Automatic intelligent duty response identification processing system and method

Abstract

The invention discloses an automatic intelligent on-duty response recognition processing system and method, relates to the technical field of intelligent question answering, and solves the problems that in the prior art, existing machine question answering processing usually answers user questions according to a universal processing template or a universal neural network model; and bad user experience is easily generated in the process. Based on the response of the dynamic response strategy adjustment module, the response strategy is dynamically adjusted according to the historical interaction of the user, the current dialogue content, the system self-learning, the open API and the third-party channel, and the personalized response strategy is generated. That is to say, by providing personalized response content, the user experience is improved, the working process is more efficient, rapid processing is carried out on the edge, the working efficiency is improved, by dynamically adjusting the response strategy, better service can be provided, by means of continuous self-learning and optimization, the system can be continuously improved, and the user experience is improved. And more accurate response content is provided, so that the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent question-answering technology, and in particular to an automated intelligent duty answer recognition processing system and method. Background Art

[0002] With the continuous development of machine learning, the application of automated voice response systems in various industries has become more and more extensive. From traditional call centers to smart homes, from voice assistants to automatic customer service, automated voice response systems have become an indispensable tool for people in many aspects. However, most of the existing automated voice response systems still have some defects and potential risks. For example, in the public service field, such as power companies and water companies, the voice recognition accuracy of automated response systems is not high, which may cause misunderstandings or misinterpretations, resulting in a decline in customer service quality. In addition, these systems may not be able to correctly identify customers' urgent needs in certain emergency situations, resulting in delayed processing or more serious consequences.

[0003] Moreover, as users' personalized needs are increasing, users require that the responses to machine questions and answers must not only be quick but also meet their personal preferences. However, current machine question and answer processing usually only answers user questions according to general processing templates or general neural network models. In this way, the answers users get will only be applicable to general situations or have no personal bias, which can easily lead to a bad user experience.

[0004] In view of this, an automated intelligent duty response recognition and processing system and method is needed. Summary of the invention

[0005] In view of the fact that the current machine question-answering processing in the prior art usually only answers user questions according to a general processing template or a general neural network model, which easily leads to a poor user experience in the process, the present invention provides an automated intelligent on-duty answer recognition processing system and method, which can continuously improve the system through continuous self-learning and optimization, provide more accurate answer content, and thus improve the user experience. The specific technical solution is as follows:

[0006] An automated intelligent duty response recognition and processing system, comprising:

[0007] A voice receiving module, used to receive a voice signal input by a user;

[0008] The edge computing module is equipped with a trained deep learning model to recognize voice signals and generate text responses. After receiving the voice signal, the edge computing module first compares the voice signal with the historical data. If the similarity between the recognition result and the historical data is greater than the set threshold, the response content in the historical data is directly called; if the similarity between the recognition result and the historical data is less than the set threshold, the deep learning model is called for processing.

[0009] A dynamic response strategy adjustment module is connected to the edge computing module, and dynamically adjusts the neural network model based on historical interactions, thereby dynamically adjusting the text response strategy. The dynamic adjustment of the neural network model includes collecting user feedback. If the user provides positive feedback, the weight of similar results in the deep learning model is increased. If the user provides negative feedback, the weight of similar results is reduced.

[0010] The response generation and feedback processing module translates the text response content output by the dynamic response strategy adjustment module into the language used by the user and uses the corresponding tone, and then sends it to the user terminal.

[0011] Preferably, after receiving the voice signal input by the user, the voice receiving module also performs noise reduction processing, and the noise processing method includes a noise reduction encoder, subtracting the original audio code from the reference audio code through the noise reduction encoder to generate a noise reduction signal, and the training steps of the noise reduction encoder are:

[0012] A noise reduction encoder model is obtained by establishing a signal noise reduction model based on audio code for training;

[0013] The training process of the noise reduction encoder is as follows: the audio code is modeled using a mixed Gaussian model based on the output signal of the encoder to obtain a mixed Gaussian model;

[0014] The improved EM algorithm is used to estimate the parameters of the Gaussian mixture model;

[0015] Subtracting the reference audio code from the original audio code to obtain a noise reduction signal;

[0016] The encoding process of the noise reduction encoder is modeled by using a hidden Markov model, the number of states is determined, and a hidden Markov model of the encoding process of the noise reduction encoder is obtained;

[0017] The improved EM algorithm is used to estimate the model parameters of the hidden Markov model of the encoding process of the denoising encoder;

[0018] After obtaining the model parameters and probability distribution of the denoising encoder and the mixed Gaussian model, a final denoising encoder model is generated;

[0019] After the training is completed, the input original audio code is processed with the denoising encoder model to obtain the denoised signal.

