For oral situational dialogue simulation system

Abstract

This invention provides a spoken language situational dialogue simulation system, relating to the field of machine question-answering technology. It includes a spoken language proficiency data acquisition module: collecting user spoken language proficiency feature data; an ability calculation module: preprocessing the collected spoken language proficiency feature data to obtain a spoken language ability index; a first speech rate calibration module: setting an initial speech rate benchmark and obtaining a speech rate calibration value; a trend analysis module: calculating the change value between the spoken language ability index corresponding to adjacent fixed time windows; and a second speech rate calibration module: calculating the average change value of spoken language ability within a judgment period, determining whether a secondary adjustment of the speech rate calibration value is needed, and selecting to apply either an upward or downward speech rate adjustment coefficient to adjust the speech rate calibration value. This invention avoids user frustration caused by excessively fast speech rate and also avoids limiting user improvement potential by excessively slow speech rate, thus allowing each learner to conduct efficient spoken language training at their most suitable pace.

Description

Technical Field

[0001] This invention relates to the field of machine question-answering technology, specifically to a spoken situational dialogue simulation system. Background Technology

[0002] In oral situational dialogue simulation training systems, the ability to provide different speaking speeds is crucial to the training process: an appropriate speaking speed helps learners gradually improve their listening comprehension and speaking rhythm within a comprehensible input range; too fast a speaking speed can cause cognitive overload and frustration, leading to training interruptions or reduced effectiveness; while too slow a speaking speed fails to provide sufficient challenge, easily causing learners to become complacent and limiting their oral progress. Therefore, providing a suitable dialogue simulation speaking speed can improve training efficiency and learner engagement.

[0003] In existing technology, CN116092352B discloses a dialogue training method. This method includes: acquiring dialogue audio from a target user; extracting speech rate information from the dialogue audio; stopping the current round of dialogue training when an abnormality in the speech rate information is detected; generating a corresponding target interruption statement for voice playback; and starting the next round of dialogue training based on the target interruption statement. In this way, by continuously detecting speech rate information, when an abnormality in the speech rate information is detected, a target interruption statement can be quickly generated and played for voice playback to interrupt the training, creating a tense atmosphere and achieving a better training effect.

[0004] However, the aforementioned existing technologies use a fixed speaking speed or rely on simple rules for rough adjustment. They cannot dynamically calibrate the initial speaking speed based on the user's ability index, and they lack a periodic analysis and secondary adjustment mechanism for changes in oral ability. Therefore, the speaking speed often does not match the user's actual level: users with strong oral skills will find the speaking speed too slow and lacking in challenge, thus failing to improve, while users with weak oral skills will be discouraged by the speaking speed, resulting in poor training effects. Moreover, they cannot respond in a timely manner to the user's progress or regression. Overall, they have low personalization and poor adaptability, making it difficult to achieve truly individualized and efficient training.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a spoken language situational dialogue simulation system to solve the problems mentioned in the background art. This invention avoids user frustration caused by excessively fast speaking speed, and also avoids limiting user improvement potential by speaking too slowly, thus allowing each learner to conduct efficient spoken language training at a pace most suitable for them.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] For oral situational dialogue simulation systems, including:

[0009] Oral proficiency data acquisition module: Collects user oral proficiency feature data, including pause frequency, grammatical error rate and vocabulary diversity, and each oral proficiency feature data is acquired in real time from the simulation;

[0010] Ability calculation module: preprocesses the collected oral proficiency feature data and establishes a mathematical conversion model. The mathematical conversion model adopts a machine learning algorithm and includes an input layer, a hidden layer and an output layer. The above sets of data are input into the input layer of the mathematical conversion model, and the oral proficiency index is obtained through the output layer.

[0011] First speech rate calibration module: Set an initial speech rate benchmark, construct a speech rate calibration formula, retrieve the oral communication ability index, use the oral communication ability index as a calibration coefficient, and simultaneously input the calibration coefficient and the initial speech rate benchmark into the speech rate calibration formula to obtain the speech rate calibration value.

