Intelligent virtual companion robot contextual dialogue pacing control system

Abstract

The application discloses a smart virtual accompanying robot scene dialogue rhythm control system and belongs to the technical field of robot dialogue. The system comprises the following modules: a data acquisition module for acquiring voice data of historical speeches; a historical vector extraction module for storing low-dimensional rhythm shape vectors in a vector database; an online clustering module for generating multiple low-dimensional rhythm clusters and response templates; a current vector extraction module for extracting rhythm shape vectors; a similarity acquisition module for acquiring similarity; a rhythm response module for calling rhythm response templates; and a new cluster generation module for generating new rhythm clusters online and creating new templates. The application compresses user voice rhythm into low-dimensional shape vectors, forms exclusive rhythm clusters through online clustering, and matches in a sliding slice manner with zero delay in real-time dialogue, so that the technical effect of instant response to dialogue rhythm and long-term memory of personal style is achieved, and the problems of rhythm response lag and lack of context long-term memory in the prior art are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of robot dialogue, and in particular to a scenario dialogue rhythm control system for an intelligent virtual companion robot. BACKGROUND

[0002] With the increasing application of intelligent virtual companion robots in home, education and medical scenarios, the naturalness and fluency of human-computer dialogue have become important indicators of user experience. To meet the needs of users for efficient response and personalized interaction of companion robots, the dialogue rhythm must be precisely controlled and dynamically adapted. Existing intelligent virtual companion robot systems analyze the rhythm of user speech by full-sentence statistical analysis based on voice activity detection and acoustic feature threshold to determine the starting time and pause position of the robot response.

[0003] In the prior art, there is a significant delay in extracting rhythm parameters for user fast speech, and there is a lack of long-term memory and personalized processing of user historical speech style, resulting in rhythm parameter response lag and lack of long-term context memory. SUMMARY

[0004] The embodiments of the present application provide an intelligent virtual companion robot scenario dialogue rhythm control system, which solves the problems of rhythm parameter response lag and lack of long-term context memory in the prior art, and realizes millisecond-level prediction and triggering of rhythm response in a multi-user concurrent dialogue scenario and continuous personalized adaptation of user long-term communication style.

[0005] The embodiments of the present application provide an intelligent virtual companion robot scenario dialogue rhythm control system, which includes: a data acquisition module for acquiring voice data of user historical speech and a unique user identifier;

[0006] A historical vector extraction module is configured to extract a low-dimensional rhythm shape vector from the acquired voice data, and store the low-dimensional rhythm shape vector in association with the unique user identifier in a vector database;

[0007] An online clustering module is configured to perform online clustering on the low-dimensional rhythm shape vector, and dynamically generate a plurality of low-dimensional rhythm clusters and their associated low-dimensional rhythm response templates, which refer to the stored low-dimensional rhythm shape vectors corresponding to the rhythm style;

[0008] A current vector extraction module is configured to continuously extract a rhythm shape vector from the user's voice in the current dialogue;

[0009] A similarity acquisition module is configured to acquire the similarity between the rhythm shape vector and the low-dimensional rhythm cluster center vector, and compare the similarity with a preset dynamically adjustable matching threshold;

[0010] Rhythm response module: for determining a match success when the similarity is not less than the matching threshold, and calling the rhythm response template associated with the low-dimensional rhythm cluster of the matching success;

[0011] New cluster generation module: for determining a matching failure when an unmatched new rhythm pattern appears and the occurrence frequency reaches the preset frequency threshold, enabling a default rhythm template and generating a new rhythm cluster online and creating a new template.

[0012] Further, the step of extracting a low-dimensional rhythm shape vector from the obtained voice data includes:

[0013] Standardizing the obtained voice data, and performing frame processing on the standardized voice data;

[0014] Dividing the continuous voice stream into a plurality of short-time voice frames, and pre-processing the short-time voice frames;

[0015] Based on the pre-processed short-time voice frames, extracting acoustic features reflecting the rhythm characteristics of the voice, the acoustic features including the fundamental frequency variation, energy fluctuation, speech rate information and sound length distribution of the voice, and quantitatively analyzing the acoustic features;

[0016] Integrating and reducing the extracted acoustic features, mapping the high-dimensional acoustic features to a low-dimensional space to form a low-dimensional rhythm shape vector.

