A method and system for monitoring the operation status of a stereo garage based on voiceprint

Abstract

The present invention discloses a method and system for monitoring the operating status of a stereo garage based on voiceprints, comprising: first, collecting the main and related operating voiceprint data of the equipment to be monitored in the stereo garage. Using a pre-trained expert model, multiple rounds of feature extraction are performed on these voiceprint data to generate a data set containing voiceprint features of multiple different granularity levels. Subsequently, a first comprehensive voiceprint feature data set is obtained through feature integration processing, and a lightweight model is used to perform feature reconstruction to generate a second comprehensive voiceprint feature data set. Finally, through feature comparison processing, the information of the two data sets is integrated to evaluate the operating status of the garage equipment to be monitored, and then the operating status of the entire stereo garage is determined. Such a design effectively integrates multi-source voiceprint information and multi-granularity features, thereby improving the accuracy and robustness of the operating status monitoring of the stereo garage.

Description

Technical Field

[0001] The present invention relates to the field of acoustic measurement, and in particular to a method and system for monitoring the operating status of a stereo garage based on voiceprints. Background Art

[0002] As an important parking facility in modern cities, the operation status monitoring of the three-dimensional parking garage is crucial to ensure parking safety and improve user experience. Traditional monitoring methods mostly rely on sensor data and visual monitoring, but they are easily interfered with in complex environments and are costly. Summary of the invention

[0003] The object of the present invention is to provide a method and system for monitoring the operating status of a stereoscopic parking garage based on voiceprint.

[0004] In a first aspect, an embodiment of the present invention provides a method for monitoring the operation status of a stereo garage based on voiceprint, comprising:

[0005] Acquire main operation voiceprint data and at least one associated operation voiceprint data of the to-be-monitored garage equipment included in the stereo garage;

[0006] Performing multiple rounds of voiceprint feature extraction on the main operation voiceprint data according to the expert model to obtain a first voiceprint feature data set, and performing multiple rounds of voiceprint feature extraction on the at least one associated operation voiceprint data according to the expert model to obtain a second voiceprint feature data set, wherein the expert model is a pre-trained model, the first voiceprint feature data set includes voiceprint features of multiple different granularity levels, the second voiceprint feature data set includes voiceprint features of multiple different granularity levels, and the number of rounds of voiceprint feature extraction is consistent with the number of granularity levels;

[0007] Performing feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain a first comprehensive voiceprint feature data set of the garage equipment to be monitored;

[0008] Performing feature reconstruction processing based on the first voiceprint feature dataset and the second voiceprint feature dataset according to the lightweight model to obtain a second comprehensive voiceprint feature dataset;

[0009] Performing feature comparison processing according to the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set to obtain an equipment operation status evaluation result of the to-be-monitored garage equipment;

[0010] The operating status of the three-dimensional parking garage is determined according to the equipment operating status evaluation result.

[0011] In a second aspect, an embodiment of the present invention provides a readable storage medium, wherein the readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method described in the first aspect.

[0012] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a method and system for monitoring the operating status of a stereo garage based on voiceprints disclosed in the present invention, by collecting the main and related operating voiceprint data of the equipment to be monitored in the stereo garage. Using a pre-trained expert model, these voiceprint data are subjected to multiple rounds of feature extraction to generate a data set containing voiceprint features of multiple different granularity levels. Subsequently, a first comprehensive voiceprint feature data set is obtained through feature integration processing, and a lightweight model is used to perform feature reconstruction to generate a second comprehensive voiceprint feature data set. Finally, through feature comparison processing, the information of the two data sets is integrated to evaluate the operating status of the garage equipment to be monitored, and then the operating status of the entire stereo garage is determined. Such a design effectively integrates multi-source voiceprint information and multi-granularity features, and improves the accuracy and robustness of the operating status monitoring of the stereo garage. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0014] Figure 1 A schematic diagram of the steps of a method for monitoring the operating status of a stereoscopic parking garage based on voiceprints provided in an embodiment of the present invention;

[0015] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0017] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0018] In order to solve the technical problems in the aforementioned background technology, Figure 1A flow chart of a method for monitoring the operating status of a stereoscopic car park based on voiceprints provided in an embodiment of the present disclosure is provided below. The method for monitoring the operating status of a stereoscopic car park based on voiceprints is introduced in detail.

[0019] Step S201, obtaining main operation voiceprint data and at least one associated operation voiceprint data of the garage equipment to be monitored included in the stereo garage;

[0020] Step S202, performing multiple rounds of voiceprint feature extraction on the main operation voiceprint data according to the expert model to obtain a first voiceprint feature data set, and performing multiple rounds of voiceprint feature extraction on the at least one associated operation voiceprint data according to the expert model to obtain a second voiceprint feature data set, wherein the expert model is a model that has been trained in advance, the first voiceprint feature data set includes voiceprint features of multiple different granularity levels, the second voiceprint feature data set includes voiceprint features of multiple different granularity levels, and the number of rounds of voiceprint feature extraction is consistent with the number of granularity levels;

[0021] Step S203, performing feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain a first comprehensive voiceprint feature data set of the garage equipment to be monitored;

[0022] Step S204, performing feature reconstruction processing based on the first voiceprint feature data set and the second voiceprint feature data set according to the lightweight model to obtain a second comprehensive voiceprint feature data set;

[0023] Step S205, performing feature comparison processing according to the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set to obtain an equipment operation status evaluation result of the to-be-monitored garage equipment;

[0024] Step S206, determining the operating status of the stereoscopic parking garage according to the equipment operating status evaluation result.

[0025] In the embodiment of the present invention, exemplarily, the server first obtains the main operation voiceprint data of the garage equipment to be monitored (such as elevators, conveyor belts, etc.) and at least one operation voiceprint data associated therewith from the sensor network in the stereo garage. For example, for the elevator, the main operation voiceprint data can be its sound pattern during the normal lifting process, and the associated operation voiceprint data can be the sound pattern of its motor or cable. The server uses the expert model that has been trained in advance to perform multiple rounds of voiceprint feature extraction on the main operation voiceprint data obtained to obtain a first voiceprint feature data set containing multiple voiceprint features of different granularity levels. Similarly, multiple rounds of voiceprint feature extraction are also performed on the associated operation voiceprint data to obtain a second voiceprint feature data set. For example, the first round of extraction can identify the frequency characteristics of the sound, the second round of extraction can identify the intensity change characteristics of the sound, and so on. The server performs feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain a first comprehensive voiceprint feature data set of the garage equipment to be monitored. This step is to integrate all the extracted voiceprint features together to form a comprehensive feature set. The server uses a lightweight model to perform feature reconstruction processing based on the first voiceprint feature data set and the second voiceprint feature data set to obtain a second comprehensive voiceprint feature data set. This step aims to reconstruct the voiceprint features through the model to further refine and enhance the expressiveness of the features. The server performs feature comparison processing on the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set to obtain the equipment operation status evaluation results of the monitored garage equipment. This step is to identify any anomalies or changes by comparing the two comprehensive feature data sets, thereby evaluating the operating status of the equipment. Finally, the server determines the overall operating status of the stereo garage based on the equipment operation status evaluation results. If the operating status of most devices is evaluated as normal, the operating status of the stereo garage is judged to be normal; if there is an abnormality in any device, the operating status of the stereo garage is judged to be abnormal and may require further inspection and maintenance. Through this series of steps, the server can effectively use voiceprint data to monitor the operating status of the stereo garage, discover potential problems in a timely manner, and ensure the safe and efficient operation of the garage.

[0026] In order to more clearly describe the solution provided by the embodiment of the present application, a more detailed description is given below.

[0027] In an embodiment of the present invention, exemplarily, a stereo garage is a multi-storey parking facility that moves vehicles vertically or horizontally to different parking spaces through mechanical or automated equipment to maximize the utilization of parking space. In a stereo garage, equipment such as elevators, conveyor belts, motors, sensors, etc. that need to be monitored for operating status. In a stereo garage, elevators are used to move vehicles vertically, and conveyor belts may be used to move vehicles horizontally. These are all garage equipment to be monitored. The main operating soundprint data refers to the sound pattern or characteristics generated by the garage equipment to be monitored during normal operation, and these data can reflect the operating status and performance of the equipment. For an elevator, its main operating soundprint data may include the sound of the motor running, the sound of the chain or cable moving, etc., which are stable and predictable when the elevator is operating normally. Associated operating soundprint data Other sound data associated with the main operating soundprint data may come from other components or systems connected to or affecting the main equipment. Continuing with the elevator as an example, the associated operating soundprint data may include the sound of the sensor on the floor where the elevator is located, the sound of the power supply system connected to the elevator, etc. Although these data are not directly generated by the elevator, they are related to its operating status and may provide additional monitoring information.

