, this indicator can be quantified through the established fluency scoring model, thereby more objectively evaluating the user's eloquence fluency level. By quantifying the intensity and diversity of emotional expression of real-time training data through the emotional expression quantification model, the user's emotional expression ability can be objectively evaluated, thereby enabling the user to better convey his or her intentions and feelings and enhance the effect of the speech. The coefficient of the power ratio of the data in each frequency band reflects the processing and activity of the user's brain on different information frequency bands during the expression process. Therefore, through the cognitive load analysis module, the cognitive load status of the user when processing information and expressing ideas can be revealed, providing a scientific basis for subsequent feedback and guidance.

[0292] The target subunit is used to construct a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load according to the expression data;

[0293] The path subunit is used to solve the multi-objective optimization problem and obtain the personalized training path P * ;

[0294] The stage subunit is used to plan long-term development goals and milestones based on the user's current status and target status using a dynamic programming method, thereby establishing a stage-by-stage comprehensive value G that takes into account skill growth. longTerm ;

[0295] Development subunit for training according to individualized training path P * and the phased comprehensive value G longTerm Generate personalized development recommendations;

[0296] Among them, the personalized training path is:

[0297]

[0298] The comprehensive value of the phase is:

[0299]

[0300] Among them, U fluency and U emotion is the measure of the improvement of eloquence fluency and emotional expression by the training path, U cogLoad is the expected increase in cognitive load; W is the weight used to balance cognitive load and skill improvement; P represents the training path;

[0301] C is the cost function, S t is the state of stage t, A t is the action taken and T is the time period required to achieve the final goal.

[0302] The target subunit, path subunit, stage subunit and development subunit of this embodiment can simultaneously consider the optimization needs of these three aspects by constructing a multi-objective optimization problem about eloquence expression, emotional expression and cognitive load, avoiding the problem of focusing on only a single indicator in the traditional single-objective optimization method, and ensuring that the established personalized training path can better meet the personalized development of users; by comparing the user's current status and target status, the user's strengths and weaknesses in skills, as well as the aspects that need further improvement can be clearly identified, so personalized development suggestions are generated according to the personalized training path and the stage-by-stage comprehensive value, and the personalized training path learning methods and strategies can be adjusted in time to ensure the effectiveness of the personalized development suggestions.

[0303] Skill units are used to calculate the expected skill level of the user according to the value S desired and the user's current skill level value S in the real-time training data current , generate feedback strategy value Δπ;

[0304] A mapping subunit is used to map the feedback strategy value Δπ to the personalized development suggestion according to the mapping rule; for example, if the feedback strategy value Δπ is positive, it may mean that the current personalized development suggestion is too conservative and the user needs more challenging or more intensive exercises; on the contrary, if the feedback strategy value Δπ is negative, it may be necessary to reduce the difficulty of the exercise or provide more supportive feedback; wherein the feedback strategy value Δπ includes the gain K of the proportional controller p , the gain K of the integral controller i and controls the gain K of the derivative controller d ;

[0305] The adjustment subunit is used to adjust the difficulty coefficient of the training task, the detail of the feedback information, the incentive intensity of the reward system, etc. according to the numerical value and sign of the feedback strategy value according to the adjustment algorithm to obtain the eloquence expression optimization plan; specifically: if the gain K of the proportional controller p If the deviation is large, the direct intensity of the personalized development suggestion is adjusted according to the deviation; through the gain K of the integral controller i Accumulate past deviations to determine whether longer-term strategy adjustments are needed, such as adjusting the difficulty of the course plan; control the gain K of the derivative controller d Respond quickly to changes in user skill levels, such as increasing difficulty to maintain challenge as users progress rapidly.

[0306] Among them, the generation feedback strategy value is:

[0307]

[0308] Among them, K p , K i and Kd are the gain of the proportional controller, the gain of the integral controller and the gain of the differential controller respectively; S desired is the expected skill level, S current is the current skill level value, and S is reduced by adjusting the feedback strategy π desired and S current The difference between them is used to obtain the final feedback strategy value Δπ.

[0309] It should be noted that the above mapping rules and adjustment algorithms are specifically as follows:

[0310] (I) Mapping rules

[0311] The proportional gain (P) directly affects the instant adjustment strength of the personalized development suggestions, that is, the strength of the suggestions. If the feedback strategy value is positive, it means that the user's current training intensity is not enough to achieve the expected skill level. At this time, the load intensity of the personalized development suggestions should be increased to encourage the user to accept more challenging tasks. On the contrary, if the feedback strategy value is negative, it means that the user may face too much pressure, and the training difficulty should be reduced and more supportive feedback should be given.