[0020] Preferably, the deep learning model training process is as follows:

[0021] Collect and organize speech data sets, and perform necessary preprocessing and annotation work;

[0022] Perform data augmentation, including adding noise, changing volume, speed, and dividing the dataset into training, validation, and test sets;

[0023] Extract effective acoustic features from the preprocessed speech signal, including MFCC and FBank;

[0024] Select a deep learning framework and build a speech recognition model;

[0025] Use the feature vectors and the corresponding text labels to train the speech recognition model. Adjust the model parameters through the back propagation algorithm to optimize the model performance;

[0026] Use the test data set to evaluate the model performance and adjust and optimize the model based on the evaluation results.

[0027] Preferably, the dynamic response strategy adjustment module performs the following operations:

[0028] After the model is deployed, collect user feedback on the recognition results and distinguish between positive and negative feedback;

[0029] If the user provides positive feedback, adjust the loss function or increase the weight of similar samples during training to achieve this; if the user provides negative feedback, adjust the loss function or reduce the frequency of similar samples in the training set;

[0030] Retrain or fine-tune the model using new training data based on the adjusted weights;

[0031] Use user feedback as a continuous feedback loop to continuously optimize and adjust the model to improve model accuracy and user satisfaction.

[0032] Preferably, the response generation and feedback processing module collects interaction data between the user and the system, maps features into latent space by analyzing historical interaction information of the user, obtains a representation of a feature vector, compares the feature vector with the current conversation content, obtains similarity, and dynamically adjusts the response strategy to obtain personalized response content.

[0033] Preferably, a self-learning module is also included, which is used to continuously collect user interaction data and system feedback data, and train and optimize the deep learning model based on the continuously collected user interaction data and system feedback data.

[0034] Preferably, it includes a user terminal, an edge computing unit and a central control unit, the edge computing unit is connected to the user terminal, the central control unit is connected to the edge computing unit, the voice receiving module is mounted on the user terminal, the edge computing module, the dynamic response strategy adjustment module and the response generation and feedback processing module are mounted on the edge computing unit, and the self-learning module is mounted on the central control unit.

[0035] An automated intelligent on-duty response recognition and processing method, applied to the system as described above, comprises the following steps:

[0036] Based on the historical data of interactions between users and the system, the system can understand the user's behavioral habits and habitual language patterns, generate personalized response strategies to better understand the user's intentions, and give corresponding feedback.

[0037] The current conversation content will be fed back to the system in real time, enabling the system to adjust strategies in real time and respond to user needs more quickly.

[0038] The response strategy can be combined with the system's self-learning mechanism to enable the automatic response system to have learning and adaptability, and can automatically correct the strategy based on user feedback to improve accuracy and efficiency.

[0039] The system can use open APIs to share data and exchange information with third-party channels. By integrating multiple data sources, it can better understand and judge user intentions and further optimize response strategies.

[0040] Open APIs make third-party channel data one of the data sources for adjusting response strategies. This data can include data from social networks, search engines, etc. By integrating data from third-party channels, the system can understand user needs more accurately, adjust strategies more quickly, and improve work efficiency.

[0041] A computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the automated intelligent duty response recognition and processing system as described above.

[0042] A processor is used to run a program, wherein the program, when running, executes the automated intelligent duty response recognition and processing system as described above.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] The present invention improves the user experience by providing personalized response content, makes the workflow more efficient, performs rapid processing at the edge, improves work efficiency, and can provide better services by dynamically adjusting the response strategy. Through continuous self-learning and optimization, the system can be continuously improved and provide more accurate response content, thereby improving the user experience. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0047] It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0048] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0049] In one embodiment of the present invention, an automated intelligent on-duty response recognition and processing system is provided, comprising the following modules:

[0050] The voice receiving module is used to receive the voice signal input by the user and perform preliminary preprocessing on the voice signal; the edge computing module recognizes the voice signal after preliminary processing and generates a text response; the dynamic response strategy adjustment module dynamically generates personalized response content based on historical interactions, current dialogue information, system self-learning, open APIs and third-party channels; the response generation and feedback processing module generates personalized voice responses based on the personalized response content of the dynamic response strategy adjustment module and sends them to the user; the self-learning module collects user interaction data and system feedback data, performs data analysis and optimizes and adjusts the response strategy.