[0028] In addition, the expert model is a model that has been pre-trained and has professional knowledge and capabilities, and is used to extract useful features from voiceprint data. In the scenario of monitoring the operating status of a stereo garage, the expert model can understand and identify the sound features related to the operating status of garage equipment. Multi-round voiceprint feature extraction is that the model processes the voiceprint data multiple times, and each processing extracts features at different levels. Each round of extraction may focus on different aspects of the sound, such as frequency, intensity, rhythm, etc. In the first round of extraction, the model may focus on the frequency characteristics of the sound and identify which frequency components are most significant in the normal operation of the garage equipment. In the second round of extraction, the model may turn to focus on the intensity changes of the sound and identify the differences in the sound intensity of the equipment at different operating stages. The first voiceprint feature dataset is a feature set extracted from the main operating voiceprint data. It contains multiple features of different granularity levels, which together describe the main operating status of the garage equipment. Suppose there is a lift, and its main operating voiceprint data is extracted with features of three granularity levels: frequency features, intensity features, and rhythm features. These features together constitute the first voiceprint feature dataset, which is used to describe the operating status of the lift. The second voiceprint feature data set is a feature set extracted from at least one associated running voiceprint data. Similar to the first voiceprint feature data set, it also contains features of multiple different granularity levels, but these features describe the operating status of other components or systems associated with the main equipment. Continuing with the elevator as an example, assume that there is also a power supply system connected to the elevator. Features extracted from the sound of the power supply system (such as sound features of voltage fluctuations, sound features of current changes, etc.) constitute the second voiceprint feature data set. Granularity levels refer to different levels or degrees of detail of features. In voiceprint feature extraction, granularity levels can represent different feature levels from coarse to fine. In sound features, frequency can be a granularity level that describes the basic tone of the sound. The change pattern of frequency (such as whether it is stable, whether there is periodic change, etc.) can be another finer granularity level. The number of rounds of voiceprint feature extraction is consistent with the number of granularity levels, which means that each round of voiceprint feature extraction corresponds to a specific granularity level. The number of rounds determines the richness of the feature details that can be extracted. If the expert model of the embodiment of the present invention is designed to perform three rounds of feature extraction, then each round will correspond to a different granularity level. The first round can extract the coarsest features (such as the main frequency components of the sound), the second round extracts more detailed features (such as the intensity change pattern of the sound), and the third round extracts the most subtle features (such as the tiny fluctuations and transient features of the sound).

[0029] In addition, feature integration processing is to merge feature data sets from different sources (here, the first voiceprint feature data set and the second voiceprint feature data set) into a more comprehensive feature set. This process involves the splicing, fusion or selection of features to ensure that all useful information is included. In the feature integration processing, the server will splice the motor running frequency feature of the elevator with the voltage fluctuation sound feature of the power system to form a comprehensive feature vector containing information from both. This process involves standardization or normalization of the features to ensure that the features from different sources are numerically comparable. The first comprehensive voiceprint feature data set is the result data set obtained through feature integration processing. It contains all the features from the first voiceprint feature data set and the second voiceprint feature data set, which is used to comprehensively describe the operating status of the garage equipment to be monitored. In the operating status monitoring scenario of the stereo garage, the first comprehensive voiceprint feature data set can include the motor running frequency, sound intensity, rhythm features of the elevator, as well as the voltage fluctuation and current change sound features of the power system. This comprehensive feature data set provides a comprehensive information basis for subsequent monitoring and analysis. Lightweight model A model with a simpler structure, fewer parameters, and lower computational requirements. In voiceprint feature processing, the lightweight model can efficiently process feature data while maintaining good performance. Feature reconstruction processing is a process in which the lightweight model uses the first voiceprint feature data set and the second voiceprint feature data set as input, and generates a new feature data set through a series of calculations and transformations. This process aims to refine and enhance the expressiveness of the features so as to better describe the operating status of the garage equipment to be monitored. In the feature reconstruction processing, the lightweight model will fuse or transform the motor operating frequency features of the elevator with the voltage fluctuation sound features of the power supply system in some form to generate a new feature vector. This new feature vector can contain information that is not in the original features, or highlight certain important features. The second comprehensive voiceprint feature data set is the result data set obtained through feature reconstruction processing. It contains the features processed by the lightweight model and is used to more comprehensively describe the operating status of the garage equipment to be monitored. In the operating status monitoring scenario of the stereo garage, the second comprehensive voiceprint feature data set can contain the elevator motor operating features, sound intensity features, and voltage fluctuations and current changes of the power supply system after being processed by the lightweight model. This comprehensive feature data set provides more refined and useful information for subsequent monitoring and analysis.

[0030] In addition, feature comparison processing refers to comparing and analyzing the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set to identify the similarities and differences between the two. This process is intended to evaluate whether the current operating state of the garage equipment to be monitored is consistent with the expected or historical state. In the feature comparison processing, the server compares the same features (such as the motor operating frequency of the elevator) in the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set, and analyzes their similarity. If the feature similarity in the two data sets is high, it means that the equipment is in a stable operating state; if the similarity is low, it may indicate that the equipment is abnormal. The equipment operating state evaluation result is the result obtained through feature comparison processing, indicating whether the current operating state of the garage equipment to be monitored is normal, whether there is an abnormality or failure, etc. The equipment operating state evaluation result can be a report containing multiple indicators, such as the similarity of the motor operating frequency, the range of sound intensity, etc. These indicators together constitute a comprehensive evaluation of the equipment operating state. The operating state of the stereo garage is a conclusion drawn based on the equipment operating state evaluation result, indicating whether the overall operating state of the stereo garage is normal, whether there are problems that need maintenance or repair, etc. If the equipment operation status evaluation results show that the operation status of all the monitored garage equipment is normal, then the operation status of the three-dimensional garage is normal. If any equipment is abnormal or fails, then the operation status of the three-dimensional garage is considered abnormal and requires further inspection and maintenance.

[0031] In the embodiment of the present invention, the step of performing feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain the first comprehensive voiceprint feature data set of the garage equipment to be monitored can be implemented through the following example.

[0032] Determining a mutual influence coefficient according to a first voiceprint feature and a second voiceprint feature, wherein the first voiceprint feature and the second voiceprint feature correspond to a granularity level of the same level, the first voiceprint feature is derived from the first voiceprint feature data set, and the second voiceprint feature is derived from the second voiceprint feature data set;

[0033] Multiplying the second voiceprint feature by the mutual influence coefficient to obtain a multiplication result;

[0034] Accumulating the multiplication result and the first voiceprint feature to obtain a first comprehensive voiceprint feature, wherein the first comprehensive voiceprint feature is classified into the first comprehensive voiceprint feature data set;

[0035] The first voiceprint feature data set and the second voiceprint feature data set are polled to obtain a first comprehensive voiceprint feature data set.

[0036] In an embodiment of the present invention, illustratively, the server first identifies features with the same or consistent granularity level in the first voiceprint feature data set and the second voiceprint feature data set. For example, the motor operating frequency feature of the elevator and the voltage fluctuation sound feature of the power system both belong to the granularity level of "frequency". Then, the server analyzes the mutual influence relationship between these features and determines the mutual influence coefficient between them. This coefficient reflects the degree to which another feature may be affected when one feature changes. For example, if an increase in the motor operating frequency usually causes a slight drop in the power system voltage, then the mutual influence coefficient between the two features is a negative number, indicating that there is a negative correlation between them. The server multiplies each feature in the second voiceprint feature data set with the corresponding mutual influence coefficient. The purpose of this operation is to adjust the value of the second voiceprint feature to reflect the mutual influence between them and the first voiceprint feature. Taking the voltage fluctuation sound feature of the power system as an example, if the mutual influence coefficient between it and the motor operating frequency feature is -0.5, then the server will multiply the value of the voltage fluctuation sound feature by -0.5 to obtain the multiplication result. The server adds the multiplication result obtained by the multiplication operation to the corresponding feature in the first voiceprint feature data set. The purpose of this operation is to merge the adjusted second voiceprint feature with the original first voiceprint feature to form a comprehensive feature representation. Continuing with the example of the motor operating frequency feature and the voltage fluctuation sound feature, the server will add the value of the adjusted voltage fluctuation sound feature to the value of the motor operating frequency feature to obtain the first comprehensive voiceprint feature. The server repeats the above steps, polling all features in the first voiceprint feature data set and the second voiceprint feature data set until a complete first comprehensive voiceprint feature data set is obtained. This data set contains all the features that have been adjusted and merged after mutual influence, and is used to comprehensively describe the operating status of the garage equipment to be monitored. For example, in addition to the motor operating frequency and voltage fluctuation sound features, other comprehensive features such as sound intensity and rhythm can also be included. Through the above steps, the server can obtain the first comprehensive voiceprint feature data set of the garage equipment to be monitored based on the integrated processing of the voiceprint features, providing basic data for subsequent equipment operating status evaluation and stereo garage operating status monitoring.

[0037] In an embodiment of the present invention, the step of performing feature reconstruction processing based on the first voiceprint feature dataset and the second voiceprint feature dataset according to a lightweight model to obtain a second comprehensive voiceprint feature dataset can be implemented through the following example.