[0312] Integral gain (I) focuses on historical performance and is used to adjust the long-term strategy of personalized development suggestions. Integral gain helps the system understand the user's learning curve by accumulating past deviations and decides whether the difficulty of the course plan needs to be adjusted. For example, if the user fails to reach the goal in the past period of time, the integral gain will gradually increase, prompting the system to provide more coaching and support.

[0313] Differential gain (D) focuses on the prediction of future states, ensuring that personalized development recommendations can quickly respond to changes in the user's skills. Differential gain is adjusted according to the speed at which the user's skill level changes. If the user progresses rapidly, the differential gain will increase so that the task difficulty can be increased in time to maintain the challenge; conversely, if the progress is slow, the differential gain will be reduced to prevent the user from losing confidence due to excessive task difficulty.

[0314] (II) Adjustment algorithm

[0315] The proportional gain (P) directly affects the immediate adjustment strength of the personalized development suggestions, the integral gain (I) focuses on historical performance and is used to adjust the long-term strategy of personalized development suggestions, and the differential gain (D) focuses on the prediction of future states to ensure that personalized development suggestions can quickly respond to user skill changes;

[0316] Furthermore, the algorithm needs to be adjusted to take into account the gap between the user’s current skill level and the desired skill level, as well as the user’s past performance. The core of the algorithm is to calculate the feedback strategy value, which is determined by the difference between the desired skill level value and the current skill level value.

[0317] The skill unit, mapping subunit and adjustment subunit of this embodiment can provide specific directions for adjusting personalized development suggestions by mapping feedback strategy values ​​to personalized development suggestions; wherein the gain of the proportional controller focuses on the immediate adjustment of the current state, the gain of the integral controller considers the cumulative effect of the historical state, and the gain of the differential controller focuses on the prediction and trend of the future state. Such multi-dimensional adjustment can comprehensively consider the user's current performance, historical progress and future potential, thereby formulating a more comprehensive and effective eloquence expression optimization plan;

[0318] In addition, the mapping and adjustment of feedback strategy values ​​are key links in ensuring the accuracy and practicality of the eloquence expression optimization plan; the proportional gain, integral gain and differential gain correspond to the load intensity, difficulty and response speed of personalized development suggestions respectively; through the formulation of clear rules and algorithms, effective mapping and adjustment can be achieved, and the feedback strategy value can be ensured to accurately reflect the user's eloquence level and training needs.

[0319] An optimization unit, which is used to train users in eloquence according to the eloquence optimization plan and continuously collect user feedback and performance data;

[0320] The optimization unit is also used to further refine the eloquence expression optimization scheme according to user feedback and performance data, and to optimize the calculation parameters (i.e., K p , K i and K d ) to make fine adjustments so that the eloquence optimization plan can be dynamically optimized as the user progresses and the environment changes, and an optimized eloquence optimization plan can be obtained; for example, if the user responds well to positive feedback and makes significant progress, the proportion of positive incentives can be increased; if the user shows an increased cognitive load when encountering a specific type of challenge, the plan can be adjusted to reduce the frequency of such tasks or provide auxiliary tools to reduce the burden;

[0321] The optimization unit is also used to analyze users' responsiveness and adoption rates to different feedback strategies, evaluate the effectiveness of strategies through methods such as A / B testing, and adjust the strategy generation logic accordingly to ensure continuous improvement and personalized matching of feedback mechanisms and eloquence expression optimization plans.

[0322] The visualization unit is used to construct a multi-dimensional achievement indicator system based on user feedback during training, including but not limited to skill scores, cognitive load improvement rate and training time, and use scatter plots, line graphs, etc. to visualize the multi-dimensional achievement indicator system.

[0323] Configuration unit, used to protect data security, specifically: encrypt user data using Advanced Encryption Standard (AES) or other encryption technologies to ensure data security during transmission and storage; and implement strict access control policies to ensure that only authorized users can access their personal data;

[0324] The configuration unit is also used to set up data management, specifically: to set up personal preferences for the first feedback plan and eloquence expression optimization plan in the plan according to user preferences, such as the style, difficulty and content of eloquence training; and to allow users to manage their own privacy settings through the user interface, including data sharing and usage preferences.

[0325] Configuration units are also used to track data, specifically: using logs to record data usage, providing transparency so that users can track how their data is used; providing clear privacy policy explanations for data to ensure that users understand how their data is processed and protected.