[0012] Trend Analysis Module: Set a fixed time window t, obtain the oral proficiency index of the corresponding stage within the fixed time window t, collect N consecutive fixed time windows, and take the N consecutive fixed time windows as a judgment period, obtain the oral proficiency index of adjacent fixed time windows within a judgment period, and calculate the change value between the oral proficiency index of adjacent fixed time windows.

[0013] The second speech rate calibration module calculates the average change in spoken ability within a judgment period, sets a speech rate increase coefficient and a speech rate decrease coefficient, and determines whether the speech rate calibration value needs to be adjusted a second time based on the average change in spoken ability. If no second adjustment is needed, the speech rate calibration value output by the first speech rate calibration module continues to be used. When it is determined that the speech rate calibration value needs to be adjusted a second time, the system selects to call the speech rate increase coefficient or the speech rate decrease coefficient to adjust the speech rate calibration value.

[0014] Furthermore, the data collection process for each oral proficiency feature in the oral proficiency data collection module is as follows:

[0015] Pause frequency acquisition: Real-time monitoring of the user's spoken output stream, identification and recording of silence intervals in the user's speech, setting a pause threshold, detecting short pauses between continuous speech, marking nodes with short pause times greater than or equal to the pause threshold, and calculating the ratio of the number of pause nodes to the total speaking time as the pause frequency.

[0016] Syntax error rate collection: Establish a grammar rule base, perform real-time text analysis based on dialogue content, the system identifies grammatical errors, and calculates the percentage of error events to total discourse units as the grammar error rate;

[0017] Lexical diversity collection: Real-time dynamic evaluation is performed through type-tag ratio calculation, and the proportion of different words to the total vocabulary is calculated as lexical diversity data.

[0018] Furthermore, the preprocessing of spoken language proficiency feature data in the capability calculation module includes data cleaning, standardization, and feature integration: data cleaning includes detecting and correcting outliers or missing values ​​in pause frequency, grammatical error rate, and lexical diversity, and identifying invalid data points through statistical methods for imputation or removal; the standardization stage converts features of different dimensions into numerical ranges between 0 and 1 using a normalization algorithm to eliminate scale differences; in the feature integration stage, the cleaned and standardized spoken language proficiency feature data are combined into structured feature vectors for processing in the input layer of the mathematical transformation model.

[0019] 4. The oral dialogue simulation system according to claim 1, characterized in that: the formula used in the ability calculation module to calculate the oral ability index is:

[0020]

[0021] in:

[0022] SAI stands for Spoken English Proficiency Index;

[0023] LD represents lexical diversity data;

[0024] PF represents the pause frequency;

[0025] GER stands for syntax error rate;

[0026] w is a mathematical constant used to ensure that the formula holds true and to prevent the denominator from being zero, and w satisfies: w > 0;

[0027] The pause frequency, grammatical error rate, and lexical diversity data, after preprocessing, range between [0,1].

[0028] Furthermore, the training process of the mathematical transformation model is based on a supervised learning paradigm, using a historical dataset labeled with real spoken language ability levels, minimizing the mean squared error loss function between the predicted index and the real score, employing the Adam optimizer for backpropagation to iteratively update the weight parameters, and implementing an early stopping strategy during training to prevent overfitting.

[0029] Furthermore, the speech rate calibration formula in the first speech rate calibration module is:

[0030]

[0031] in:

[0032] This is the speech rate calibration value;

[0033] B represents the initial speech rate benchmark, indicating the system's preset default speech rate value, which is set based on the average level of multiple users and serves as the starting point for adjustment.

[0034] k is the sensitivity coefficient;

[0035] This serves as a reference threshold for spoken language proficiency.

[0036] It is an exponential function.

[0037] Furthermore, it also includes the process for obtaining the sensitivity coefficient k:

[0038] Historical dialogue data samples are collected, which include user actual speech rate adaptation feedback corresponding to different oral proficiency indices. A parameter optimization objective function is established based on a preset speech rate calibration formula. The objective function is to minimize the weighted sum of user discomfort score and speech rate matching error. The objective function is solved using the gradient descent algorithm under the constraint of k>0. The sensitivity coefficient k value with the best generalization is determined by cross-validation and embedded into the system.