[0038] Pre-adjusting the third voiceprint feature and the fourth voiceprint feature respectively to obtain a first pre-adjusted voiceprint feature and a second pre-adjusted voiceprint feature, wherein the third voiceprint feature and the fourth voiceprint feature have corresponding granularity levels of the same level, the third voiceprint feature is a voiceprint feature obtained by the expert model performing a subsequent round of voiceprint feature extraction on the main operation voiceprint data, and the fourth voiceprint feature is a voiceprint feature obtained by the expert model performing a subsequent round of voiceprint feature extraction on the at least one associated operation voiceprint data; the subsequent round of voiceprint feature extraction is a next round of voiceprint feature extraction after the last round of voiceprint feature extraction of the multiple rounds of voiceprint feature extraction;

[0039] Performing feature integration processing on the first pre-adjusted voiceprint feature and the second pre-adjusted voiceprint feature to obtain a first pre-adjusted comprehensive voiceprint feature;

[0040] Performing feature reconstruction processing based on the first pre-adjusted integrated voiceprint feature to obtain a second integrated voiceprint feature, wherein the second integrated voiceprint feature is derived from the second integrated voiceprint feature data set, and the granularity level of the second integrated voiceprint feature is consistent with the granularity level of the integrated voiceprint feature output by the expert model for multiple rounds;

[0041] Pre-adjusting the fifth voiceprint feature and the sixth voiceprint feature respectively to obtain a third pre-adjusted voiceprint feature and a fourth pre-adjusted voiceprint feature, wherein the fifth voiceprint feature and the sixth voiceprint feature have corresponding granularity levels of the same level, the fifth voiceprint feature is a voiceprint feature obtained by the expert model performing multiple rounds of voiceprint feature extraction on the main running voiceprint data, and the sixth voiceprint feature is a voiceprint feature obtained by the expert model performing multiple rounds of voiceprint feature extraction on the at least one associated running voiceprint data;

[0042] Performing feature integration processing on the third pre-adjusted voiceprint feature and the fourth pre-adjusted voiceprint feature to obtain a second pre-adjusted comprehensive voiceprint feature;

[0043] Performing feature reconstruction processing based on the second pre-adjusted comprehensive voiceprint feature to obtain a third comprehensive voiceprint feature, the third comprehensive voiceprint feature originates from the second comprehensive voiceprint feature data set, and the granularity level of the third comprehensive voiceprint feature is consistent with the granularity level of the comprehensive voiceprint feature output by the preceding voiceprint feature extraction of the expert model; the preceding voiceprint feature extraction is a previous round of voiceprint feature extraction of the last round of voiceprint feature extraction of the multiple rounds of voiceprint feature extraction;

[0044] The above process is executed repeatedly to obtain the second comprehensive voiceprint feature data set.

[0045] In the embodiment of the present invention, exemplarily, the server first performs pre-adjustment processing on the third voiceprint feature and the fourth voiceprint feature. These voiceprint features are obtained by extracting the main operation voiceprint data and the associated operation voiceprint data in subsequent rounds (i.e., the rounds after the last round of multiple rounds of voiceprint feature extraction) through the expert model. For example, the server can pre-adjust the motor operation frequency feature (third voiceprint feature) of the elevator and the voltage fluctuation sound feature (fourth voiceprint feature) of the power supply system, such as removing noise, standardization, etc., to obtain the first pre-adjusted voiceprint feature and the second pre-adjusted voiceprint feature. Then, the server performs feature integration processing on the first pre-adjusted voiceprint feature and the second pre-adjusted voiceprint feature. This process may include splicing, fusion or selection of features to ensure that all useful information is included. For example, the server will splice the pre-adjusted motor operation frequency feature with the pre-adjusted voltage fluctuation sound feature to form a first pre-adjusted comprehensive voiceprint feature containing information of both. Then, the server performs feature reconstruction processing based on the first pre-adjusted comprehensive voiceprint feature. This process involves further refining and enhancing the features using a lightweight model to obtain a more refined and useful second comprehensive voiceprint feature. For example, the lightweight model will perform some form of transformation or encoding on the first pre-adjusted comprehensive voiceprint feature to generate a second comprehensive voiceprint feature, the granularity level of which is consistent with the granularity level of the comprehensive voiceprint features output by the expert model in multiple rounds. The server repeats the above steps, but this time the fifth and sixth voiceprint features are pre-adjusted. These voiceprint features are obtained by performing multiple rounds of voiceprint feature extraction on the main operating voiceprint data and the associated operating voiceprint data by the expert model. For example, the server can perform the same pre-adjustment processing on the motor running frequency feature (fifth voiceprint feature) and the voltage fluctuation sound feature (sixth voiceprint feature) of the power system extracted in another round. Next, the server performs feature integration processing on the third pre-adjusted voiceprint feature and the fourth pre-adjusted voiceprint feature to obtain the second pre-adjusted comprehensive voiceprint feature. This process is similar to the process of obtaining the first pre-adjusted comprehensive voiceprint feature, but uses voiceprint features from different rounds. Then, the server performs feature reconstruction processing based on the second pre-adjusted comprehensive voiceprint feature to obtain a third comprehensive voiceprint feature. The granularity level of this feature is consistent with the granularity level of the comprehensive voiceprint feature extracted and output by the expert model's previous voiceprint feature. For example, the lightweight model will perform a different transformation or encoding on the second pre-adjusted comprehensive voiceprint feature to generate a third comprehensive voiceprint feature. The server repeats the above process to perform pre-adjustment, feature integration, and feature reconstruction on different voiceprint features until a complete second comprehensive voiceprint feature data set is obtained. This data set contains all the comprehensive voiceprint features that have been pre-adjusted, integrated, and reconstructed, and is used to more comprehensively describe the operating status of the garage equipment to be monitored.Through the above steps, the server can obtain the second comprehensive voiceprint feature data set of the garage equipment to be monitored based on the lightweight model and feature reconstruction processing, providing more refined data support for subsequent equipment operation status evaluation and three-dimensional parking garage operation status monitoring.

[0046] In the embodiments of the present invention, the following implementation modes are also provided.

[0047] Acquire a sample instance data set and an original lightweight model, wherein the sample instance data set includes target voiceprint instance data and associated running voiceprint data;

[0048] Performing multiple rounds of voiceprint feature extraction on the target voiceprint instance data based on the expert model to obtain multiple voiceprint features of the target voiceprint instance data;

[0049] Performing multiple rounds of voiceprint feature extraction on the associated operation voiceprint data based on the expert model to obtain multiple voiceprint features of the associated operation voiceprint data;

[0050] Performing feature integration processing on the multiple voiceprint features of the target voiceprint instance data and the multiple voiceprint features of the associated running voiceprint data to obtain multiple training comprehensive voiceprint features;

[0051] Inputting a plurality of voiceprint features of the target voiceprint instance data and a plurality of voiceprint features of the associated running voiceprint data into the original lightweight model to perform feature reconstruction processing to obtain a plurality of reconstructed voiceprint features;

[0052] Perform feature comparison processing according to the multiple reconstructed voiceprint features and the multiple trained comprehensive voiceprint features to obtain a cost parameter;

[0053] A training process is performed on the original lightweight model according to the cost parameter to obtain the lightweight model.

[0054] In the embodiment of the present invention, exemplarily, the server first obtains a sample instance data set from a database, the data set including target voiceprint instance data and associated operation voiceprint data. For example, the target voiceprint instance data may be the sound emitted by a normally operating elevator, and the associated operation voiceprint data may be the sound emitted by a power supply system or a lighting system that is running at the same time. At the same time, the server loads an original lightweight model, which is untrained and is used for subsequent feature reconstruction processing. The server uses an expert model to perform multiple rounds of voiceprint feature extraction on the target voiceprint instance data. Each round of extraction will extract different voiceprint features. For example, the first round can extract the motor running frequency feature, and the second round can extract the sound intensity feature. The server also uses the expert model to perform multiple rounds of voiceprint feature extraction on the associated operation voiceprint data. For example, for the power supply system, the voltage fluctuation sound feature and the current change sound feature can be extracted. The server performs feature integration processing on multiple voiceprint features of the target voiceprint instance data and multiple voiceprint features of the associated operation voiceprint data. This process may include feature splicing, fusion or selection. For example, the motor running frequency feature and the voltage fluctuation sound feature are spliced ​​to form a training comprehensive voiceprint feature containing information of both. The server inputs multiple voiceprint features of the target voiceprint instance data and multiple voiceprint features of the associated running voiceprint data into the original lightweight model and performs feature reconstruction processing. This process aims to further refine and enhance the features through the model. For example, the model will perform some form of transformation or encoding on the input features to generate reconstructed voiceprint features. The server performs feature comparison processing on multiple reconstructed voiceprint features and multiple training comprehensive voiceprint features. This process aims to evaluate the similarity or difference between the reconstructed voiceprint features and the training comprehensive voiceprint features. For example, a certain distance metric (such as Euclidean distance) can be used to calculate the difference between them as a cost parameter. Finally, the server performs a training process on the original lightweight model according to the cost parameter. This process aims to adjust the parameters of the model so that the difference between the reconstructed voiceprint features and the training comprehensive voiceprint features is minimized. After multiple iterations of training, the server obtains a trained lightweight model that can be used for feature reconstruction processing in subsequent monitoring of the operation status of a stereo garage.

[0055] In the embodiments of the present invention, the following implementation modes are also provided.

[0056] Obtain a first sample instance; wherein the first sample instance includes first target voiceprint instance data and a corresponding voiceprint target value;

[0057] Performing voiceprint semantic information construction processing on the first target voiceprint instance data to obtain voiceprint semantic information of the first target voiceprint instance data;

[0058] Integrate the voiceprint target value of the first target voiceprint instance data with the voiceprint semantic information to obtain integrated description information;

[0059] Perform voiceprint reverse construction processing according to the integrated description information to obtain second target voiceprint instance data;

[0060] Constructing a second sample instance according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data;

[0061] The sample instance data set is constructed according to the first sample instance and the second sample instance.