[0051] 1. Voice receiving module: users input voice through smart devices, and the front-end module receives the voice signal; converts the voice signal into audio format, performs noise reduction on the audio signal, and sends the noise-reduced voice signal to the edge computing module, which processes the voice signal in real time. During the voice input and preliminary processing, there is no need for system participation, and it is only transmitted between the edge device and the front-end module, with fast transmission speed and high efficiency.

[0052] 1.1 The voice receiving module performs noise reduction processing on the audio signal. The noise reduction processing method includes a noise reduction encoder. The original audio code is subtracted from the reference audio code through the noise reduction encoder to generate a noise reduction signal. The training steps of the noise reduction encoder are:

[0053] A noise reduction encoder model is obtained by establishing a signal noise reduction model based on audio code for training;

[0054] The training process of the noise reduction encoder is as follows: the audio code is modeled using a mixed Gaussian model based on the output signal of the encoder to obtain a mixed Gaussian model;

[0055] The improved EM algorithm is used to estimate the parameters of the Gaussian mixture model;

[0056] Subtracting the reference audio code from the original audio code to obtain a noise reduction signal;

[0057] The encoding process of the noise reduction encoder is modeled by using a hidden Markov model, the number of states is determined, and a hidden Markov model of the encoding process of the noise reduction encoder is obtained;

[0058] The improved EM algorithm is used to estimate the model parameters of the hidden Markov model of the encoding process of the denoising encoder;

[0059] After obtaining the model parameters and probability distribution of the denoising encoder and the mixed Gaussian model, a final denoising encoder model is generated;

[0060] After the training is completed, the input original audio code is processed with the denoising encoder model to obtain the denoised signal.

[0061] 2. The edge computing module performs speech recognition on the initially processed voice signal, transmits the voice signal to the edge device in 16khz, 16-bit linear PCM format for noise reduction; uses a deep learning model to perform speech recognition on the signal, selects the corresponding model according to system requirements, and generates a text response; in edge computing, the use of deep learning algorithms can process signals faster.

[0062] 2.1 The edge computing module compares the voice signal with the historical data. If the similarity between the recognition result and the historical data is greater than the set threshold, the response content in the historical data is directly called; if the similarity between the recognition result and the historical data is less than the set threshold, the deep learning model is called for processing.

[0063] 3. Dynamic response strategy adjustment module, dynamically adjusts the response strategy and generates personalized response content based on the user's historical interactions, current conversation content, system self-learning, and open API; dynamically adjusts by recording the user's historical input, records each user input to generate the corresponding feature vector, compares the historical data with the current conversation content, records the differences, selects the appropriate response strategy based on the corresponding data set, and generates personalized response content.

[0064] 3.1 The dynamic response strategy adjustment module includes a dynamic adjustment module and a personalized response generation module. According to the historical interactions and the current conversation content, the dynamic adjustment module dynamically adjusts the response strategy according to the context information of the user interaction and the current conversation content, and generates personalized response content; the personalized response generation module uses natural language generation technology to generate personalized voice response content based on the response content generated by the dynamic adjustment module; the generated voice response content is transmitted to the edge device, and the device's vocoder is used to convert the text response content into a voice signal and send it to the terminal.

[0065] 3.1.1 The dynamic response strategy adjustment module adjusts the response strategy according to the historical feedback information provided by the user in the previous interaction; if the user provides positive feedback, the weight of the similar results in the deep learning model is increased; if the user provides negative feedback, the weight of the similar results is reduced; the response strategy is adjusted according to the historical feedback information provided by the user to improve the accuracy of the response.

[0066] 3.1.2 The dynamic response strategy adjustment module also obtains external data through open APIs and third-party channels, uses deep learning models to pre-process the external data, integrates the external data into the response strategy, and dynamically adjusts the response strategy.