[0039] Furthermore, in the trend analysis module, multiple fixed time windows t are marked sequentially as follows: Extract the spoken language proficiency index at the end of each window and record it as . For any pair of adjacent windows: The change value of its oral proficiency index is Obtained through difference operations:

[0040]

[0041] in: For the nth fixed time window, the oral proficiency index is used. For the (n+1)th fixed time window, the oral proficiency index is used.

[0042] The change value is used to quantify the instantaneous fluctuation direction and magnitude of a user's spoken fluency within adjacent time periods, and its meaning is as follows:

[0043] This indicates an improvement in the user's fluency in spoken conversation;

[0044] This indicates that the user's fluency in spoken conversation remains unchanged;

[0045] This indicates a decrease in the fluency of the user's spoken conversation.

[0046] Furthermore, the second speech rate calibration module calculates the average change in spoken ability over a judgment period as follows:

[0047]

[0048] in: denoted as the average change in oral communication ability, and N as the total number of fixed time windows within a judgment period.

[0049] Furthermore, the average change threshold for oral communication ability is set as follows: And satisfy ;

[0050] when At that time, the speech rate calibration value is determined using the formula: A second adjustment was made, among which The final speech rate after the second adjustment is given, where α is the speech rate adjustment coefficient. This indicates that the user's training status in the previous judgment period has shown significant improvement. Therefore, the speech rate in the system's simulated situational dialogue is further increased, and a maximum speech rate threshold is set. and satisfy ;

[0051] when When this occurs, it indicates that the user's training state in the previous judgment period is stable. Therefore, the system's simulated dialogue speed remains unchanged and continues to use the speed calibration value. The simulated speech rate of the situational dialogue serves as the basis for the next judgment cycle;

[0052] when At that time, the speech rate calibration value is determined using the formula: A second adjustment is made, where β is the speech rate reduction coefficient. This indicates that the user's training status in the previous judgment period was too difficult; therefore, the system's simulated speech rate in the situational dialogue is further reduced, and a minimum speech rate threshold is set. and satisfy .

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] This invention dynamically and intelligently adjusts the speaking speed according to the user's own fluency level, rather than using a fixed speaking speed. It assesses the user's true level by collecting data on the user's pause frequency, grammatical error rate, and vocabulary diversity in real time, and calibrates the initial speaking speed accordingly. At the same time, it can analyze the overall trend of the user's speaking ability at regular intervals. If the user makes significant progress, the speaking speed is appropriately increased to enhance the challenge; if the user finds it difficult, the speaking speed is promptly reduced to alleviate pressure. This dual calibration mechanism achieves personalized adaptive adjustment that conventional solutions cannot provide. It avoids the frustration caused by speaking too fast and the limitation of improvement caused by speaking too slow, thus allowing each learner to conduct efficient oral training at a pace that best suits them. Attached Figure Description

[0055] Figure 1 This is a block diagram illustrating the principle of the oral situational dialogue simulation system of the present invention.

[0056] Figure 2 This is a schematic diagram of the operation process of the oral situational dialogue simulation system of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0058] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0059] Example:

[0060] Please see Figures 1-2 The present invention provides the following technical solutions:

[0061] For oral situational dialogue simulation systems, including:

[0062] Oral proficiency data acquisition module: Collects user oral proficiency feature data, including pause frequency, grammatical error rate and vocabulary diversity, and each oral proficiency feature data is acquired in real time from the simulation;

[0063] The data collection process for each oral proficiency feature in the oral proficiency data collection module is as follows:

[0064] Pause Frequency Acquisition: The system monitors the user's spoken output stream in real time, carefully identifying and recording various silence intervals that occur during the user's speech. First, a reasonable pause threshold is set. Then, short pauses that occur between continuous speech are detected. Nodes with pause times that reach or exceed the threshold are clearly marked. Finally, the ratio of the total number of marked pauses to the user's overall speaking time is calculated, and this ratio is used as the current pause frequency.