[0062] In the embodiment of the present invention, exemplarily, the server obtains a first sample instance from the database, the instance including the first target voiceprint instance data and the corresponding voiceprint target value. For example, the first target voiceprint instance data may be the sound emitted by a normally operating elevator, and the corresponding voiceprint target value may be a label indicating the normal operating state of the elevator. The server performs voiceprint semantic information construction processing on the first target voiceprint instance data. This process aims to extract semantic information from the sound, that is, meaningful sound features related to the operating state of the elevator. For example, the server will identify semantic information such as the stability of the motor operation and the rhythm of the sound. The server integrates the voiceprint target value of the first target voiceprint instance data with the voiceprint semantic information. This process may include associating or splicing the voiceprint target value (such as the label of the normal operating state) with the extracted semantic information (such as the stability of the motor operation) to form an integrated description information containing both information. The server performs voiceprint reverse construction processing based on the integrated description information. This process aims to generate a new voiceprint instance data, which is semantically similar to the original first target voiceprint instance data, but may contain some minor changes or noise. For example, the server generates a simulated elevator operation sound that is similar to the original sound as a whole, but contains some minor changes caused by environmental factors or equipment aging. The server constructs a second sample instance based on the second target voiceprint instance data obtained by the voiceprint reverse construction process and the voiceprint target value of the first target voiceprint instance data. This second sample instance is consistent with the first sample instance in the voiceprint target value, but different in the voiceprint instance data because the second target voiceprint instance data is obtained through reverse construction. Finally, the server constructs a sample instance dataset based on the first sample instance and the second sample instance. This dataset contains a variety of voiceprint instance data, including both original voiceprint data and simulated voiceprint data obtained through reverse construction, all of which are associated with the same voiceprint target value. Such a dataset can be used to train a more robust and generalized lightweight model to better cope with the challenges in actual stereo garage operation status monitoring.

[0063] In the embodiment of the present invention, the step of performing voiceprint reverse construction processing according to the integrated description information to obtain the second target voiceprint instance data can be implemented through the following example.

[0064] Acquire a preset voiceprint feature of first interfering voiceprint data;

[0065] Performing a de-interference operation on the voiceprint features of the first interfering voiceprint data according to a de-interference filtering strategy to obtain the voiceprint features of the second target voiceprint instance data; wherein the de-interference filtering strategy includes the semantic description features of the integrated description information;

[0066] The second target voiceprint instance data is derived according to the voiceprint features of the second target voiceprint instance data.

[0067] In the embodiment of the present invention, exemplarily, the server obtains the preset first interference voiceprint data from the database, which may be the elevator operation sound collected under different environmental conditions (such as temperature and humidity changes) or different equipment states (such as equipment aging and damage). The server uses an expert model or feature extraction algorithm to extract the voiceprint features of the first interference voiceprint data to obtain its voiceprint features. These features may include the frequency, amplitude, rhythm, etc. of the sound. The server performs a de-interference operation on the voiceprint features of the first interference voiceprint data according to the de-interference filtering strategy. This de-interference filtering strategy is constructed based on the semantic description features of the integrated description information, and is intended to remove or weaken those sound features that are inconsistent with the normal elevator operation state. For example, if the integrated description information indicates that the sound of a normal elevator should be stable and have a uniform rhythm, then the de-interference filtering strategy will remove those sound features with large frequency fluctuations and uneven rhythms. After removing the interference, the server uses voiceprint synthesis technology or a reverse construction algorithm to derive the second target voiceprint instance data based on the voiceprint features of the remaining second target voiceprint instance data. This derivation process involves converting the voiceprint features into actual sound waveforms, or generating simulated elevator operation sounds through some form of voiceprint reconstruction. Finally, the server obtains the second target voiceprint instance data, which is a simulated sound data that is semantically similar to the original first target voiceprint instance data but contains some minor changes caused by environmental factors or device status. Through such voiceprint reverse construction processing, the server can generate diverse voiceprint instance data for training a more robust and generalized lightweight model to better cope with various challenges in actual parking garage operation status monitoring.

[0068] In the embodiment of the present invention, before performing the interference removal operation on the voiceprint features of the first interfering voiceprint data according to the interference removal filtering strategy, the embodiment of the present invention further provides the following implementation manner.

[0069] Performing interference adding processing on the voiceprint features of the first target voiceprint instance data to obtain the voiceprint features of the second interference voiceprint data;

[0070] Performing a de-interference operation on the voiceprint features of the second interfering voiceprint data according to the de-interference filtering strategy through a de-interference model to obtain the voiceprint features of the de-interference voiceprint data;

[0071] Determine a cost parameter according to the interference information added corresponding to the interference adding process and the interference information inferred by the interference removal operation, and train a interference removal model according to the cost parameter;

[0072] The step of performing a de-interference operation on the voiceprint features of the first interfering voiceprint data according to the de-interference filtering strategy to obtain the voiceprint features of the second target voiceprint instance data includes:

[0073] The voiceprint features of the first interfering voiceprint data are de-interfered by the trained de-interference model according to the de-interference filtering strategy to obtain the voiceprint features of the second target voiceprint instance data.

[0074] In an embodiment of the present invention, illustratively, the server first obtains the first target voiceprint instance data, which is the sound data emitted by the elevator in normal operation. The server uses an expert model or a feature extraction algorithm to extract the voiceprint features of the first target voiceprint instance data. Then, the server performs interference addition processing on these voiceprint features to simulate the interference that may be introduced under different environmental conditions or in different states of the device. For example, the server will add some background noise, frequency fluctuations or rhythm changes to simulate the interference that may be encountered in actual situations. The server uses a pre-built interference removal model to perform interference removal operations on the voiceprint features of the second interference voiceprint data after adding interference according to the interference removal filtering strategy. This interference removal filtering strategy is constructed based on the semantic description features of the integrated description information, and is intended to remove or weaken those sound features that are inconsistent with the normal operation state of the elevator. The interference removal model will try to identify and remove those simulated interferences, such as background noise, frequency fluctuations, etc., to restore the voiceprint features close to the original first target voiceprint instance data. The server compares the interference information corresponding to the interference addition processing with the interference information inferred by the interference removal operation, and determines a cost parameter. This cost parameter reflects the performance of the interference removal model in removing interference. The server uses this cost parameter to train the de-interference model, and improves the accuracy and robustness of the model in removing interference through an iterative optimization process. After training the de-interference model, the server uses the model to perform de-interference operations on the voiceprint features of the first interference voiceprint data according to the de-interference filtering strategy. This first interference voiceprint data is sound data containing various interferences that may be encountered in actual collection. The de-interference model attempts to remove or weaken these interferences to obtain the voiceprint features of the second target voiceprint instance data that is close to the normal elevator operation state. Through such a processing flow, the server can generate a purer second target voiceprint instance data that is close to the normal elevator operation state for subsequent voiceprint reverse construction processing and model training process. This helps to improve the accuracy and robustness of the stereo garage operation status monitoring system.

[0075] In an embodiment of the present invention, the de-interference model includes a first feature combination unit; the step of performing a de-interference operation on the voiceprint features of the second interfering voiceprint data according to the de-interference filtering strategy through the de-interference model to obtain the voiceprint features of the de-interference voiceprint data can be implemented through the following example.

[0076] Performing a feature combination operation on the semantic description features of the integrated description information and the voiceprint features of the second interfering voiceprint data through a first feature combination unit in the interference removal model, so as to infer and filter the interference information in the voiceprint features of the second interfering voiceprint data, and obtain the voiceprint features of the interference-removed voiceprint data;

[0077] The step of training the interference removal model according to the cost parameter comprises:

[0078] The first feature combination unit in the interference removal model is trained according to the cost parameter.

[0079] In the embodiment of the present invention, exemplarily, the server first obtains the semantic description features of the integrated description information, which describe the voiceprint features of the normal elevator operation state. The server also obtains the voiceprint features of the second interference voiceprint data, which are obtained by adding interference to the voiceprint features of the first target voiceprint instance data. The server uses the first feature combination unit in the de-interference model to perform a feature combination operation on the semantic description features of the integrated description information and the voiceprint features of the second interference voiceprint data. This operation is intended to compare and combine the normal voiceprint features with the voiceprint features containing interference, so as to infer which parts are interference information. Through the feature combination operation, the first feature combination unit can identify and filter out the interference information in the voiceprint features of the second interference voiceprint data, thereby obtaining the voiceprint features of the de-interference voiceprint data. The server uses the cost parameter determined previously to train the first feature combination unit in the de-interference model. The training process includes adjusting the parameters and algorithms of the first feature combination unit to more accurately identify and filter out the interference information. Through the iterative training process, the server continuously optimizes the performance of the first feature combination unit until a satisfactory de-interference effect is achieved. Through such a processing flow, the server can effectively remove the interference information in the second interference voiceprint data by using the first feature combination unit in the interference removal model, and obtain a purer voiceprint feature of the interference-free voiceprint data. These voiceprint features after interference removal can be used for subsequent voiceprint reverse construction processing and model training process, further improving the accuracy and robustness of the three-dimensional parking garage operation status monitoring system.

