A method and system for printer voice interaction implementation

Abstract

The application provides a printer voice interaction implementation method and system, and belongs to the technical field of printing voice interaction. The method comprises the following steps: traversing various OID nodes of the printer to form the content of the OID column of the general intent mapping table; forming the special function program column of the special intent mapping table according to the special functions that need to be completed by the printer; marking the operation intent and supplementary information of the general intent mapping table and the special intent mapping table; constructing a training sample and training an intent recognition model by using the training sample; deploying the intent recognition model to the printer; converting the collected voice information into text data, obtaining the operation intent and / or supplementary parameters corresponding to the text data by using the intent recognition model; determining the OID operation item or the special program by searching the general intent mapping table and the special intent mapping table, executing the OID operation or executing the special program, and obtaining the result text of the OID operation or the special program execution and voice broadcasting.

Description

Technical Field

[0001] This invention proposes a method and system for implementing voice interaction in printing devices, belonging to the field of printing voice interaction technology. Background Technology

[0002] Voice interaction has become a new way to manage and control printing equipment. By sending voice commands, users can query printer properties (e.g., "Does the printer support duplex printing?") and printing status (e.g., "How many more pages can the printer print with the remaining toner?"), and also configure the printing equipment (e.g., "Set the printer's network address to 192.168.6.6"). After analyzing the voice to obtain the intended command (e.g., setting the network address) and supplementary information (e.g., setting it to "192.168.6.6"), the user's intent needs to be executed to complete the specific operation on the printer. Since there are numerous commands for querying and configuring printers (approximately dozens to hundreds), developing code to implement each intent would not only be a huge development workload but also unable to handle new user command requirements. Summary of the Invention

[0003] This invention provides a method and system for implementing voice interaction in a printing device, to solve the technical problems existing in the prior art. The technical solution adopted is as follows:

[0004] A method for implementing voice interaction in a printer, the method comprising:

[0005] Iterate through all types of OID nodes corresponding to the printer, select the target OID node, construct a general intent mapping table for the target OID node, and form the content of the OID column of the general intent mapping table;

[0006] The operation intent and supplementary information corresponding to each OID node in the general intent mapping table are labeled;

[0007] As needed, dedicated functions that can be performed by voice control of the printer are mapped into a dedicated intent map of dedicated function programs, and their corresponding operation intents and supplementary information are labeled.

[0008] Based on the operation intentions and supplementary information listed in the general intent mapping table and the special intent mapping table, training samples for the intent recognition model are constructed, and the intent recognition model is trained and tested using the training samples. The intent recognition model corresponding to the completed training and testing is then deployed to the printer.

[0009] The system collects voice information sent by the user to the printer in real time, converts the voice information into text data, and obtains the corresponding operation intention and supplementary parameters of the text data by reasoning through the intent recognition model.

[0010] The OID operation item or special procedure is determined by looking up the general intent mapping table and the special intent mapping table, the OID operation or special procedure is executed, the result text of the OID operation or special procedure is obtained, and the result text is converted into voice broadcast and feedback to the user.

[0011] Furthermore, by traversing the various OID nodes corresponding to the printer, a general intent mapping table is formed, corresponding to the contents of the OID column of the general intent mapping table, including:

[0012] The OID traversal module performs traversal operations on various OID nodes, sequentially traversing printer manufacturer-specific nodes, printer-related general nodes, and device-related general nodes; and selecting OID nodes suitable for printer voice control as target OID nodes.

[0013] By collecting the OID information of the target OID node, a general intent mapping table is formed; wherein, the general intent mapping table contains OID items and corresponding OID information; wherein, the OID information is used to match the user's operation intent, and the OID information corresponds to the label of the operation intent of the user command during model training.

[0014] Furthermore, the operation intent and supplementary information corresponding to each OID node in the general intent mapping table are labeled, including:

[0015] Iterate through each row of the general intent map table to obtain each OID item contained in the general intent map table;

[0016] For each OID entry in the general intent mapping table, determine the corresponding operation intent and supplementary information for the general intent mapping table;

[0017] The operation intent and supplementary information corresponding to each OID item in the general intent mapping table are labeled.

[0018] Furthermore, a dedicated function program sequence corresponding to the dedicated functions performed by voice-controlled printers as needed is formed into a dedicated intent mapping table, and its corresponding operation intent and supplementary information are labeled, including:

[0019] Based on the current printer's attributes and the various queries and configuration operations that can be performed, determine the dedicated functions that need to be completed through voice control of the printer;

[0020] A dedicated intent mapping table is formed for dedicated functions that can be completed by voice control of the printer as needed. The dedicated intent mapping table contains a dedicated function program list corresponding to the dedicated functions.