[0065] Syntax error rate collection: The system establishes a complete grammar rule base. The construction process of this grammar rule base includes knowledge extraction from a large-scale corpus, obtaining common grammar patterns and error types, and processing the extracted grammar knowledge to form structured, machine-readable rule entries to support the accurate identification of grammar errors in real-time text analysis.

[0066] In the above data collection process, real-time text analysis first divides user speech into independent lexical units through word segmentation, and then uses grammatical analysis technology to determine the correctness of syntactic structure, thereby identifying grammatical errors. At the same time, the calculation of lexical diversity relies on word frequency statistics. After completing word segmentation, the system counts the number of occurrences of each word to accurately calculate the type-tag ratio.

[0067] The system performs real-time text analysis on user speech based on the dialogue content, automatically identifies grammatical errors, and calculates the percentage of units occupied by these grammatical error events relative to the total number of user speech units, using this as the grammatical error rate. On this basis, the system can further combine semantic analysis technology to determine whether there are logical contradictions or semantic ambiguities in the user's expression, thereby assisting in the comprehensive evaluation of the grammatical error rate and making the dimensions of dialogue quality analysis more comprehensive.

[0068] Lexical diversity collection: The system uses the type-tag ratio calculation method to dynamically evaluate users' vocabulary usage in real time. That is, it counts the ratio of the number of different types of words used by users in the conversation to the total number of words, and uses this ratio as data to measure lexical diversity.

[0069] The ability calculation module: After fully preprocessing the collected spoken language proficiency feature data, a mathematical transformation model is established. This mathematical transformation model adopts a machine learning algorithm, and its overall structure includes an input layer, several hidden layers, and an output layer. As an alternative implementation, the hidden layers of the mathematical transformation model can adopt a fully connected network structure in deep learning, which automatically extracts high-order interaction relationships between spoken language features through multi-layer nonlinear transformations to improve the accuracy of spoken language proficiency index prediction.

[0070] The preprocessing stage primarily involves cleaning, standardizing, and integrating the original spoken language feature data to eliminate noise and dimensional differences, ensuring that all features can participate in calculations on the same scale. After preprocessing, the system inputs the processed spoken language proficiency feature data into the input layer of the mathematical transformation model. The data propagates forward layer by layer within the model, undergoing nonlinear transformations and feature extraction in the hidden layers. Finally, the output layer calculates and outputs a comprehensive spoken language ability index to quantitatively characterize the user's current spoken language proficiency.

[0071] The preprocessing of spoken language proficiency feature data in the ability calculation module mainly consists of three sequential steps. The data cleaning step specifically involves: detecting and correcting outliers or missing values ​​in the three types of feature data: pause frequency, grammatical error rate, and lexical diversity; identifying invalid data points using statistical methods; and performing appropriate imputation or direct removal of these invalid data points based on the actual situation, thereby ensuring the authenticity and completeness of the input data. In the next standardization stage, feature data with different dimensions are uniformly transformed to a numerical range between 0 and 1 using a normalization algorithm, eliminating scale differences caused by different original units of measurement and ensuring the fairness and stability of the subsequent modeling process. Finally, in the feature integration stage, the cleaned and standardized spoken language proficiency feature data are organically combined into a structured feature vector. This feature vector is then fully passed to the input layer of the mathematical transformation model for further analysis and processing.

[0072] The formula used to calculate the oral proficiency index in the ability calculation module is as follows:

[0073]

[0074] in:

[0075] SAI (Spoken English Ability Index) is used to comprehensively quantify a user's current spoken English level. A higher SAI value indicates stronger spoken English ability, and vice versa.

[0076] LD (Lexical Diversity) data reflects the richness of a user's vocabulary. A higher LD indicates a broader vocabulary and more flexible expression.