[0080] In an embodiment of the present invention, the de-interference filtering strategy also includes the voiceprint features of the first target voiceprint instance data; the de-interference model includes a first feature combination unit and a second feature combination unit; the step of performing a de-interference operation on the voiceprint features of the second interference voiceprint data according to the de-interference filtering strategy through the de-interference model to obtain the voiceprint features of the de-interference voiceprint data can be implemented through the following example.

[0081] Performing a feature combination operation on the semantic description feature of the integrated description information and the voiceprint feature of the second interfering voiceprint data through a first feature combination unit in the interference removal model, so as to infer and filter the interference information in the voiceprint feature of the second interfering voiceprint data, and obtain a first sub-voiceprint feature of the interference-removed voiceprint data;

[0082] Performing a feature combination operation on the voiceprint features of the first target voiceprint instance data and the voiceprint features of the second interference voiceprint data through a second feature combination unit in the interference removal model, so as to infer and filter interference information in the voiceprint features of the second interference voiceprint data, and obtain a second sub-voiceprint feature of the interference-removed voiceprint data;

[0083] Performing feature fusion processing on the first sub-voiceprint feature of the de-interferenced voiceprint data and the second sub-voiceprint feature to obtain the voiceprint feature of the de-interferenced voiceprint data;

[0084] The step of training the interference removal model according to the cost parameter comprises:

[0085] The second feature combination unit in the interference removal model is trained according to the cost parameter.

[0086] In the embodiment of the present invention, exemplarily, the server obtains the semantic description features of the integrated description information, which describe the voiceprint features of the normal elevator operation state. The server also obtains the voiceprint features of the second interference voiceprint data, which are obtained by adding interference to the voiceprint features of the first target voiceprint instance data. The server uses the first feature combination unit in the de-interference model to perform a feature combination operation on the semantic description features of the integrated description information and the voiceprint features of the second interference voiceprint data. This operation is intended to compare and combine the normal voiceprint features with the voiceprint features containing interference, so as to infer which parts are interference information. Through the feature combination operation, the first feature combination unit can identify and filter out part of the interference information in the voiceprint features of the second interference voiceprint data, thereby obtaining the first sub-voiceprint features of the de-interference voiceprint data. The server obtains the voiceprint features of the first target voiceprint instance data, which are the features of the sound data emitted by the normal elevator. The server uses the voiceprint features of the second interference voiceprint data again. The server uses the second feature combination unit in the de-interference model to perform a feature combination operation on the voiceprint features of the first target voiceprint instance data and the voiceprint features of the second interference voiceprint data. This operation is intended to further compare and combine voiceprint features in order to infer more interference information. Through the feature combination operation, the second feature combination unit can identify and filter out more interference information in the voiceprint features of the second interfering voiceprint data, thereby obtaining the second sub-voiceprint features of the de-interfering voiceprint data. The server performs feature fusion processing on the first sub-voiceprint features of the de-interfering voiceprint data and the second sub-voiceprint features. This processing process is intended to merge the useful information in the two sub-voiceprint features to obtain more complete and accurate voiceprint features of the de-interfering voiceprint data. Through feature fusion processing, the server obtains voiceprint features of the de-interfering voiceprint data, which are closer to the voiceprint features of the normal elevator operation state. The server uses the previously determined cost parameters to train the second feature combination unit in the de-interference model. The training process includes adjusting the parameters and algorithms of the second feature combination unit to more accurately identify and filter out interference information. Through the iterative training process, the server continuously optimizes the performance of the second feature combination unit until a satisfactory de-interference effect is achieved. Through such a processing flow, the server can effectively remove the interference information in the second interference voiceprint data by using the first feature combination unit and the second feature combination unit in the interference removal model, and obtain the voiceprint features of the purer interference removal voiceprint data. These voiceprint features after interference removal can be used for the subsequent voiceprint reverse construction processing and model training process, further improving the accuracy and robustness of the three-dimensional parking garage operation status monitoring system.

[0087] In the embodiment of the present invention, before the second feature combination unit in the interference removal model performs a feature combination operation on the voiceprint features of the first target voiceprint instance data and the voiceprint features of the second interference voiceprint data, the embodiment of the present invention further provides the following implementation manner.

[0088] Performing a feature mapping operation on the voiceprint features of the first target voiceprint instance data by a feature mapping unit; wherein the voiceprint features after feature mapping of the first target voiceprint instance data are used to perform a feature combination operation;

[0089] The method further comprises:

[0090] The feature mapping unit is trained according to the cost parameter.

[0091] In the embodiment of the present invention, exemplarily, the server first obtains the voiceprint features of the first target voiceprint instance data, which are the features of the sound data emitted by the normally operating elevator. The server uses the feature mapping unit to perform a feature mapping operation on the voiceprint features of the first target voiceprint instance data. This operation is intended to convert the original voiceprint features into a new feature space so as to better compare and combine with the voiceprint features of the second interference voiceprint data. Through the feature mapping operation, the server obtains the voiceprint features of the first target voiceprint instance data after feature mapping. The server obtains the voiceprint features of the second interference voiceprint data, which are obtained by adding interference to the voiceprint features of the first target voiceprint instance data. The server uses the second feature combination unit in the de-interference model to perform a feature combination operation on the voiceprint features of the first target voiceprint instance data after feature mapping and the voiceprint features of the second interference voiceprint data. This operation is intended to further compare and combine the voiceprint features so as to infer more interference information. Through the feature combination operation, the second feature combination unit can identify and filter out more interference information in the voiceprint features of the second interference voiceprint data, thereby obtaining the second sub-voiceprint features of the de-interference voiceprint data. The server uses the cost parameter determined previously to train the feature mapping unit. The training process includes adjusting the parameters and algorithms of the feature mapping unit in order to generate feature-mapped voiceprint features that are more suitable for comparison and combination with the voiceprint features of the second interfering voiceprint data. Through the iterative training process, the server continuously optimizes the performance of the feature mapping unit until a satisfactory feature mapping effect is achieved. Through such a processing flow, the server can use the feature mapping unit to convert the voiceprint features of the first target voiceprint instance data into a feature space that is more suitable for comparison and combination with the voiceprint features of the second interfering voiceprint data. Then, the server uses the second feature combination unit in the de-interference model to effectively remove the interference information in the second interfering voiceprint data, and obtains purer voiceprint features of the de-interfering voiceprint data. These de-interfered voiceprint features can be used in subsequent voiceprint reverse construction processing and model training processes to further improve the accuracy and robustness of the stereo parking garage operation status monitoring system.

[0092] In the embodiment of the present invention, the step of performing interference adding processing on the voiceprint features of the first target voiceprint instance data to obtain the voiceprint features of the second interference voiceprint data can be implemented through the following example.

[0093] Execute a multi-cycle interference adding loop and perform the following process in the target cycle interference adding loop process:

[0094] Add interference information to the voiceprint features of the target period adding interference cycle input, and obtain the voiceprint features of the subsequent period adding interference cycle input; the subsequent period adding interference cycle is the period adding interference cycle after the last period adding interference cycle in the multi-period adding interference cycle;

[0095] The voiceprint features of the first target voiceprint instance data are used as the voiceprint features of the first periodic interference cycle input; the voiceprint features obtained by adding interference cycles to the target periodicity are the voiceprint features of the second interference voiceprint data;

[0096] The step of performing a de-interference operation on the voiceprint features of the second interfering voiceprint data according to the de-interference filtering strategy using the de-interference model to obtain the voiceprint features of the de-interference voiceprint data includes:

[0097] Execute a multi-cycle interference removal loop and perform the following processing during the target cycle interference removal loop:

[0098] Inferring interference information in the voiceprint features of the target period de-interference cycle input according to the de-interference filtering strategy through the de-interference model, and filtering the inferred interference information in the voiceprint features of the target period de-interference cycle input to obtain the voiceprint features of the subsequent period de-interference cycle input;

[0099] Among them, the voiceprint features of the second interference voiceprint data are used as the voiceprint features input into the first periodic interference removal loop; the voiceprint features obtained in the target periodic interference removal loop are the voiceprint features of the interference-removed voiceprint data.

[0100] In an embodiment of the present invention, exemplarily, the server obtains the voiceprint feature of the first target voiceprint instance data, which is the feature of the sound data emitted by the normal operation of the elevator. The server uses this voiceprint feature as the voiceprint feature input for the first cycle of adding interference. The server starts to execute a multi-cycle adding interference cycle. In each target cycle adding interference cycle, the server adds interference information to the voiceprint feature currently input. After adding the interference information, the server obtains the voiceprint feature input for the subsequent cycle adding interference cycle. The server continues to execute the next cycle adding interference cycle until the preset number of cycles is reached. In the last target cycle adding interference cycle, the voiceprint feature obtained by the server is the voiceprint feature of the second interference voiceprint data. The server uses the voiceprint feature of the second interference voiceprint data as the voiceprint feature input for the first cycle removing interference cycle. The server starts to execute a multi-cycle removing interference cycle. In each target cycle removing interference cycle, the server uses the removing interference model to infer the interference information in the voiceprint feature currently input according to the removing interference filtering strategy. The server filters the inferred interference information in the voiceprint feature currently input and obtains the voiceprint feature input for the subsequent cycle removing interference cycle. The server continues to execute the next cycle removing interference cycle until the preset number of cycles is reached. In the last target cycle de-interference cycle, the voiceprint features obtained by the server are the voiceprint features of the de-interference voiceprint data. Through such a processing flow, the server can perform interference addition processing on the voiceprint features of the first target voiceprint instance data in a simulated interference environment, and obtain the voiceprint features of the second interference voiceprint data. Then, the server uses the de-interference model to perform multi-cycle de-interference operations on the voiceprint features of the second interference voiceprint data according to the de-interference filtering strategy, and finally obtains the voiceprint features of the de-interference voiceprint data. These de-interferenced voiceprint features can be used for subsequent voiceprint reverse construction processing and model training processes to further improve the accuracy and robustness of the stereo garage operation status monitoring system.