[0017] Further, the step of online clustering the low-dimensional rhythm shape vector and dynamically generating a plurality of low-dimensional rhythm clusters corresponding to the stored low-dimensional rhythm shape vectors of the rhythm style and the associated low-dimensional rhythm response templates includes:

[0018] Calculating the centroid distance of each vector, grouping the vectors with a distance less than a set threshold into the same set, and averaging the set to obtain a new low-dimensional rhythm cluster center vector;

[0019] Establishing a unique low-dimensional rhythm cluster identifier in each new low-dimensional rhythm cluster center vector, and creating a rhythm response template with the same name as the low-dimensional rhythm cluster identifier;

[0020] The centroid distance of each vector is obtained by the centroid distance calculation formula:

[0021] ;

[0022] In the formula, is a dynamic time sequence weighted centroid distance, is an M-dimensional component of the current rhythm shape vector, is an M-dimensional component of the corresponding rhythm cluster center vector, is the weight coefficient of the first dimensional feature, is a time decay coefficient, is a timestamp of the current vector, is a nominal timestamp of the cluster center vector, is a time normalization constant, is a natural constant, is a current rhythm shape vector of the first dimensional component, is a first dimensional component of the rhythm cluster center vector of the first dimensional component,

[0023] Further, the step of averaging the set to obtain a new low-dimensional rhythm cluster center vector comprises:

[0024] The new low-dimensional rhythm cluster center vector is obtained by calculating a weighted average formula:

[0025] ;

[0026] In the formula, is a vector number, is a first dimensional component of the first vector in the set, is the total number of vectors used to update the cluster center, is a time decay-based weight, wherein is the timestamp of the first vector, is the current time when the current clustering update is triggered, is a time decay coefficient, is a time normalization constant, is a weight coefficient of the first dimensional feature.

[0027] Further, the step of obtaining the similarity between the rhythm shape vector and the low-dimensional rhythm cluster center vector comprises:

[0028] Obtaining the latest low-dimensional rhythm shape vector continuously extracted from the user's voice in the current dialogue as input data for similarity calculation;

[0029] Retrieving all generated center vectors of low-dimensional rhythm clusters from the vector database;

[0030] Using a preset similarity calculation formula, the similarity between the extracted latest low-dimensional rhythm shape vector and each retrieved low-dimensional rhythm cluster center vector is calculated.

[0031] Further, the similarity calculation formula is:

[0032] ;

[0033] Where, is the weighted cosine similarity with temporal decay, is the currently extracted low-dimensional rhythm shape vector, is the low-dimensional shape vector corresponding to the center of the rhythm cluster, For the dimensional feature weights, To form the basis of weighted cosine similarity, is the timing attenuation factor, is the time series decay rate parameter, is the timestamp of the current vector, is the timestamp of the cluster center vector, is the time normalization constant, is the current rhythm shape vector No. dimensional component, is the rhythm cluster center vector No. dimensional component, is the total number of feature dimensions.

[0034] Furthermore, the step of dynamically adjusting the matching threshold includes:

[0035] Continuously monitor the overall performance of the robot's scenario dialogue rhythm control system, including the rhythm pattern matching success rate and matching failure rate;

[0036] According to the preset dynamic adjustment formula and combined with the monitored overall operation performance data, the matching threshold is adjusted in real time;

[0037] The dynamically adjusted matching threshold is applied to the subsequent similarity comparison process between the rhythm shape vector and the low-dimensional rhythm cluster center vector.

[0038] Furthermore, the preset dynamic adjustment formula is:

[0039] ;

[0040] Where, for Dynamic matching threshold updated at all times, is the initial baseline threshold, for Real-time matching success rate at each moment, is the target matching success rate, for Real-time matching failure rate at each moment, is the target matching failure rate, for a user feedback score at the moment, a target user feedback score, a success rate sensitivity coefficient, a failure rate sensitivity coefficient, a user feedback weight coefficient, a timestamp of the last threshold adjustment, a time decay coefficient, a time normalization constant.

[0041] Further, when the similarity is not less than the matching threshold, it is determined that the matching is successful, and the step of calling the rhythm response template associated with the successfully matched low-dimensional rhythm cluster includes:

[0042] comparing the obtained similarity value with the matching threshold;

[0043] when the similarity value is not less than the matching threshold, it is determined that the current user rhythm mode is successfully matched with one of the low-dimensional rhythm clusters, and the successfully matched low-dimensional rhythm cluster is identified;

[0044] the low-dimensional rhythm response template associated with the rhythm cluster is called and enabled to guide the robot to generate dialogue content conforming to the user rhythm.