[0077] PF represents the pause frequency, reflecting the percentage of pauses that indicate a lack of fluency in a user's speech. A higher PF indicates a user is more prone to pausing, has longer thought intervals, and exhibits poorer spoken fluency. Because it's in the denominator, PF is negatively correlated with SAI: a higher pause frequency corresponds to a lower spoken ability index.

[0078] GER (Grammatical Error Rate) reflects the proportion of grammatical errors in a user's speech. A higher GER indicates a weaker grasp of grammar and lower accuracy in expression. Also in the denominator, GER is negatively correlated with SAI (Spoken English Ability Index): a higher grammatical error rate corresponds to a lower spoken English ability index.

[0079] w is a mathematical constant used to ensure that the formula holds true and to prevent the denominator from being zero, and w satisfies: w > 0;

[0080] The pause frequency, grammatical error rate, and lexical diversity data, after preprocessing, range between [0,1].

[0081] In the training process of the mathematical transformation model, a supervised learning paradigm is used for model training. A historical dataset with labeled real spoken language ability levels is prepared. During training, the model continuously calculates the error between its predicted spoken language ability index and the actual scores in the dataset, and minimizes the mean squared error loss function between the predicted index and the actual scores as the optimization objective. To efficiently find the optimal model parameters, this embodiment uses the Adam optimizer for backpropagation calculation, iteratively updating the various weight parameters in the model to gradually reduce the value of the loss function. Simultaneously, to prevent overfitting during training—that is, the model excessively memorizing training data and losing its generalization ability to new data—the system implements an early stopping strategy during training. Once the performance on the validation set no longer improves, training is stopped promptly, thus ensuring that the model has good generalization performance.

[0082] The first speech rate calibration module sets an initial speech rate benchmark as the starting point for subsequent adjustments. It constructs a speech rate calibration formula to map the user's spoken language level to the speech rate adjustment amount. The system retrieves the current user's spoken language ability index from the ability calculation module and uses this index as the core calibration coefficient. The system simultaneously inputs this calibration coefficient and the previously set initial speech rate benchmark into the speech rate calibration formula. After calculation and conversion, a preliminary adjusted speech rate calibration value is obtained to guide the system in outputting a speech rate adapted to the current user's spoken language level.

[0083] The speech rate calibration formula is:

[0084]

[0085] in:

[0086] This is the speech rate calibration value, representing the output speech rate after initial calibration. A higher value indicates a faster speech rate, while a lower value indicates a slower speech rate. The purpose is to adjust the speech rate to match the user's current spoken language ability level.

[0087] B represents the initial speech rate benchmark, indicating the system's preset default speech rate value, set based on the average level of multiple users, and serving as the starting point for adjustments. and Positive correlation: The larger the value, the faster the output speech rate under the same conditions;

[0088] k is the sensitivity coefficient, and the value of k is obtained by optimizing historical data to ensure user comfort;

[0089] This serves as a reference threshold for spoken language ability, used to determine whether the current user's spoken language level is higher or lower than the baseline. When the index is positive, the speech rate is higher than the baseline; when At that time, the index was negative, and the speech rate was lower than the benchmark.

[0090] The purpose of using an exponential function is to allow the speech rate to increase rapidly but not indefinitely when the user's ability is much higher than the reference threshold (in conjunction with the subsequent maximum speech rate limit); and to allow the speech rate to decrease significantly but always remain greater than zero when the user's ability is much lower than the reference threshold.

[0091] The process of obtaining the sensitivity coefficient k:

[0092] Historical dialogue data samples are collected. These samples not only record the user's performance under different oral proficiency indices but also include corresponding user feedback on actual speech rate adaptation, such as the user's comfort level or discomfort level at the current speech rate. After obtaining these samples, the system establishes a parameter optimization objective function based on a preset speech rate calibration formula. This objective function is set to minimize the weighted sum of the user's discomfort score and the speech rate matching error, thereby balancing the user's subjective feelings and the objective speech rate matching effect overall. The system uses a gradient descent algorithm to solve the objective function under the constraint k > 0, iteratively updating the k value until the optimal solution is found. To ensure that the coefficient has good generalization ability and is not only effective for training data, the system also verifies the solution results through cross-validation. Finally, the sensitivity coefficient k value with the best generalization performance is selected and embedded into the system.