[0021] For each dedicated function in the dedicated intent mapping table, determine the corresponding operation intent and supplementary information for that dedicated intent mapping table;

[0022] The operation intents and supplementary information corresponding to the dedicated functions contained in the dedicated intent mapping table are labeled.

[0023] Further, training samples for the intent recognition model are constructed, the intent recognition model is trained using the training samples, and the completed training and testing intent recognition model is deployed to the printer, including:

[0024] Construct training samples, which include user command text and operation intent labels and / or supplementary parameter labels corresponding to the user command text. The operation intent labels and / or supplementary parameter labels are matched one-to-one with the operation intents and / or supplementary parameters in the general intent mapping table and the special intent mapping table.

[0025] The intent recognition model is trained based on training samples;

[0026] Construct test samples, test the intent recognition model based on the test samples, and calculate the label inference probability of each operation intent label and / or supplementary parameter label. When all label inference probabilities are less than 60%, it is determined that the text intent is not highly relevant to all existing labels, and the message "Sorry, I can only understand print settings related instructions" is output.

[0027] The intent recognition model is trained and tested multiple times based on training and test samples until the intent accuracy and supplementary parameter accuracy reach the preset qualified standards. Then the training of the intent recognition model is considered complete.

[0028] Deploy the completed intent recognition model to the printer.

[0029] Furthermore, the system collects voice information sent by the user to the printer in real time and converts the voice information into text data, including:

[0030] The user sends a voice command to the printer. The voice acquisition and command text recognition module captures the voice information and performs noise processing on the voice information to obtain the voice information after noise processing.

[0031] The noise-processed speech information is converted into text data.

[0032] Furthermore, the voice acquisition and command text recognition module captures voice information and performs voice noise processing, including:

[0033] Automatically initiate noise sampling according to the noise sampling window, collect current ambient noise, and extract baseline features:

[0034] Environmental noise is classified based on baseline characteristics to determine noise types, including steady-state noise, transient noise, and mixed noise.

[0035] For steady-state noise, a high-frequency retention filtering strategy is adopted to deeply suppress the noise peak frequency band of steady-state noise; wherein, the deep suppression refers to the suppression of the noise peak frequency band by an attenuation of 7dB-10dB;

[0036] For transient noise, only the transient noise portion of the speech information whose noise duration is less than a preset specific duration t is moderately suppressed. The moderate suppression refers to the suppression of the noise peak frequency band by an attenuation of 4dB-6dB. The preset specific duration t ranges from 180ms to 220ms.

[0037] For mixed noise, noise suppression is performed using the noise energy parameter E corresponding to the time period of the mixed noise.

[0038] Furthermore, for mixed noise, noise suppression is performed using the noise energy parameter E corresponding to the time period of the mixed noise, including:

[0039] Extract each frame of the time period corresponding to the mixed noise, and obtain the noise energy parameter E using the frame energy and energy change rate of each frame contained in the time period corresponding to the mixed noise;

[0040] When the noise energy parameter exceeds the preset energy parameter threshold, the noise peak frequency band corresponding to the mixed noise is suppressed by an attenuation of 8dB-12dB.

[0041] When the noise energy parameter does not exceed the preset energy parameter threshold, the noise peak frequency band corresponding to the mixed noise is suppressed by an attenuation of 5dB-7dB.

[0042] Furthermore, the operation intent and supplementary parameters corresponding to the text data are obtained by reasoning through the intent recognition model, including:

[0043] The text data is input into the instruction text intent recognition module; wherein, the instruction text intent recognition module is equipped with an intent recognition model.

[0044] The intent recognition model deployed in the instruction text intent recognition module is used to reason about the text data and extract the operation intent and supplementary parameters from the text data.

[0045] Furthermore, by searching the general intent mapping table and the dedicated intent mapping table, the system performs voice information correspondence queries and operation intent determination, and provides voice broadcast and feedback on the query results and operation intent, including:

[0046] The instruction execution scheduling module independently queries the general intent mapping table and the special intent mapping table respectively;

[0047] If an OID item corresponding to the operation intent is found in the general intent mapping table, the control OID management module executes the corresponding operation through the get / set standard command of the SNMP protocol according to the OID item matched in the general intent mapping table, and passes the query and setting operation results to the speech synthesis module; the speech synthesis module synthesizes speech broadcast content based on the matching results and broadcasts it.

[0048] If no OID item corresponding to the operation intent is found in the general intent mapping table, the instruction execution scheduling module returns "Unable to execute your instruction" to the speech synthesis module, and the speech synthesis module then reads the content aloud.