[0045] Further, the step of determining that the matching fails, enabling a default rhythm template, and generating a new rhythm cluster online and creating a new template includes:

[0046] after the similarity comparison, if the current rhythm shape vector fails to be successfully matched with all existing low-dimensional rhythm clusters, it is determined that a new rhythm mode that is not matched appears;

[0047] the appearance frequency of the new mode is continuously tracked and counted, and when the appearance frequency reaches a preset frequency threshold, it is confirmed that it is a new rhythm style worthy of recognition;

[0048] a preset default rhythm template is enabled, and a new low-dimensional rhythm cluster is generated online according to the characteristics of the confirmed unmatched new rhythm mode, and a new low-dimensional rhythm response template is created and associated with the new cluster.

[0049] The one or more technical solutions provided in the embodiments of the application have at least the following technical effects or advantages:

[0050] 1. By collecting user speeches in the background and compressing multi-dimensional rhythm fluctuations into low-dimensional shape vectors, high-speed extraction and matching of user speech rhythm characteristics are realized, and millisecond-level prediction and triggering of rhythm responses of the system in a multi-user concurrent dialogue scene are realized, effectively solving the problems of rhythm parameter response lag and insufficient real-time in the prior art.

[0051] 2. By dynamically adjusting the threshold based on the matching success rate and drift degree, accurate recognition and classification of various rhythm patterns can be achieved, thereby enabling the system to continuously remember the user's long-term communication style and adapt to the personalized rhythm, effectively solving the problem of the lack of long-term contextual memory and rhythm style adaptation capabilities in existing technologies.

[0052] 3. By triggering the generation of new clusters based on the frequency threshold of unmatched new rhythm patterns and creating corresponding response templates online, the system can achieve self-learning and online evolution of emerging or changing rhythm styles, thereby enabling the robot to continuously provide a coherent conversation experience that matches the user's rhythm in different users and conversation environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a structural diagram of the intelligent virtual companion robot scenario dialogue rhythm control system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The embodiments of the present application solve the problems of rhythm parameter response lag and long-term context memory loss in the prior art by providing an intelligent virtual companion robot scenario dialogue rhythm control system. By silently collecting and compressing low-dimensional rhythm shape vectors in the background and combining online clustering to dynamically match rhythm response templates, the system achieves millisecond-level prediction and triggering of rhythm responses in multi-user concurrent dialogue scenarios and continuous personalized adaptation to users' long-term communication styles.

[0055] In order to better understand the above technical solution, the above technical solution will be described in detail below with reference to the accompanying drawings and specific implementation methods.

[0056] like Figure 1 The figure shows a schematic diagram of the structure of the intelligent virtual companion robot scenario dialogue rhythm control system provided by an embodiment of the present application, including: a data acquisition module: used to obtain the voice data of the user's historical speech and a unique user identifier, the voice data including the speech duration, speaking speed and pause information; the unique user identifier is generated during the voiceprint registration stage;

[0057] A history vector extraction module is used to extract a low-dimensional rhythm shape vector from the acquired speech data using Fourier transform technology, and associate the low-dimensional rhythm shape vector with a unique user identifier and store it in a vector database;

[0058] Online clustering module: used to perform online clustering on stored low-dimensional rhythm shape vectors to identify common low-dimensional rhythm patterns, dynamically generate and maintain multiple low-dimensional rhythm clusters that refer to the rhythm styles corresponding to the stored low-dimensional rhythm shape vectors and their associated low-dimensional rhythm response templates;

[0059] Current vector extraction module: for continuously extracting rhythm shape vectors from the user's speech in the current dialogue through a sliding window;

[0060] Similarity acquisition module: for acquiring the similarity between the rhythm shape vector and the low-dimensional rhythm cluster center vector, and comparing the similarity with a preset dynamically adjustable matching threshold; the matching threshold is adaptively adjusted according to the user rhythm drift degree and the historical success rate.

[0061] Rhythm response module: for determining that the matching is successful when the similarity is not less than the matching threshold, and calling the rhythm response template associated with the successfully matched low-dimensional rhythm cluster;

[0062] New cluster generation module: for determining that the matching fails when a new rhythm pattern that is not matched appears and the appearance frequency reaches a preset frequency threshold, enabling a default rhythm template and generating a new rhythm cluster online and creating a new template.