[0326] Overall, this embodiment has the following beneficial effects:

[0327] The present invention can understand the possible behavior patterns of users in different situations by performing probability distribution calculations on the feedback strategies of users in preset situations. Therefore, the first probability distribution provides users with quantitative information about behavior patterns, which helps to formulate more effective strategies. Combining the first probability distribution with the user's current performance data, the first feedback plan can be customized for the user more accurately based on the user's current performance and targeting the weak points of eloquence, thereby obtaining the real-time training data fed back by the user after training; for eloquence training, these real-time training data are valuable resources. By analyzing these data, problems and deficiencies in the training process can be discovered at the first time. Therefore, an eloquence expression optimization plan is generated based on the user's expected skill level value and the user's current skill level value in the real-time training data. The user's expected skill level value and the current skill level value can be compared, and then a clear understanding of the user's gaps and the areas in which further improvement is needed can be obtained. Therefore, the eloquence expression optimization plan can accurately locate the user's weak links, provide targeted improvement suggestions for the user, and conduct further tracking training for the user;

[0328] In addition, by combining the user's historical performance and real-time training data, the training content and difficulty are dynamically adjusted to provide users with a learning path that best suits their current ability level and progress rate. This adaptability ensures the continued challenge of training, avoids overload or inefficient learning, and promotes long-term stable development. When processing user data, the system uses advanced encryption technology and strict data access control mechanisms to ensure the security and privacy of user information. At the same time, users are allowed to personalize data sharing permissions, which increases users' trust in the system and willingness to use it.

[0329] Embodiment three:

[0330] An embodiment of the present invention 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 method for optimizing eloquence based on instant feedback and tracking;

[0331] Wherein, the eloquence optimization method based on instant feedback and tracking, 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 such an 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. Wherein, 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, etc. 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.

[0332] The above are preferred embodiments of the present invention. It should be noted 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 method for optimizing eloquence based on instant feedback and tracking, characterized in that: include: Get the user's speech data; According to the speech data, the probability distribution of the feedback strategy adapted by the user in a preset situation is calculated to obtain a first probability distribution; wherein the first probability distribution is the probability distribution of the most suitable feedback strategy obtained through iterative updating based on the feedback action of the user in a specific situation and a preset learning rate; specifically: feature extraction is performed on the speech data to obtain an optimized feature set; the probability distribution of the feedback strategy is calculated under the given conditions of the optimized feature set and the current situation feature to obtain the initial probability distribution of the feedback strategy adapted by the user; wherein the current situation feature refers to the speech environment feature of the speech data; the initial probability distribution is iteratively updated according to the feedback action and learning rate of the user to obtain the first probability distribution of the feedback strategy adapted by the user; The first probability distribution is combined with the user's current performance data in the speech data to generate a first feedback scheme, and the user is trained according to the first feedback scheme to obtain real-time training data; wherein the first probability distribution is combined with the user's current performance data in the speech data to generate the first feedback scheme, specifically: using a personalized feedback model, according to the first probability distribution, the user's historical speech data and the user's current performance data in the speech data, to generate the first feedback scheme; wherein the model parameters of the personalized feedback model are obtained by iteratively updating according to the user's learning rate and feedback loss function; An eloquence expression optimization plan is generated according to the expected skill level value of the user and the current skill level value of the user in the real-time training data, specifically: the expressive ability of the real-time training data is quantified through a user growth model to obtain expression data; a multi-objective optimization problem is constructed according to the expression data, and personalized development suggestions are obtained by solving the problem; a feedback strategy value is generated according to the expected skill level value of the user and the current skill level value of the user in the real-time training data; and the eloquence expression optimization plan is generated according to the personalized development suggestions and the feedback strategy value.

2. The method for optimizing eloquence based on instant feedback and tracking as claimed in claim 1, characterized in that: The user growth model is specifically: Establish a fluency scoring model by calculating the sentence length and pause ratio of preset data; Establishing an emotion expression quantification module by quantifying the emotion expression intensity and diversity of the preset data; Establishing a cognitive load analysis module according to the coefficient of the power ratio of each frequency band in the preset data; The user growth model is composed of the fluency scoring model, the emotion expression quantification module and the cognitive load analysis module.

3. The method for optimizing eloquence based on instant feedback and tracking as claimed in claim 1, characterized in that: A multi-objective optimization problem is constructed based on the expression data, and the solution is used to obtain personalized development suggestions, specifically: constructing a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load according to the expression data; Solving the multi-objective optimization problem to obtain a personalized training path; Calculate a phased comprehensive value considering skill growth according to the current state and target state of the user; The personalized development suggestion is generated according to the personalized training path and the phased comprehensive value.

4. The method for optimizing eloquence based on instant feedback and tracking as claimed in claim 1, characterized in that: The eloquence expression optimization scheme is generated according to the personalized development suggestions and the feedback strategy value, specifically: Mapping the feedback strategy value to the personalized development suggestion; wherein the feedback strategy value includes proportional gain, integral gain and differential gain; According to the numerical value of the feedback strategy value, the load intensity, difficulty and response speed of the personalized development suggestion are adjusted according to the proportional gain, integral gain and differential gain respectively to obtain the eloquence expression optimization plan.