[0067] 3.2 Dynamic adjustment module, generates user personalized information based on historical interaction information, uses text retrieval algorithm to retrieve historical interaction information to generate user personalized information; based on user personalized information, uses neural network architecture to generate personalized response content, uses user personalized information and user historical interaction information as input of neural network architecture, and generates personalized response content through neural network training.

[0068] 3.3 Personalized response generation module, using a neural network architecture, models the user's input and output according to the response content output by the dynamic adjustment module, inputs the user's input and output samples into the neural network architecture, obtains the model results through training, automatically outputs the response content according to the model results, and uses a text retrieval algorithm to select personalized response content based on the correlation between the user's input and historical interaction information.

[0069] The training process of the deep learning model mentioned above includes the following key steps:

[0070] Data preparation: First, collect and organize speech data sets, and perform necessary preprocessing and annotation. The quality and quantity of data are crucial to the model training effect; second, data enhancement, such as adding noise, changing volume, speed, etc., and dividing the data set into training set, validation set, and test set.

[0071] Feature extraction: Extract effective acoustic features from the preprocessed speech signal, such as MFCC, FBank, etc.

[0072] Model construction: Select a deep learning framework (such as TensorFlow, PyTorch, etc.) to build a speech recognition model. Depending on the specific task requirements, you can choose different types of model structures such as DNN, CNN, RNN, etc.

[0073] Model training: Use feature vectors and corresponding text labels to train the speech recognition model. Adjust model parameters through the back propagation algorithm to optimize model performance.

[0074] Model evaluation: Use the test dataset to evaluate the model performance and adjust and optimize the model based on the evaluation results.

[0075] In addition, the process of optimizing the model based on user feedback is as follows:

[0076] The process of adjusting the model weights based on user feedback can be done as follows:

[0077] Collect user feedback: After the model is deployed, collect user feedback on the recognition results, distinguishing between positive and negative feedback.

[0078] Adjust model weights: If the user provides positive feedback, the model can be strengthened in similar situations by increasing the weight of similar results in the model. This can be done by adjusting the loss function or increasing the weight of similar examples during training. If the user provides negative feedback, the model can be optimized by reducing the weight of similar results in the model to avoid repeating the same mistakes in future predictions. This can be done by adjusting the loss function or reducing the frequency of similar examples in the training set.

[0079] Retrain the model: Retrain or fine-tune the model using new training data based on the adjusted weights.

[0080] Model evaluation: After adjusting the weights and retraining the model, you need to reevaluate the model's performance on the test set to ensure that the model's improvements are as expected.

[0081] Continuous iteration: Use user feedback as a continuous feedback loop to continuously optimize and adjust the model to improve model accuracy and user satisfaction.

[0082] This process involves continuous learning and adaptation of the deep learning model. It is a dynamic and iterative process that requires constant updating of the model based on new data and feedback.

[0083] 4. The answer generation and feedback processing module performs reply processing based on the personalized answer content obtained by the dynamic answer strategy adjustment module, obtains a personalized voice answer, and sends it to the user; the answer generation and feedback processing module translates the personalized answer content into the language and tone used by the user, generates personalized voice answer content, and then sends the answer content to the terminal.

[0084] 4.1 The response generation and feedback processing module collects the interaction data between users and the system, maps the features into the latent space by analyzing the user's historical interaction information, obtains the representation of the feature vector, compares the feature vector with the current conversation content, obtains the similarity, and dynamically adjusts the response strategy to obtain personalized response content.

[0085] 5. The self-learning module continuously collects user interaction data and system feedback data, performs data analysis and optimizes the response strategy, performs self-learning and self-adjustment based on user feedback, and generates a more accurate response strategy.

[0086] 5.1 Based on the operation results of the self-learning module and the dynamic response strategy, adjustments are made to generate a deep learning model. After passing the system verification, the deep learning model is used to recognize and judge voice signals and process user voice responses. The trained model is used to recognize and process voice input, and the trained deep learning model is used to dynamically adjust the response strategy to obtain better response results.

[0087] In one embodiment of the present invention, the system includes a user terminal, an edge computing unit and a central control unit, the edge computing unit is connected to the user terminal, the central control unit is connected to the edge computing unit, and each unit module in the previous embodiment is respectively mounted in the user terminal, the edge computing unit and the central control unit. Among them, the voice receiving module is mounted in the user terminal, the edge computing module, the dynamic response strategy adjustment module and the response generation and feedback processing module are mounted in the edge computing unit, and the self-learning module is mounted in the central control unit.