[0063] Further, the step of extracting a low-dimensional rhythm shape vector from the obtained speech data includes:

[0064] Standardizing the obtained speech data, including performing a noise reduction operation on the speech signal to remove background noise, and performing frame processing on the standardized speech data;

[0065] Dividing the continuous speech stream into a plurality of short-time speech frames, and pre-processing the short-time speech frames to ensure the accuracy of subsequent feature extraction;

[0066] Based on the pre-processed short-time speech frames, acoustic features reflecting the rhythm characteristics of the speech are extracted, including the fundamental frequency variation, energy fluctuation, speech rate information and sound length distribution of the speech, and the acoustic features are quantitatively analyzed to capture the inherent rhythm pattern of the user's speech;

[0067] The extracted acoustic features are integrated and dimensionally reduced, and the high-dimensional acoustic features are mapped to a low-dimensional space to form a low-dimensional rhythm shape vector, which can effectively represent the unique speaking rhythm characteristics of the user and provide basic data for subsequent rhythm pattern recognition.

[0068] Further, the step of online clustering the low-dimensional rhythm shape vector to dynamically generate a plurality of low-dimensional rhythm clusters corresponding to the stored low-dimensional rhythm shape vectors of the rhythm style and the associated low-dimensional rhythm response templates includes:

[0069] Calculate the centroid distance of each vector, and group the vectors with a distance less than a set threshold into the same set, and average the set to obtain a new low-dimensional rhythm cluster center vector;

[0070] Establishing a unique low-dimensional rhythm cluster identifier in each new low-dimensional rhythm cluster center vector, and creating a rhythm response template with the same name as the low-dimensional rhythm cluster identifier, to obtain a low-dimensional rhythm cluster that refers to the rhythm style corresponding to the stored low-dimensional rhythm shape vector and its associated low-dimensional rhythm response template;

[0071] The centroid distance of each vector is obtained by the centroid distance calculation formula:

[0072] ;

[0073] Where, is the dynamic time series weighted centroid distance, which is used to measure the difference between the current vector and the cluster center. is the M-dimensional component of the current rhythm shape vector, is the M-dimensional component of the center vector of the corresponding rhythm cluster, For the The weight coefficient of the dimension feature is automatically assigned by the online clustering module according to the inverse of the variance of the dimension within the cluster, which is used to highlight the importance of low-variance dimensions in distance calculation. is the time attenuation coefficient, is the timestamp of the current vector, is the calibration timestamp of the cluster center vector, is the time normalization constant, is the current rhythm shape vector No. dimensional component, is the rhythm cluster center vector No. dimensional component, is the total number of feature dimensions;

[0074] Using the original rhythm fluctuation sequence corresponding to the center vector of the low-dimensional rhythm cluster as a reference, the position and length of each pause point in the sequence are extracted to obtain the average speaking rate and average pause interval of the sequence. The average speaking rate is mapped to the response delay time, and the pause position sequence is retained as a pause position list. The computer combines the response delay time and the pause position list into a rhythm response template, and the template name is consistent with the low-dimensional rhythm cluster identifier;

[0075] After the computer completes the creation, it immediately writes the low-dimensional rhythm cluster center vector, the low-dimensional rhythm cluster identifier, and the rhythm response template into the cluster partition of the vector database for subsequent real-time call.

[0076] Furthermore, the step of averaging the set to obtain a new low-dimensional rhythm cluster center vector includes:

[0077] The new low-dimensional rhythm cluster center vector is obtained by calculating the weighted average formula:

[0078] ;

[0079] Where, is the vector number, For the set The first vector dimensional component, is the total number of vectors used to update cluster centers, is the weight based on time decay, where It is timestamps of a vector, is the current time when this cluster update is triggered, is the time attenuation coefficient, is the time normalization constant, For the The weight coefficient of the dimension feature is used to emphasize the impact of the dimension with smaller variance within the corresponding cluster on the center update.

[0080] Furthermore, the step of obtaining the similarity between the rhythm shape vector and the low-dimensional rhythm cluster center vector includes:

[0081] Obtaining the latest low-dimensional rhythm shape vector continuously extracted from the user's speech in the current conversation, wherein the latest low-dimensional rhythm shape vector represents the user's current speaking rhythm style as input data for similarity calculation;

[0082] Retrieve the center vectors of all generated low-dimensional rhythm clusters from the vector database. These center vectors represent different, identified rhythm style patterns and provide reference points for subsequent similarity comparisons.