[0049] If a matching item corresponding to the operation intent is found in the dedicated intent mapping table, the instruction execution scheduling module calls the program execution content corresponding to the matching item in the dedicated execution module set to complete the query and setting operation results, and passes the query and setting operation results to the speech synthesis module; the speech synthesis module synthesizes speech broadcast content based on the matching results and broadcasts it.

[0050] If no matching item corresponding to the operation intent is found in the dedicated intent mapping table, the instruction execution scheduling module returns "Unable to execute your instruction" to the speech synthesis module, which then reads the content aloud.

[0051] A system for implementing voice interaction for a printer, the system comprising:

[0052] The general intent mapping table forming module is used to traverse the various OID nodes corresponding to the printer, select the target OID node, construct a general intent mapping table for the target OID node, and form the content of the OID column of the general intent mapping table;

[0053] The standard module of the general intent mapping table is used to annotate the operation intent and supplementary information corresponding to each OID node in the general intent mapping table;

[0054] The dedicated intent mapping table generation module is used to generate a dedicated function program list for dedicated intent mapping tables based on the dedicated functions that can be performed by voice-controlled printers as needed, and to annotate the corresponding operation intents and supplementary information.

[0055] The dedicated intent mapping table standard module is used to construct training samples for the intent recognition model based on the operation intents and supplementary information listed in the general intent mapping table and the dedicated intent mapping table, and to use the training samples to train and test the intent recognition model, and then deploy the intent recognition model corresponding to the completed training and testing to the printer.

[0056] The voice acquisition and voice semantic data processing module is used to acquire voice information sent by the user to the printer in real time, convert the voice information into text data, and obtain the operation intention and / or supplementary parameters corresponding to the text data by reasoning through the intent recognition model.

[0057] The query and voice feedback control module is used to query the corresponding voice information and determine the operation intent by looking up the general intent mapping table and the special intent mapping table, and to broadcast and provide feedback on the query results and operation intent via voice.

[0058] Beneficial effects of this invention:

[0059] This invention proposes a method and system for implementing voice interaction in printing devices. By extensively querying and setting operation intentions for printers and implementing them through standard SNMP protocol commands, the amount of programming required to implement printer operation commands is reduced. On one hand, by annotating the operation intentions and supplementary information associated with OIDs in the general intent mapping table, a precise mapping mechanism from voice commands and operation intentions to device OIDs is established. On the other hand, by combining text data reasoning to obtain operation intentions and supplementary parameters, the core needs of user voice commands can be effectively identified, reducing misunderstandings caused by differences or ambiguities in command expression. Simultaneously, through a dual query mechanism combining priority querying of the general table with matching of the dedicated table, it is ensured that user commands can quickly locate the corresponding execution path, i.e., OID operation or dedicated program call, effectively reducing the probability of command execution failure and thus improving the reliability of voice interaction. Attached Figure Description

[0060] Figure 1 The process of the method described in this invention Figure 1 ;

[0061] Figure 2 The process of the method described in this invention Figure 2 ;

[0062] Figure 3 This is a schematic diagram of the system principle of the system described in this invention;

[0063] Figure 4 This is a schematic diagram illustrating the principle and structure of the intent recognition model described in this invention. Detailed Implementation

[0064] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0065] This invention proposes a method for implementing voice interaction in a printer, such as... Figure 1 and Figure 2 As shown, the method includes:

[0066] Traverse all types of OID nodes corresponding to printer 100, select target OID node, construct a general intent mapping table 111 for the target OID node, and form the content of the OID column of the general intent mapping table 111.

[0067] The operation intent and supplementary information corresponding to each OID node in the general intent mapping table 111 are labeled;

[0068] As needed, the dedicated functions that can be performed by voice control of the printer are used to form a dedicated function program list in dedicated intent mapping table 112, and the corresponding operation intent and supplementary information are marked.

[0069] Based on the operation intentions and supplementary information listed in the general intent mapping table 111 and the special intent mapping table 112, training samples for the intent recognition model are constructed, and the intent recognition model is trained using the training samples. The intent recognition model corresponding to the completed training and testing is then deployed to the printer 100.

[0070] The system collects voice information sent by the user to the printer 100 in real time, converts the voice information into text data, and obtains the operation intent and supplementary parameters corresponding to the text data by reasoning through the intent recognition model.

[0071] The system performs voice information matching and operation intent determination by searching the general intent mapping table 111 and the special intent mapping table 112, and then broadcasts and provides feedback on the query results and operation intents via voice.