[0093] Trend Analysis Module: A fixed time window length, denoted as t, is defined. Within each fixed time window t, the system continuously collects and calculates the spoken language proficiency index at the end of that period, thereby obtaining the spoken language proficiency assessment result corresponding to the current time window. Then, the system continuously collects N such fixed time windows in chronological order and merges these N consecutive fixed time windows into a complete judgment period. Within a judgment period, the spoken language proficiency index corresponding to each pair of adjacent fixed time windows is sequentially acquired, and the spoken language proficiency indices of each pair of adjacent windows are compared. The difference between them is calculated; this difference is the change in the spoken language proficiency index between adjacent fixed time windows, reflecting the fluctuation of the user's spoken language proficiency within adjacent time periods.

[0094] In the trend analysis module, multiple fixed time windows t are labeled according to the sequence. Extract the spoken language proficiency index at the end of each window and record it as . For any pair of adjacent windows: The change value of its oral proficiency index is Obtained through difference operations:

[0095]

[0096] in: For the nth fixed time window, the oral proficiency index is used. For the (n+1)th fixed time window, the oral proficiency index is used.

[0097] The change value is used to quantify the instantaneous fluctuation direction and magnitude of a user's spoken fluency within adjacent time periods, and its meaning is as follows:

[0098] This indicates an improvement in the user's fluency in spoken conversation;

[0099] This indicates that the user's fluency in spoken conversation remains unchanged;

[0100] This indicates a decrease in the fluency of the user's spoken conversation.

[0101] The second speech rate calibration module calculates and obtains the average change in spoken language ability over a complete assessment period. This average change reflects the overall fluctuation trend of the user's spoken language level during that period. Speech rate adjustment coefficients are set to adjust the speech rate accordingly when the user's performance improves or regresses.

[0102] The system determines whether a secondary adjustment to the speech rate calibration value output by the first speech rate calibration module is needed based on the calculated average change in spoken language ability. If the average change is within a stable range, it indicates that the user's current training state is good and adapted to the current speech rate, and no secondary adjustment is needed. The system will continue to use the speech rate calibration value output by the first speech rate calibration module as the speech rate for subsequent dialogues. Conversely, when the system determines that a secondary adjustment to the speech rate calibration value is needed, the system will select to apply either an upward or downward adjustment coefficient based on the positive or negative direction and magnitude of the average change, adjusting the current speech rate calibration value upward or downward to make the dialogue speed more closely match the user's latest spoken language ability trends.

[0103] The process by which the second speech rate calibration module calculates and obtains the average change in spoken ability over a judgment period is as follows:

[0104]

[0105] in: The average change in spoken language ability is represented by , where N is the total number of fixed time windows within a given period. A larger value indicates a more significant improvement in the user's spoken language ability within that period; a smaller value, or even a negative value, indicates slow improvement or regression. The system uses this average change value as the core basis for secondary adjustments to the speaking speed to match the user's learning progress trend.

[0106] Set the average change threshold for oral proficiency as And satisfy ;

[0107] when At that time, the speech rate calibration value is determined using the formula: A second adjustment was made, among which The final speech rate after the second adjustment is given, where α is the speech rate increase coefficient. This indicates that the user's spoken language level has improved significantly in the previous evaluation period, demonstrating substantial progress in training. At this point, the system needs to perform a second adjustment on the current speech rate calibration value. Specifically, the adjustment method is to multiply the speech rate calibration value by the speech rate increase coefficient to obtain the final speech rate after the second adjustment, and set the maximum speech rate threshold as follows: and satisfy ;

[0108] when If the error message indicates that the user's spoken language level did not fluctuate significantly during the previous evaluation period, and the training state is stable, then the system does not need to adjust the speech rate again. Instead, it maintains the original speech rate calibration value and uses it directly as the simulated speech rate for the situational dialogue in the next evaluation period. This maintains the conversational rhythm that the user has already adapted to, and the system continues to use the original speech rate calibration value. The simulated speech rate of the situational dialogue serves as the basis for the next judgment cycle;