[0101] In the embodiment of the present invention, the embodiment of the present invention also provides the following implementation manner.

[0102] Performing voiceprint conversion on the first target voiceprint instance data by a first voiceprint conversion component to obtain voiceprint features of the first target voiceprint instance data;

[0103] Performing semantic conversion processing on the integrated description information through a semantic conversion component to obtain semantic description features of the integrated description information;

[0104] The deducing the second target voiceprint instance data according to the voiceprint features of the second target voiceprint instance data includes:

[0105] The voiceprint inverse conversion component corresponding to the first voiceprint conversion component performs voiceprint inverse conversion processing on the voiceprint features of the second target voiceprint instance data to obtain the second target voiceprint instance data.

[0106] In an embodiment of the present invention, exemplarily, the server obtains the first target voiceprint instance data from the operation status monitoring system of the stereo garage, which is the sound data emitted by the normally operating elevator. The server uses the first voiceprint conversion component to perform voiceprint conversion processing on the first target voiceprint instance data. The voiceprint conversion processing is intended to convert the original sound data into a voiceprint feature that can represent its characteristics. Through the voiceprint conversion processing, the server obtains the voiceprint feature of the first target voiceprint instance data. The server obtains the integrated description information, which is a collection of sound features describing the normal operation status of the elevator. The server uses the semantic conversion component to perform semantic conversion processing on the integrated description information. The semantic conversion processing is intended to convert the integrated description information into a semantic description feature that can represent its semantics. Through the semantic conversion processing, the server obtains the semantic description feature of the integrated description information. The server has obtained the voiceprint feature of the second target voiceprint instance data in the previous processing step, which is the pure voiceprint feature obtained after the interference removal operation. The server uses the voiceprint inverse conversion component corresponding to the first voiceprint conversion component to perform voiceprint inverse conversion processing on the voiceprint feature of the second target voiceprint instance data. The voiceprint inverse conversion processing is intended to convert the voiceprint feature back to the original sound data. Through the voiceprint inverse conversion processing, the server obtains the second target voiceprint instance data, which is the sound data after interference removal, and can be used for further analysis and processing of the three-dimensional parking garage operation status monitoring system. Through such a processing flow, the server can perform voiceprint conversion processing on the first target voiceprint instance data to obtain its voiceprint features; perform semantic conversion processing on the integrated description information to obtain its semantic description features; and convert the voiceprint features of the second target voiceprint instance data back to the original sound data through voiceprint inverse conversion processing. These processing steps provide the three-dimensional parking garage operation status monitoring system with more accurate and robust sound data processing and analysis capabilities.

[0107] In the embodiment of the present invention, the training of the interference removal model according to the cost parameter may be implemented through the following example.

[0108] Perform any of the following:

[0109] Training adjustment parameters according to the cost parameters, and optimizing network parameters of the interference removal model according to the adjustment parameters;

[0110] Training a first sub-adjustment parameter and a second sub-adjustment parameter according to the cost parameter, integrating the first sub-adjustment parameter and the second sub-adjustment parameter to obtain an adjustment parameter, and optimizing a network parameter of a de-interference model according to the adjustment parameter;

[0111] The dimension of the adjustment parameter is consistent with the dimension of the network parameter.

[0112] In an embodiment of the present invention, exemplarily, the server obtains cost parameters from the historical data of the stereo garage operation status monitoring system, which reflect the performance and error of the model in the de-interference process. The server uses the cost parameters to train the adjustment parameters. This process involves initialization and iterative update of the adjustment parameters to minimize the cost function. The dimension of the adjustment parameters is consistent with the dimension of the network parameters of the de-interference model to ensure that they can be optimized accordingly. The server optimizes the network parameters of the de-interference model according to the trained adjustment parameters. The optimization process may include algorithms such as gradient descent and back propagation to update the network parameters and reduce the model error. Similarly, the server obtains the cost parameters from the historical data of the stereo garage operation status monitoring system. The server uses the cost parameters to train the first sub-adjustment parameter and the second sub-adjustment parameter respectively. The training process of these two sub-adjustment parameters involves different algorithms or strategies, but their goal is to jointly optimize the de-interference model. The server integrates the trained first sub-adjustment parameter and the second sub-adjustment parameter to obtain the final adjustment parameter. The integration process involves weighted averaging, splicing or other mathematical operations to ensure that the integrated adjustment parameter has a suitable dimension and value range. Finally, the server optimizes the network parameters of the de-interference model based on the integrated adjustment parameters. The optimization process is the same as in the first processing step, aiming to reduce model errors and improve de-interference performance. Through such a processing flow, the server can effectively train the de-interference model based on the cost parameters and improve the performance and accuracy of the model by optimizing the network parameters. This will help the three-dimensional parking garage operation status monitoring system to more accurately identify and analyze the operation status of the elevator, thereby improving the overall performance and reliability of the system.

[0113] In the embodiment of the present invention, the voiceprint semantic information construction process is performed on the first target voiceprint instance data to obtain the voiceprint semantic information of the first target voiceprint instance data, which can be implemented through the following examples.

[0114] Performing voiceprint conversion on the first target voiceprint instance data through a second voiceprint conversion component to obtain voiceprint features of the first target voiceprint instance data;

[0115] Extracting main voiceprint features from the voiceprint features of the first target voiceprint instance data through a deep learning model;

[0116] The semantic recognition model is used to perform semantic construction processing according to the main voiceprint features to obtain voiceprint semantic information of the first target voiceprint instance data.

[0117] In an embodiment of the present invention, exemplarily, the server obtains the first target voiceprint instance data from the operation status monitoring system of the stereo garage, which is the sound data emitted by the normally operating elevator. The server uses the second voiceprint conversion component to perform voiceprint conversion processing on the first target voiceprint instance data. The voiceprint conversion processing is intended to convert the original sound data into voiceprint features that can represent its characteristics. Through the voiceprint conversion processing, the server obtains the voiceprint features of the first target voiceprint instance data. The server uses a deep learning model to extract the main voiceprint features from the voiceprint features of the first target voiceprint instance data. The deep learning model can automatically learn and identify the key information in the voiceprint features, thereby extracting the main voiceprint features. Through the extraction of the deep learning model, the server obtains the main voiceprint features of the first target voiceprint instance data. The server uses a semantic recognition model to perform semantic construction processing based on the main voiceprint features. The semantic recognition model can convert the main voiceprint features into corresponding semantic information, that is, voiceprint semantic information. Through the semantic construction processing, the server obtains the voiceprint semantic information of the first target voiceprint instance data. Through such a processing flow, the server can perform voiceprint semantic information construction processing on the first target voiceprint instance data and obtain its voiceprint semantic information. These voiceprint semantic information describes the sound characteristics of the elevator when it is operating normally, and can be used for subsequent monitoring and analysis of the operating status of the stereo garage. For example, the server can compare the voiceprint semantic information with the real-time collected elevator sound data to determine whether the elevator is operating normally.

[0118] In the embodiment of the present invention, before constructing the second sample instance according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data, the embodiment of the present invention further provides the following implementation manner.

[0119] Comparing the second target voiceprint instance data with the voiceprint target value of the first target voiceprint instance data to obtain a comparison correlation coefficient;

[0120] The step of constructing a second sample instance according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data includes:

[0121] When the comparison correlation coefficient is greater than a comparison correlation coefficient threshold, a second sample instance is constructed according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data.

[0122] In an embodiment of the present invention, exemplarily, the server obtains the second target voiceprint instance data from the operation status monitoring system of the stereo garage, which is the elevator sound data collected in real time. The server simultaneously obtains the voiceprint target value of the first target voiceprint instance data, which is the voiceprint feature representation of the normal operation elevator sound data. The server compares the second target voiceprint instance data with the voiceprint target value of the first target voiceprint instance data. The comparison process is intended to calculate the similarity or correlation between the two voiceprint instance data and obtain the comparison correlation coefficient. Through the comparison process, the server obtains the comparison correlation coefficient between the second target voiceprint instance data and the first target voiceprint instance data. The server determines whether the comparison correlation coefficient is greater than a preset comparison correlation coefficient threshold. The comparison correlation coefficient threshold is a preset threshold used to determine whether the similarity between the two voiceprint instance data is high enough to construct a valid second sample instance. When the comparison correlation coefficient is greater than the comparison correlation coefficient threshold, the server constructs a second sample instance according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data. The second sample instance is a sample containing two voiceprint instance data and their associated information, which is used for subsequent model training or analysis. By constructing the second sample instance, the server obtains a sample instance containing the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data, which is used for further monitoring and analysis of the operation status of the stereo garage. Through such a processing flow, the server can compare the voiceprint target values ​​of the second target voiceprint instance data and the first target voiceprint instance data, and obtain a comparison correlation coefficient. When the comparison correlation coefficient is greater than the threshold, the server constructs a second sample instance based on the two voiceprint instance data for subsequent model training or analysis. This helps to improve the accuracy and robustness of the stereo garage operation status monitoring system.