[0083] Using the preset similarity calculation formula, the similarity calculation is performed on the latest low-dimensional rhythm shape vector extracted and the center vector of each retrieved low-dimensional rhythm cluster. This calculation process quantifies the degree of match between the current user rhythm and the known rhythm style, and generates a similarity value that reflects the degree of closeness between the two.

[0084] Furthermore, the similarity calculation formula is:

[0085] ;

[0086] Where, is the weighted cosine similarity with time decay, dimensionless, is the currently extracted low-dimensional rhythm shape vector, each component of which is a dimensionless feature. is the low-dimensional shape vector corresponding to the center of the rhythm cluster, and each component is related to Same dimension, For the dimensionless, automatically assigned by the online clustering module inversely proportional to the within-cluster variance of the dimension, is a weighting cosine similarity base term, is a time decay factor, dimensionless, ensuring the distance similarity smoothly decreases as the time difference increases, is a time decay rate parameter, dimension of “inverse per time unit”, used to adjust the speed of time decay, is the timestamp of the current vector, dimension of time, is the timestamp of the cluster center vector, dimension of time, is a time normalization constant, dimension of time, used to normalize the time difference, is the first dimension component of the current rhythm shape vector , is the first dimension component of the rhythm cluster center vector , is the total number of feature dimensions.

[0087] Further, the step of dynamically adjusting the matching threshold value comprises:

[0088] Continuously monitoring the overall performance of the robot scenario dialogue rhythm control system, including the matching success rate, matching failure rate of the rhythm pattern, and user feedback on dialogue rhythm adjustment, which provides the basis for threshold adjustment;

[0089] According to the preset dynamic adjustment formula, combined with the monitored overall performance data, the matching threshold value is adjusted in real time, for example, when the matching success rate is too high or too low, the system will correspondingly increase or decrease the matching threshold value, to optimize the recognition accuracy and adaptability of the rhythm pattern;

[0090] The dynamically adjusted matching threshold value is applied to the subsequent similarity comparison process between the rhythm shape vector and the low-dimensional rhythm cluster center vector, ensuring that the system can flexibly determine the matching result of the rhythm pattern according to the latest running state and optimization target, thereby improving the accuracy of dialogue rhythm control.

[0091] Further, the preset dynamic adjustment formula is:

[0092] ;

[0093] wherein, is the dynamically updated dynamic matching threshold value, dimensionless, with a value range of , directly used for subsequent similarity and threshold comparison and determination, is the initial reference threshold value, is the initial reference threshold value, is the initial reference threshold value, Real-time matching success rate of the moment, dimensionless, reflecting the current system's recognition ability to the known rhythm pattern, Target matching success rate, dimensionless, representing the ideal success rate level expected to be achieved, Real-time matching failure rate of the moment, dimensionless, Target matching failure rate, dimensionless, User feedback score of the moment, dimensionless, Target user feedback score, dimensionless, Success rate sensitivity coefficient, dimensionless, Failure rate sensitivity coefficient, dimensionless, User feedback weight coefficient, dimensionless, Timestamp of the last threshold adjustment, time dimension, Time decay coefficient, dimensionless, Time normalization constant, time dimension.

[0094] Further, when the similarity is not less than the matching threshold, it is determined that the matching is successful, and the steps of calling the rhythm response template associated with the low-dimensional rhythm cluster of the matching success include:

[0095] The similarity value obtained is compared with the matching threshold to determine whether the current user rhythm pattern is consistent with the identified rhythm style pattern;

[0096] When the similarity value is not less than the matching threshold, it is determined that the current user rhythm pattern successfully matches one of the low-dimensional rhythm clusters, and the low-dimensional rhythm cluster of the successful matching is identified;

[0097] The low-dimensional rhythm response template associated with the rhythm cluster is called and enabled, which contains the dialogue response strategy and rhythm adjustment parameters for the stored low-dimensional rhythm shape vector corresponding to the rhythm style, guiding the robot to generate dialogue content consistent with the user rhythm.