[0088] The deep learning model installed in the edge computing unit is regularly updated from the central control unit. In other words, the edge computing unit is also equipped with a deep learning model, but the optimization of the deep learning model is performed using some historical interaction data of the current user, which has a small amount of calculation. The deep learning model in the central control unit is optimized using further expanded data such as practical third-party data and historical interaction data of all users. Therefore, the update of the deep learning model in the central control unit reflects the overall trend of voice recognition and answer optimization, which has a large amount of calculation, but it has its own computing space, so it does not occupy the space of the edge computing unit. The two cooperate with each other to further optimize the deep learning model, and each has its own computing space, so that the speed of information processing is guaranteed to a certain extent.

[0089] The reason for setting up the edge computing unit and the central control unit is to reduce the computing pressure of the edge computing unit so that it can return the response to the user terminal faster.

[0090] In one embodiment of the present invention, an automated intelligent on-duty answer recognition processing method is provided. Based on the dynamic answer strategy adjustment module responding to the user's voice input and the recognition result of the edge computing device, the system dynamically adjusts the answer strategy according to the user's historical interaction, current conversation content, system self-learning, open API and third-party channels to generate personalized answer strategies. These strategies may include:

[0091] 1.1 User’s historical interaction data. Based on the historical interaction data between the user and the system, the system can understand the user’s behavioral habits and habitual language patterns, generate personalized response strategies to better understand the user’s intentions, and give corresponding feedback.

[0092] 1.2 Current conversation content: The current conversation content will be fed back to the system in real time, enabling the system to adjust its strategy in real time and thus respond to user needs more quickly.

[0093] 1.3 System self-learning, the response strategy can be combined with the system's self-learning mechanism, so that the automatic response system itself has learning and adaptability, and can automatically correct the strategy based on user feedback to improve accuracy and efficiency.

[0094] 1.4 Open API. The system can use open API to share data and interact with third-party channels. By integrating multiple data sources, it can better understand and judge user intentions and further optimize response strategies.

[0095] 1.5 Third-party channels: Open APIs make third-party channel data one of the data sources for adjusting response strategies. This data can include data from social networks, search engines, and other channels. By integrating data from third-party channels, the system can understand user needs more accurately, adjust strategies more quickly, and improve work efficiency.

[0096] In summary, the present invention improves the user experience by providing personalized response content, makes the workflow more efficient, performs rapid processing at the edge, improves work efficiency, and can provide better services by dynamically adjusting the response strategy. Through continuous self-learning and optimization, the system can continuously improve and provide more accurate response content, thereby improving the user experience.

[0097] Those of ordinary skill in the art will appreciate that the units of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0098] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0099] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0100] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nlyMemory), random access memory (RAM, RandomAccessMemory), mobile hard disk, magnetic disk or optical disk, etc., which can store program code.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and specification of the present invention.

Claims

1. An automated intelligent duty response recognition and processing system, characterized in that: include: A voice receiving module, used to receive a voice signal input by a user; The edge computing module is equipped with a trained deep learning model to recognize voice signals and generate text responses. After receiving the voice signal, the edge computing module first compares the voice signal with the historical data. If the similarity between the recognition result and the historical data is greater than the set threshold, the response content in the historical data is directly called; if the similarity between the recognition result and the historical data is less than the set threshold, the deep learning model is called for processing. A dynamic response strategy adjustment module is connected to the edge computing module, and dynamically adjusts the neural network model based on historical interactions, thereby dynamically adjusting the text response strategy. The dynamic adjustment of the neural network model includes collecting user feedback. If the user provides positive feedback, the weight of similar results in the deep learning model is increased. If the user provides negative feedback, the weight of similar results is reduced. The response generation and feedback processing module translates the text response content output by the dynamic response strategy adjustment module into the language used by the user and uses the corresponding tone, and then sends it to the user terminal.