[0123] In the embodiments of the present invention, the following implementation modes are also provided.

[0124] Acquire multiple original normalized components of the expert model, wherein the multiple original normalized components correspond one-to-one to the target voiceprint instance data and the associated running voiceprint data;

[0125] The multiple original normalized components are tuned according to the cost parameter to obtain multiple normalized components, wherein the multiple normalized components correspond to different model parameters.

[0126] In an embodiment of the present invention, exemplarily, the server first identifies an expert model for monitoring the operation status of a stereo garage. This model monitors the operation status of an elevator based on voiceprint analysis. The server obtains multiple original normalized components from the expert model. These components are used to normalize the target voiceprint instance data and the associated operation voiceprint data to the same numerical range or format for subsequent processing. The original normalized components correspond one-to-one to the target voiceprint instance data and the associated operation voiceprint data to ensure that each data has proper normalization processing. The server obtains cost parameters from the historical data of the stereo garage operation status monitoring system. These parameters reflect the performance and error of the model during the monitoring process. The server performs tuning processing on multiple original normalized components according to the cost parameters. The purpose of the tuning processing is to adjust the parameters of the original normalized components to reduce the model error and improve the monitoring performance. The tuning processing involves algorithms such as iterative update and gradient descent to gradually optimize the parameters of the normalized components. After the tuning processing, the server obtains multiple normalized components. These components correspond to different model parameters and have been optimized according to the cost parameters. The normalization component can now more effectively normalize the target voiceprint instance data and the associated operation voiceprint data to the appropriate value range or format to support the subsequent monitoring and analysis of the operation status of the stereo garage. Through such a processing flow, the server can obtain multiple original normalized components of the expert model and tune them according to the cost parameters. The tuned normalized components have more optimized model parameters, which can improve the accuracy and performance of the stereo garage operation status monitoring system. This will help the server to more accurately identify and analyze the operation status of the elevator, thereby ensuring the safe and efficient operation of the stereo garage.

[0127] In an embodiment of the present invention, the steps of performing multiple rounds of voiceprint feature extraction on the main operating voiceprint data according to the expert model to obtain a first voiceprint feature data set, and performing multiple rounds of voiceprint feature extraction on the at least one associated operating voiceprint data according to the expert model to obtain a second voiceprint feature data set can be implemented through the following examples.

[0128] Extracting voiceprint features from the main operation voiceprint data according to the expert model and processing them through a corresponding normalization component to obtain the first voiceprint feature data set;

[0129] The voiceprint feature of the at least one associated running voiceprint data is extracted according to the expert model and processed through a corresponding normalization component to obtain the second voiceprint feature data set.

[0130] In an embodiment of the present invention, exemplarily, the server identifies the main operation voiceprint data from the operation status monitoring system of the stereo garage. These data are the sound data generated by the elevator during normal operation. The server uses an expert model to extract voiceprint features from the main operation voiceprint data. This process involves converting the sound data into voiceprint features that can represent its characteristics. The server processes the extracted voiceprint features through a corresponding normalization component. The purpose of the normalization component is to normalize the voiceprint features to the same numerical range or format for subsequent processing. After normalization, the server obtains a first voiceprint feature data set. This data set contains the voiceprint features of the main operation voiceprint data and has been normalized. The server identifies at least one associated operation voiceprint data from the operation status monitoring system of the stereo garage. These data are other sound data related to the operation status of the elevator, such as garage environmental noise, other equipment sounds, etc. The server uses an expert model to extract voiceprint features from the associated operation voiceprint data. Similarly, this process involves converting the sound data into voiceprint features that can represent its characteristics. The server processes the extracted voiceprint features through a corresponding normalization component. The normalization component ensures that the voiceprint features of the associated operation voiceprint data are also within the same numerical range or format as the voiceprint features of the main operation voiceprint data. After the normalization process, the server obtains a second voiceprint feature data set. This data set contains the voiceprint features of the associated operation voiceprint data and has been normalized. Through such a processing flow, the server can perform multiple rounds of voiceprint feature extraction on the main operation voiceprint data and the associated operation voiceprint data according to the expert model, and process them through the corresponding normalization components to obtain the first voiceprint feature data set and the second voiceprint feature data set. These two data sets contain the voiceprint features of the main operation voiceprint data and the associated operation voiceprint data, respectively, and have been normalized, providing a basis for subsequent monitoring and analysis of the operation status of the stereo parking garage.

[0131] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the above-mentioned method for monitoring the operation status of a stereoscopic parking garage based on voiceprint. Figure 2 As shown, Figure 2 The block diagram of the computer device 100 provided in the embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112 and a communication unit 113. To achieve data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected to each other. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0132] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best apply the present disclosure and various embodiments with different modifications to the specific application intended.

Claims

1. A method for monitoring the operation status of a stereo garage based on voiceprint, characterized in that: include: Acquire main operation voiceprint data and at least one associated operation voiceprint data of the to-be-monitored garage equipment included in the stereo garage; The main operation voiceprint data is the sound data generated by the monitored garage equipment during normal operation, which is used to reflect the operation status and performance of the equipment, and the associated operation voiceprint data is other sound data associated with the main operation voiceprint data; A first voiceprint feature data set is obtained by performing multiple rounds of voiceprint feature extraction on the main operation voiceprint data according to the expert model, and a second voiceprint feature data set is obtained by performing multiple rounds of voiceprint feature extraction on the at least one associated operation voiceprint data according to the expert model, wherein the expert model is a model that has been trained in advance, the first voiceprint feature data set includes voiceprint features of multiple different granularity levels, the second voiceprint feature data set includes voiceprint features of multiple different granularity levels, and the number of rounds of voiceprint feature extraction is consistent with the number of granularity levels; the granularity level refers to different levels or degrees of detail of features; Performing feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain a first comprehensive voiceprint feature data set of the garage equipment to be monitored; Performing feature reconstruction processing based on the first voiceprint feature dataset and the second voiceprint feature dataset according to the lightweight model to obtain a second comprehensive voiceprint feature dataset; Performing feature comparison processing according to the first comprehensive voiceprint feature data set and the second comprehensive voiceprint feature data set to obtain an equipment operation status evaluation result of the to-be-monitored garage equipment; The operating status of the three-dimensional parking garage is determined according to the equipment operating status evaluation result.

2. The method according to claim 1, characterized in that Performing feature integration processing on the first voiceprint feature data set and the second voiceprint feature data set to obtain a first comprehensive voiceprint feature data set of the garage equipment to be monitored includes: Determining a mutual influence coefficient according to a first voiceprint feature and a second voiceprint feature, wherein the first voiceprint feature and the second voiceprint feature correspond to a granularity level of the same level, the first voiceprint feature is derived from the first voiceprint feature data set, and the second voiceprint feature is derived from the second voiceprint feature data set; Multiplying the second voiceprint feature by the mutual influence coefficient to obtain a multiplication result; Accumulating the multiplication result and the first voiceprint feature to obtain a first comprehensive voiceprint feature, wherein the first comprehensive voiceprint feature is classified into the first comprehensive voiceprint feature data set; The first voiceprint feature data set and the second voiceprint feature data set are polled to obtain a first comprehensive voiceprint feature data set.

3. The method according to claim 1, characterized in that Performing feature reconstruction processing based on the first voiceprint feature data set and the second voiceprint feature data set according to the lightweight model to obtain a second comprehensive voiceprint feature data set includes: Pre-adjusting the third voiceprint feature and the fourth voiceprint feature respectively to obtain a first pre-adjusted voiceprint feature and a second pre-adjusted voiceprint feature, wherein the third voiceprint feature and the fourth voiceprint feature have corresponding granularity levels of the same level, the third voiceprint feature is a voiceprint feature obtained by the expert model performing a subsequent round of voiceprint feature extraction on the main operation voiceprint data, and the fourth voiceprint feature is a voiceprint feature obtained by the expert model performing a subsequent round of voiceprint feature extraction on the at least one associated operation voiceprint data; the subsequent round of voiceprint feature extraction is a next round of voiceprint feature extraction after the last round of voiceprint feature extraction of the multiple rounds of voiceprint feature extraction; Performing feature integration processing on the first pre-adjusted voiceprint feature and the second pre-adjusted voiceprint feature to obtain a first pre-adjusted comprehensive voiceprint feature; Performing feature reconstruction processing based on the first pre-adjusted integrated voiceprint feature to obtain a second integrated voiceprint feature, wherein the second integrated voiceprint feature is derived from the second integrated voiceprint feature data set, and the granularity level of the second integrated voiceprint feature is consistent with the granularity level of the integrated voiceprint feature output by the expert model for multiple rounds; Pre-adjusting the fifth voiceprint feature and the sixth voiceprint feature respectively to obtain a third pre-adjusted voiceprint feature and a fourth pre-adjusted voiceprint feature, wherein the fifth voiceprint feature and the sixth voiceprint feature have corresponding granularity levels of the same level, the fifth voiceprint feature is a voiceprint feature obtained by the expert model performing multiple rounds of voiceprint feature extraction on the main running voiceprint data, and the sixth voiceprint feature is a voiceprint feature obtained by the expert model performing multiple rounds of voiceprint feature extraction on the at least one associated running voiceprint data; Performing feature integration processing on the third pre-adjusted voiceprint feature and the fourth pre-adjusted voiceprint feature to obtain a second pre-adjusted comprehensive voiceprint feature; Performing feature reconstruction processing based on the second pre-adjusted comprehensive voiceprint feature to obtain a third comprehensive voiceprint feature, the third comprehensive voiceprint feature originates from the second comprehensive voiceprint feature data set, and the granularity level of the third comprehensive voiceprint feature is consistent with the granularity level of the comprehensive voiceprint feature output by the preceding voiceprint feature extraction of the expert model; the preceding voiceprint feature extraction is a previous round of voiceprint feature extraction of the last round of voiceprint feature extraction of the multiple rounds of voiceprint feature extraction; The steps of pre-adjusting the third voiceprint feature and the fourth voiceprint feature respectively to obtain the first pre-adjusted voiceprint feature and the second pre-adjusted voiceprint feature are cyclically executed, and the step of performing feature reconstruction processing based on the second pre-adjusted comprehensive voiceprint feature to obtain the third comprehensive voiceprint feature is cyclically executed.