[0098] Further, the steps of determining matching failure, enabling the default rhythm template, and generating a new rhythm cluster online and creating a new template include:

[0099] After the similarity comparison, if the current rhythm shape vector fails to successfully match all existing low-dimensional rhythm clusters, it is determined that a new rhythm pattern that does not match appears;

[0100] The frequency of the new pattern is continuously tracked and counted, and when the frequency reaches a preset frequency threshold, it is confirmed as a new rhythm style worthy of recognition;

[0101] ​​The preset default rhythm template is enabled, and a new low-dimensional rhythm cluster is generated online according to the confirmed characteristics of the unmatched new rhythm mode, and a new low-dimensional rhythm response template is created and associated for the new cluster, which defines the corresponding dialogue response strategy and rhythm adjustment parameters.

[0102] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can be embodied in the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROMs, optical storage media, etc.) having computer usable program code embodied therein.

[0103] The application is described with reference to the flowcharts and / or block diagrams of the methods, apparatus (systems), and computer program products according to embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing apparatus to produce a machine, so that the instructions, which are executed via the processor of the computer or other programmable data processing apparatus, generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in the flowchart

[0104] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including instruction means, which implement the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in the flowchart

[0105] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus, so that a series of operation steps are performed on the computer or other programmable data processing apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable data processing apparatus provide a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the functions specified in the flowchart

[0106] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0107] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.

Claims

1. An intelligent virtual companion robot contextual dialogue pacing control system, characterized by, Comprise: Data acquisition module: for acquiring user history speech voice data and unique user identification; History vector extraction module: for extracting low-dimensional rhythm shape vector from the acquired voice data, and storing the low-dimensional rhythm shape vector in the vector database in association with the unique user identification; Online clustering module: for online clustering of low-dimensional rhythm shape vectors, dynamically generating a plurality of low-dimensional rhythm clusters corresponding to the rhythm style of the stored low-dimensional rhythm shape vectors and their associated low-dimensional rhythm response templates; Current vector extraction module: for continuously extracting rhythm shape vectors from the user's voice in the current dialogue; Similarity acquisition module: for acquiring the similarity of the rhythm shape vector and the low-dimensional rhythm cluster center vector, and comparing the similarity with the preset dynamically adjustable matching threshold value; The step of dynamically adjusting the matching threshold value comprises: Continuously monitoring the overall running performance of the robot scenario dialogue rhythm control system, including the matching success rate and matching failure rate of the rhythm pattern; According to the preset dynamic adjustment formula, the overall running performance data monitored is combined to adjust the matching threshold value in real time; The dynamically adjusted matching threshold value is applied to the subsequent similarity comparison process of the rhythm shape vector and the low-dimensional rhythm cluster center vector; The preset dynamic adjustment formula is: ； In the formula, is the dynamic matching threshold updated at the moment, is the initial reference threshold, is the real-time matching success rate at the moment, is the target matching success rate, is the real-time matching failure rate at the moment, is the target matching failure rate, is the user feedback score at the moment, is the target user feedback score, is the success rate sensitivity coefficient, is the failure rate sensitivity coefficient, is the user feedback weight coefficient, is the timestamp of the last threshold adjustment, is the time decay coefficient, is the time normalization constant; Rhythm response module: for determining that the matching is successful when the similarity is not less than the matching threshold value, and calling the rhythm response template associated with the matched low-dimensional rhythm cluster; New cluster generation module: for determining that the matching fails when a new rhythm pattern appears and the occurrence frequency reaches the preset frequency threshold, enabling the default rhythm template and generating a new rhythm cluster online and creating a new template.