[0109] when At that time, the speech rate calibration value is determined using the formula: A second adjustment is performed, where β is the speech rate reduction coefficient. This indicates a significant decline in the user's spoken language proficiency during the previous evaluation period, suggesting the current dialogue is too difficult and the user struggles to adapt to the existing speech rate. The system then needs to adjust the speech rate calibration value by multiplying it by the speech rate reduction coefficient to obtain the final adjusted speech rate. This operation further reduces the simulated speech rate in the situational dialogue to decrease the user's cognitive load and help them better keep up with the conversation. Simultaneously, the system sets a minimum speech rate threshold, requiring the adjusted final speech rate not to fall below this threshold to ensure basic comprehensibility and training effectiveness. The minimum speech rate threshold is set as follows: and satisfy .

[0110] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0111] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0112] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0113] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A spoken situational dialogue simulation system, characterized in that, include: Oral proficiency data acquisition module: Collects user oral proficiency feature data, including pause frequency, grammatical error rate and vocabulary diversity, and each oral proficiency feature data is acquired in real time from the simulation; Ability calculation module: preprocesses the collected oral proficiency feature data and establishes a mathematical conversion model. The mathematical conversion model adopts a machine learning algorithm and includes an input layer, a hidden layer and an output layer. The above sets of data are input into the input layer of the mathematical conversion model, and the oral proficiency index is obtained through the output layer. First speech rate calibration module: Set an initial speech rate benchmark, construct a speech rate calibration formula, retrieve the oral communication ability index, use the oral communication ability index as a calibration coefficient, and simultaneously input the calibration coefficient and the initial speech rate benchmark into the speech rate calibration formula to obtain the speech rate calibration value. Trend Analysis Module: Set a fixed time window t, obtain the oral proficiency index of the corresponding stage within the fixed time window t, collect N consecutive fixed time windows, and take the N consecutive fixed time windows as a judgment period, obtain the oral proficiency index of adjacent fixed time windows within a judgment period, and calculate the change value between the oral proficiency index of adjacent fixed time windows. The second speech rate calibration module calculates the average change in spoken ability within a judgment period, sets a speech rate increase coefficient and a speech rate decrease coefficient, and determines whether the speech rate calibration value needs to be adjusted a second time based on the average change in spoken ability. If no second adjustment is needed, the speech rate calibration value output by the first speech rate calibration module continues to be used. When it is determined that the speech rate calibration value needs to be adjusted a second time, the system selects to call the speech rate increase coefficient or the speech rate decrease coefficient to adjust the speech rate calibration value.

2. The oral situational dialogue simulation system according to claim 1, characterized in that: The data collection process for each oral proficiency feature in the oral proficiency data collection module is as follows: Pause frequency acquisition: Real-time monitoring of the user's spoken output stream, identification and recording of silence intervals in the user's speech, setting a pause threshold, detecting short pauses between continuous speech, marking nodes with short pause times greater than or equal to the pause threshold, and calculating the ratio of the number of pause nodes to the total speaking time as the pause frequency. Syntax error rate collection: Establish a grammar rule base, perform real-time text analysis based on dialogue content, the system identifies grammatical errors, and calculates the percentage of error events to total discourse units as the grammar error rate; Lexical diversity collection: Real-time dynamic evaluation is performed through type-tag ratio calculation, and the proportion of different words to the total vocabulary is calculated as lexical diversity data.

3. The oral situational dialogue simulation system according to claim 1, characterized in that: The ability calculation module performs preprocessing on spoken language proficiency feature data, including data cleaning, standardization, and feature integration. Data cleaning includes detecting and correcting outliers or missing values ​​in pause frequency, grammatical error rate, and lexical diversity; identifying invalid data points through statistical methods and imputing or removing them; and the standardization stage converts features of different dimensions into numerical ranges between 0 and 1 through normalization algorithms to eliminate scale differences. In the feature integration stage, the cleaned and standardized spoken language proficiency feature data are combined into a structured feature vector, which is then used as the input layer of the mathematical transformation model for processing.