5. The method for optimizing eloquence based on instant feedback and tracking as claimed in claim 1, characterized in that: The feedback strategy value is specifically: Among them, K p , K i and K d are the gains of the proportional controller, the integral controller, and the differential controller, respectively. desired is the expected skill level value, S current is the current skill level value.

6. An eloquence optimization device based on instant feedback and tracking, characterized in that: Includes data module, probability module, training module and solution module; Wherein, the data module is used to obtain the user's speech data; The probability module is used to calculate the probability distribution of the feedback strategy adapted by the user in a preset situation based on the speech data, and obtain a first probability distribution; wherein the first probability distribution is the probability distribution of the most suitable feedback strategy obtained through iterative updating based on the feedback action of the user in a specific situation and a preset learning rate; specifically: feature extraction is performed on the speech data to obtain an optimized feature set; the probability distribution of the feedback strategy is calculated under the given conditions of the optimized feature set and the current situation feature, and the initial probability distribution of the feedback strategy adapted by the user is obtained; wherein the current situation feature refers to the speech environment feature of the speech data; the initial probability distribution is iteratively updated according to the feedback action and learning rate of the user, and the first probability distribution of the feedback strategy adapted by the user is obtained; The training module is used to combine the first probability distribution with the user's current performance data in the speech data to generate a first feedback scheme, and train the user according to the first feedback scheme to obtain real-time training data; wherein, the first probability distribution is combined with the user's current performance data in the speech data to generate the first feedback scheme, specifically: using a personalized feedback model, according to the first probability distribution, the user's historical speech data and the user's current performance data in the speech data, to generate the first feedback scheme; wherein the model parameters of the personalized feedback model are obtained by iteratively updating according to the user's learning rate and feedback loss function; The solution module is used to generate an eloquence expression optimization solution based on the expected skill level value of the user and the current skill level value of the user in the real-time training data, specifically: quantifying the expressive ability of the real-time training data through a user growth model to obtain expression data; constructing a multi-objective optimization problem based on the expression data, and solving it to obtain personalized development suggestions; generating a feedback strategy value based on the expected skill level value of the user and the current skill level value of the user in the real-time training data; generating the eloquence expression optimization solution based on the personalized development suggestions and the feedback strategy value.

7. The eloquence optimization device based on instant feedback and tracking as claimed in claim 6, characterized in that: The user growth model is specifically: Establish a fluency scoring model by calculating the sentence length and pause ratio of preset data; Establishing an emotion expression quantification module by quantifying the emotion expression intensity and diversity of the preset data; Establishing a cognitive load analysis module according to the coefficient of the power ratio of each frequency band in the preset data; The user growth model is composed of the fluency scoring model, the emotion expression quantification module and the cognitive load analysis module.

8. The eloquence optimization device based on instant feedback and tracking as claimed in claim 6, characterized in that: The solution module includes a target subunit, a path subunit, a stage subunit, and a development subunit; Wherein, the target subunit is used to construct a multi-objective optimization problem including eloquence expression, emotional expression and cognitive load according to the expression data; The path subunit is used to solve the multi-objective optimization problem and obtain a personalized training path; The stage subunit is used to calculate a stage-by-stage comprehensive value considering skill growth according to the current state and target state of the user; The development subunit is used to generate the personalized development suggestion according to the personalized training path and the stage-by-stage comprehensive value.

9. The eloquence optimization device based on instant feedback and tracking as claimed in claim 6, characterized in that: The scheme module includes a mapping subunit and an adjustment subunit; Wherein, the mapping subunit is used to map the feedback strategy value to the personalized development suggestion; wherein the feedback strategy value includes proportional gain, integral gain and differential gain; The adjustment subunit is used to adjust the load intensity, difficulty and response speed of the personalized development suggestions according to the numerical value of the feedback strategy value, the proportional gain, the integral gain and the differential gain, so as to obtain the eloquence expression optimization plan.

10. The eloquence optimization device based on instant feedback and tracking as claimed in claim 6, characterized in that: The feedback strategy value is specifically: Among them, K p , K i and K d are the gains of the proportional controller, the integral controller, and the differential controller, respectively. desired is the expected skill level value, S current is the current skill level value.

11. A storage medium, characterized in that: The storage medium stores a computer program, which is called and executed by a computer to implement any one of the eloquence optimization methods based on instant feedback and tracking as claimed in claims 1 to 5.

Images