2. The intelligent virtual companion robot contextual dialogue pacing control system of claim 1, wherein, The step of extracting low-dimensional rhythm shape vectors from the acquired voice data comprises: Standardizing the acquired voice data, and performing frame processing on the standardized voice data; Divide the continuous voice stream into a plurality of short-time voice frames, and pre-process the short-time voice frames; Based on the pre-processed short-time voice frames, extract acoustic features reflecting the rhythm characteristics of the voice, including the fundamental frequency variation, energy fluctuation, speech rate information and sound length distribution of the voice, and perform quantitative analysis on the acoustic features; Integrate and reduce the extracted acoustic features, map the high-dimensional acoustic features to a low-dimensional space, and form a low-dimensional rhythm shape vector. 3.The intelligent virtual companion robot contextual dialogue pacing control system of claim 1, wherein, The step of online clustering of low-dimensional rhythm shape vectors, dynamically generating a plurality of low-dimensional rhythm clusters corresponding to the rhythm style of the stored low-dimensional rhythm shape vectors and their associated low-dimensional rhythm response templates comprises: Calculate the centroid distance of each vector, and group the vectors with a distance less than a set threshold into the same set, and average the set to obtain a new low-dimensional rhythm cluster center vector; Establish a unique low-dimensional rhythm cluster identifier in each new low-dimensional rhythm cluster center vector, and create a rhythm response template with the same name as the low-dimensional rhythm cluster identifier; The centroid distance of each vector is obtained by the centroid distance calculation formula: ； Where, is the dynamic time series weighted centroid distance, is the M-dimensional component of the current rhythm shape vector, is the M-dimensional component of the center vector of the corresponding rhythm cluster, For the The weight coefficient of the dimension feature, is the time attenuation coefficient, is the timestamp of the current vector, is the calibration timestamp of the cluster center vector, is the time normalization constant, is a natural constant, is the current rhythm shape vector No. dimensional component, is the rhythm cluster center vector No. dimensional component, is the total number of feature dimensions.

4. The intelligent virtual companion robot contextual dialogue pacing control system of claim 3, wherein, The step of averaging the set to obtain a new low-dimensional rhythm cluster center vector comprises: The new low-dimensional rhythm cluster center vector is obtained by the weighted average formula: ； Where, is the vector number, For the set The first vector dimensional component, is the total number of vectors used to update cluster centers, is the weight based on time decay, where It is timestamps of a vector, is the current time when this cluster update is triggered, is the time attenuation coefficient, is the time normalization constant, For the The weight coefficient of the dimension feature. 5.The intelligent virtual companion robot contextual dialogue pacing control system of claim 1, wherein, The step of obtaining the similarity between the rhythm shape vector and the low-dimensional rhythm cluster center vector comprises: Obtaining the latest low-dimensional rhythm shape vector continuously extracted from the user voice in the current dialogue as input data for similarity calculation; Retrieving the center vectors of all generated low-dimensional rhythm clusters from the vector database; Using a preset similarity calculation formula, the similarity between the extracted latest low-dimensional rhythm shape vector and each retrieved low-dimensional rhythm cluster center vector is calculated. 6.The intelligent virtual companion robot contextual dialogue pacing control system of claim 5, wherein, The similarity calculation formula is: ； Where, is the weighted cosine similarity with temporal decay, is the currently extracted low-dimensional rhythm shape vector, is the low-dimensional shape vector corresponding to the center of the rhythm cluster, For the dimensional feature weights, To form the basis of weighted cosine similarity, is the timing attenuation factor, is the time series decay rate parameter, is the timestamp of the current vector, is the timestamp of the cluster center vector, is the time normalization constant, is the current rhythm shape vector No. dimensional component, is the rhythm cluster center vector No. dimensional component, is the total number of feature dimensions.

7. The intelligent virtual companion robot contextual dialogue pacing control system of claim 1, wherein, When the similarity is not less than the matching threshold, it is determined that the matching is successful, and the steps of calling the rhythm response template associated with the successfully matched low-dimensional rhythm cluster comprise: Comparing the obtained similarity value with the matching threshold; When the similarity value is not less than the matching threshold, it is determined that the current user rhythm mode is successfully matched with one of the low-dimensional rhythm clusters, and the successfully matched low-dimensional rhythm cluster is identified; The low-dimensional rhythm response template associated with the rhythm cluster is called and enabled to guide the robot to generate dialogue content conforming to the user rhythm.

8. The intelligent virtual companion robot contextual dialogue pacing control system of claim 1, wherein, The steps of determining that the matching fails, enabling the default rhythm template, and generating a new rhythm cluster online and creating a new template comprise: After the similarity comparison, if the current rhythm shape vector fails to be successfully matched with all existing low-dimensional rhythm clusters, it is determined that a new rhythm mode that is not matched appears; The frequency of the appearance of the new rhythm mode is continuously tracked and counted, and when the frequency reaches a preset frequency threshold, it is confirmed that it is a new rhythm style worthy of recognition; A preset default rhythm template is enabled, and a new low-dimensional rhythm cluster is generated online according to the characteristics of the confirmed unmatched new rhythm mode, and a new low-dimensional rhythm response template is created and associated with the new cluster.

Images