4. The oral situational dialogue simulation system according to claim 1, characterized in that: The formula used in the ability calculation module to calculate the oral proficiency index is as follows: in: SAI stands for Spoken English Proficiency Index; LD represents lexical diversity data; PF represents the pause frequency; GER stands for syntax error rate; w is a mathematical constant used to ensure that the formula holds true and to prevent the denominator from being zero, and w satisfies: w > 0; The pause frequency, grammatical error rate, and lexical diversity data, after preprocessing, range between [0,1].

5. The oral situational dialogue simulation system according to claim 4, characterized in that: The training process of the mathematical transformation model is based on a supervised learning paradigm. It uses a historical dataset labeled with real oral proficiency levels, minimizes the mean squared error loss function between the predicted index and the real score, uses the Adam optimizer for backpropagation to iteratively update the weight parameters, and implements an early stopping strategy during training to prevent overfitting.

6. The oral situational dialogue simulation system according to claim 1, characterized in that: The speech rate calibration formula in the first speech rate calibration module is: in: This is the speech rate calibration value; B represents the initial speech rate benchmark, indicating the system's preset default speech rate value, which is set based on the average level of multiple users and serves as the starting point for adjustment. k is the sensitivity coefficient; This serves as a reference threshold for spoken language proficiency. It is an exponential function.

7. The oral situational dialogue simulation system according to claim 6, characterized in that: It also includes the process of obtaining the sensitivity coefficient k: Historical dialogue data samples are collected, which include user actual speech rate adaptation feedback corresponding to different oral proficiency indices. A parameter optimization objective function is established based on a preset speech rate calibration formula. The objective function is to minimize the weighted sum of user discomfort score and speech rate matching error. The objective function is solved using the gradient descent algorithm under the constraint of k>0. The sensitivity coefficient k value with the best generalization is determined by cross-validation and embedded into the system.

8. The oral situational dialogue simulation system according to claim 7, characterized in that: In the trend analysis module, multiple fixed time windows t are marked according to a sequence. Extract the spoken language proficiency index at the end of each window and record it as . For any pair of adjacent windows: The change value of its oral proficiency index is Obtained through difference operations: in: For the nth fixed time window, the oral proficiency index is used. For the (n+1)th fixed time window, the oral proficiency index is used. The change value is used to quantify the instantaneous fluctuation direction and magnitude of a user's spoken fluency within adjacent time periods, and its meaning is as follows: This indicates an improvement in the user's fluency in spoken conversation; This indicates that the user's fluency in spoken conversation remains unchanged; This indicates a decrease in the fluency of the user's spoken conversation.

9. The oral situational dialogue simulation system according to claim 8, characterized in that: The second speech rate calibration module calculates the average change in spoken ability over a judgment period as follows: in: denoted as the average change in oral communication ability, and N as the total number of fixed time windows within a judgment period.

10. The oral situational dialogue simulation system according to claim 9, characterized in that: Set the average change threshold for oral proficiency as And satisfy ; when At that time, the speech rate calibration value is determined using the formula: A second adjustment was made, among which The final speech rate after the second adjustment is given, where α is the speech rate adjustment coefficient. This indicates that the user's training status in the previous judgment period has shown significant improvement. Therefore, the speech rate in the system's simulated situational dialogue is further increased, and a maximum speech rate threshold is set. and satisfy ; when When this occurs, it indicates that the user's training state in the previous judgment period is stable. Therefore, the system's simulated dialogue speed remains unchanged and continues to use the speed calibration value. The simulated speech rate of the situational dialogue serves as the basis for the next judgment cycle; when At that time, the speech rate calibration value is determined using the formula: A second adjustment is made, where β is the speech rate reduction coefficient. This indicates that the user's training status in the previous judgment period was too difficult; therefore, the system's simulated speech rate in the situational dialogue is further reduced, and a minimum speech rate threshold is set. and satisfy .

Images