4. The method according to any one of claims 1 to 3, characterized in that The method further comprises: Acquire a sample instance data set and an original lightweight model, wherein the sample instance data set includes target voiceprint instance data and associated running voiceprint data; Performing multiple rounds of voiceprint feature extraction on the target voiceprint instance data based on the expert model to obtain multiple voiceprint features of the target voiceprint instance data; Performing multiple rounds of voiceprint feature extraction on the associated operation voiceprint data based on the expert model to obtain multiple voiceprint features of the associated operation voiceprint data; Performing feature integration processing on the multiple voiceprint features of the target voiceprint instance data and the multiple voiceprint features of the associated running voiceprint data to obtain multiple training comprehensive voiceprint features; Inputting a plurality of voiceprint features of the target voiceprint instance data and a plurality of voiceprint features of the associated running voiceprint data into the original lightweight model to perform feature reconstruction processing to obtain a plurality of reconstructed voiceprint features; Perform feature comparison processing according to the multiple reconstructed voiceprint features and the multiple trained comprehensive voiceprint features to obtain a cost parameter; A training process is performed on the original lightweight model according to the cost parameter to obtain the lightweight model.

5. The method according to claim 4, characterized in that The method further comprises: Obtain a first sample instance; wherein the first sample instance includes first target voiceprint instance data and a corresponding voiceprint target value; Performing voiceprint conversion on the first target voiceprint instance data through a second voiceprint conversion component to obtain voiceprint features of the first target voiceprint instance data; Extracting main voiceprint features from the voiceprint features of the first target voiceprint instance data through a deep learning model; Performing semantic construction processing according to the main voiceprint features through a semantic recognition model to obtain voiceprint semantic information of the first target voiceprint instance data; Integrate the voiceprint target value of the first target voiceprint instance data with the voiceprint semantic information to obtain integrated description information; Acquire a preset voiceprint feature of first interfering voiceprint data; Performing interference adding processing on the voiceprint features of the first target voiceprint instance data to obtain the voiceprint features of the second interference voiceprint data; Performing a de-interference operation on the voiceprint features of the second interfering voiceprint data according to a de-interference filtering strategy using a de-interference model to obtain the voiceprint features of the de-interference voiceprint data; Determine a cost parameter according to the interference information added corresponding to the interference adding process and the interference information inferred by the interference removal operation, and train a interference removal model according to the cost parameter; Performing a de-interference operation on the voiceprint features of the first interfering voiceprint data according to the de-interference filtering strategy through the trained de-interference model to obtain the voiceprint features of the second target voiceprint instance data; wherein the de-interference filtering strategy includes the semantic description features of the integrated description information; Derived the second target voiceprint instance data according to the voiceprint features of the second target voiceprint instance data; Compare the second target voiceprint instance data with the voiceprint target value of the first target voiceprint instance data to obtain a comparison correlation coefficient When the comparison correlation coefficient is greater than the comparison correlation coefficient threshold, constructing a second sample instance according to the second target voiceprint instance data and the voiceprint target value of the first target voiceprint instance data; The sample instance data set is constructed according to the first sample instance and the second sample instance.

6. The method according to claim 5, characterized in that The interference removal model includes a first feature combination unit; the interference removal model performs interference removal operation on the voiceprint features of the second interference voiceprint data according to the interference removal filtering strategy to obtain the voiceprint features of the interference removal voiceprint data, including: Performing a feature combination operation on the semantic description features of the integrated description information and the voiceprint features of the second interfering voiceprint data through a first feature combination unit in the interference removal model, so as to infer and filter the interference information in the voiceprint features of the second interfering voiceprint data, and obtain the voiceprint features of the interference-removed voiceprint data; The step of training the interference removal model according to the cost parameter comprises: Training a first feature combination unit in a de-interference model according to the cost parameter; The de-interference filtering strategy also includes the voiceprint features of the first target voiceprint instance data; the de-interference model includes a first feature combination unit and a second feature combination unit; the de-interference model performs a de-interference operation on the voiceprint features of the second interference voiceprint data according to the de-interference filtering strategy to obtain the voiceprint features of the de-interference voiceprint data, and also includes: Performing a feature combination operation on the semantic description feature of the integrated description information and the voiceprint feature of the second interfering voiceprint data through a first feature combination unit in the interference removal model, so as to infer and filter the interference information in the voiceprint feature of the second interfering voiceprint data, and obtain a first sub-voiceprint feature of the interference-removed voiceprint data; The feature mapping unit performs a feature mapping operation on the voiceprint features of the first target voiceprint instance data; wherein the voiceprint features after feature mapping of the first target voiceprint instance data are used to perform a feature combination operation Performing a feature combination operation on the voiceprint features of the first target voiceprint instance data and the voiceprint features of the second interference voiceprint data through a second feature combination unit in the interference removal model, so as to infer and filter interference information in the voiceprint features of the second interference voiceprint data, and obtain a second sub-voiceprint feature of the interference-removed voiceprint data; Performing feature fusion processing on the first sub-voiceprint feature of the de-interferenced voiceprint data and the second sub-voiceprint feature to obtain the voiceprint feature of the de-interferenced voiceprint data; The step of training the interference removal model according to the cost parameter comprises: The second feature combination unit in the interference removal model is trained according to the cost parameter.

7. The method according to claim 5, characterized in that The step of performing interference adding processing on the voiceprint features of the first target voiceprint instance data to obtain the voiceprint features of the second interference voiceprint data includes: Execute a multi-cycle interference adding loop and perform the following process in the target cycle interference adding loop process: Add interference information to the voiceprint features of the target period adding interference cycle input, and obtain the voiceprint features of the subsequent period adding interference cycle input; the subsequent period adding interference cycle is the period adding interference cycle after the last period adding interference cycle in the multi-period adding interference cycle; The voiceprint features of the first target voiceprint instance data are used as the voiceprint features of the first periodic interference cycle input; the voiceprint features obtained by adding interference cycles to the target periodicity are the voiceprint features of the second interference voiceprint data; The step of performing a de-interference operation on the voiceprint features of the second interfering voiceprint data according to the de-interference filtering strategy using the de-interference model to obtain the voiceprint features of the de-interference voiceprint data includes: Execute a multi-cycle interference removal loop and perform the following processing during the target cycle interference removal loop: Inferring interference information in the voiceprint features of the target period de-interference cycle input according to the de-interference filtering strategy through the de-interference model, and filtering the inferred interference information in the voiceprint features of the target period de-interference cycle input to obtain the voiceprint features of the subsequent period de-interference cycle input; Among them, the voiceprint features of the second interference voiceprint data are used as the voiceprint features input into the first periodic interference removal loop; the voiceprint features obtained in the target periodic interference removal loop are the voiceprint features of the interference-removed voiceprint data.

8. The method according to claim 5, characterized in that The method further comprises: Performing voiceprint conversion on the first target voiceprint instance data by a first voiceprint conversion component to obtain voiceprint features of the first target voiceprint instance data; Performing semantic conversion processing on the integrated description information through a semantic conversion component to obtain semantic description features of the integrated description information; The deducing the second target voiceprint instance data according to the voiceprint features of the second target voiceprint instance data includes: The voiceprint inverse conversion component corresponding to the first voiceprint conversion component performs voiceprint inverse conversion processing on the voiceprint features of the second target voiceprint instance data to obtain the second target voiceprint instance data.

9. The method according to claim 6, characterized in that The step of extracting multiple rounds of voiceprint features from the main operation voiceprint data according to the expert model to obtain a first voiceprint feature data set, and extracting multiple rounds of voiceprint features from the at least one associated operation voiceprint data according to the expert model to obtain a second voiceprint feature data set includes: Extracting voiceprint features from the main operation voiceprint data according to the expert model and processing them through a corresponding normalization component to obtain the first voiceprint feature data set; The voiceprint feature of the at least one associated running voiceprint data is extracted according to the expert model and processed through a corresponding normalization component to obtain the second voiceprint feature data set.

10. A server system, characterized in that: The method comprises a server, wherein the server is used to execute the method described in any one of claims 1 to 9.

Images