2. The automated intelligent duty response recognition and processing system according to claim 1 is characterized in that: After receiving the voice signal input by the user, the voice receiving module also performs noise reduction processing. The noise processing method includes a noise reduction encoder, and the original audio code is subtracted from the reference audio code through the noise reduction encoder to generate a noise reduction signal. The training steps of the noise reduction encoder are: A noise reduction encoder model is obtained by establishing a signal noise reduction model based on audio code for training; The training process of the noise reduction encoder is as follows: the audio code is modeled using a mixed Gaussian model based on the output signal of the encoder to obtain a mixed Gaussian model; The improved EM algorithm is used to estimate the parameters of the Gaussian mixture model; Subtracting the reference audio code from the original audio code to obtain a noise reduction signal; The encoding process of the noise reduction encoder is modeled by using a hidden Markov model, the number of states is determined, and a hidden Markov model of the encoding process of the noise reduction encoder is obtained; The improved EM algorithm is used to estimate the model parameters of the hidden Markov model of the encoding process of the denoising encoder; After obtaining the model parameters and probability distribution of the denoising encoder and the mixed Gaussian model, a final denoising encoder model is generated; After the training is completed, the input original audio code is processed with the denoising encoder model to obtain the denoised signal.

3. The automated intelligent duty response recognition and processing system according to claim 1, characterized in that: The deep learning model training process is as follows: Collect and organize speech data sets, and perform necessary preprocessing and annotation work; Perform data augmentation, including adding noise, changing volume, speed, and dividing the dataset into training, validation, and test sets; Extract effective acoustic features from the preprocessed speech signal, including MFCC and FBank; Select a deep learning framework and build a speech recognition model; Train the speech recognition model using feature vectors and corresponding text labels. Adjust model parameters through back-propagation algorithm to optimize model performance; Use the test data set to evaluate the model performance and adjust and optimize the model based on the evaluation results.

4. The automated intelligent duty response recognition and processing system according to claim 2, characterized in that: The dynamic response strategy adjustment module performs the following operations: After the model is deployed, collect user feedback on the recognition results and distinguish between positive and negative feedback; If the user provides positive feedback, adjust the loss function or increase the weight of similar samples during training to achieve this; if the user provides negative feedback, adjust the loss function or reduce the frequency of similar samples in the training set; Retrain or fine-tune the model using new training data based on the adjusted weights; Use user feedback as a continuous feedback loop to continuously optimize and adjust the model to improve model accuracy and user satisfaction.

5. The automated intelligent duty response recognition and processing system according to claim 1 is characterized in that: The response generation and feedback processing module collects the interaction data between the user and the system, maps the features into the latent space by analyzing the historical interaction information of the user, obtains the representation of the feature vector, compares the feature vector with the current conversation content, obtains the similarity, and dynamically adjusts the response strategy to obtain personalized response content.

6. The automated intelligent duty response recognition and processing system according to claim 1, characterized in that: It also includes a self-learning module for continuously collecting user interaction data and system feedback data, and training and optimizing the deep learning model based on the continuously collected user interaction data and system feedback data.

7. The automated intelligent duty response recognition and processing system according to claim 6, characterized in that: It includes a user terminal, an edge computing unit and a central control unit, the edge computing unit is connected to the user terminal, the central control unit is connected to the edge computing unit, the voice receiving module is mounted on the user terminal, the edge computing module, the dynamic response strategy adjustment module and the response generation and feedback processing module are mounted on the edge computing unit, and the self-learning module is mounted on the central control unit.

8. An automated intelligent duty response recognition and processing method, characterized in that: The system applied to any one of claims 1 to 7 comprises the following steps: Based on the historical data of interactions between users and the system, the system can understand the user's behavioral habits and habitual language patterns, generate personalized response strategies to better understand the user's intentions, and give corresponding feedback. The current conversation content will be fed back to the system in real time, enabling the system to adjust strategies in real time and respond to user needs more quickly. The response strategy can be combined with the system's self-learning mechanism to enable the automatic response system to have learning and adaptability, and can automatically correct the strategy based on user feedback to improve accuracy and efficiency. The system can use open APIs to share data and exchange information with third-party channels. By integrating multiple data sources, it can better understand and judge user intentions and further optimize response strategies. Open APIs make third-party channel data one of the data sources for adjusting response strategies. This data can include data from social networks, search engines, etc. By integrating data from third-party channels, the system can understand user needs more accurately, adjust strategies more quickly, and improve work efficiency.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the automated intelligent duty response recognition and processing system according to any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the automated intelligent duty response recognition and processing system described in any one of claims 1 to 7.