[0072] The working principle of the above technical solution is as follows: First, an intent recognition model is deployed on the printer to realize the voice interaction function. The intent recognition model is trained specifically using two types of samples: a general intent mapping table and a dedicated intent mapping table. After training and testing, the corresponding intent recognition model is deployed to the printer 100, enabling the printer to recognize the user's printing commands or printing intents. The general intent mapping table is deployed on the printer's target OID node to enable the querying and setting of general printer functions. The general intent mapping table contains operation intents and supplementary information. These general printing functions correspond to the functions of the printer's existing OID nodes. For additional printer-specific attributes and / or dedicated operations, dedicated programs are developed to implement them. The operation intents and supplementary information in the dedicated intent mapping table 112 are implemented by calling the corresponding dedicated applets. Since the number of printer-specific attributes and dedicated operations is usually limited, the number of dedicated programs developed is controllable.

[0073] The core of the general program is to first obtain the printer's supported attribute items and configuration items. A dedicated tool iterates through the printer's supported OID items, selects the target OID node, and builds or sets a general OID item list on the target OID node. This list includes the OID item and its corresponding description, used to correspond to specific operational instructions. The target OID node is selected based on commonly used function nodes among all existing printer OID nodes, such as 'check paper tray' or 'check ink level'. Functions such as 'check printer frequency', which are generally not of interest to users or frequently used, are not considered as target OID nodes.

[0074] The general OID list requires labeling each OID with its intended operation and supplementary parameters, forming a general intent mapping table. When a user sends a voice command like "Tell me the printer's device model," the intent recognition model first infers the user's intended operation as "query information," with the supplementary parameter being "device model." Then, it retrieves the corresponding OID from the general intent mapping table, calls the OID management program, and uses the SNMP protocol's get / set standard commands to obtain or set the OID's value. Finally, the returned result is returned to the user via speech synthesis. If no operable OID is found in the list, the corresponding result is read aloud. Furthermore, while searching the general intent mapping table, a custom-defined dedicated intent mapping table is also used to retrieve and call the corresponding proprietary program module to obtain the relevant query information or execute specific configuration operations.

[0075] The above technical solution achieves the following results: In the process of implementing printing commands via printer voice interaction, by pre-deploying printer functionality—that is, by extensively querying and setting operation intentions for the printer, and implementing them through standard SNMP protocol commands—the amount of programming required to implement printer operation commands is reduced. On one hand, by labeling the operation intentions and supplementary information associated with OIDs in the general intent mapping table, a precise mapping mechanism from voice commands and operation intentions to device OIDs is established. On the other hand, by combining text data reasoning to obtain operation intentions and supplementary parameters, the core needs of user voice commands can be effectively identified, reducing misunderstandings caused by differences or ambiguities in command expression. Simultaneously, through a dual query mechanism combining priority querying of the general table with matching of the dedicated table, it is ensured that user commands can quickly locate the corresponding execution path, i.e., OID operation or dedicated program call, effectively reducing the probability of command execution failure and thus improving the reliability of voice interaction.

[0076] In one embodiment of the present invention, the various OID nodes corresponding to the printer 100 are traversed to form a general intent mapping table 111 corresponding to the contents of the OID column of the general intent mapping table, including:

[0077] The OID traversal module 201 performs traversal operations on various OID nodes, sequentially traversing printer manufacturer-specific nodes, printer-related general nodes, and device-related general nodes; and selecting OID nodes suitable for printer voice control as target OID nodes.

[0078] By collecting OID information of OID nodes suitable for printer voice control, a general intent mapping table 111 is formed, as shown in Table 1. This table contains OID entries and corresponding OID information. The OID information is used to match the user's operational intent, and it corresponds to the annotation of the user's operational intent used during model training. As shown in the table below, each OID entry corresponds to a different operational intent and supplementary parameters, uniquely pointing to a specific operational intent instruction. For example, when the user voice outputs "query printer model," the specific OID entry will be determined based on the operational intent and / or supplementary parameters in the general intent mapping table 111, and the instruction will be executed after the instruction is determined.

[0079] Table 1

[0080]

[0081] The working principle of the above technical solution is as follows: First, the OID traversal module performs a targeted OID traversal operation, sequentially traversing the printer manufacturer's private nodes, printer-related general nodes, and device-related general nodes, systematically collecting all OID information supported by the printer; then, a general intent mapping table is constructed based on the collected OID information. The general intent mapping table contains OID items and corresponding OID information, wherein the OID information is specifically used to match user operation intents and forms a corresponding relationship with the intent labels of user instructions during model training, thus building a mapping bridge between OID information and user operation intents.

[0082] The technical effects of the above solution are as follows: It performs targeted traversal of three types of core OID nodes, covering both printer manufacturer-specific private nodes and general nodes related to printing and devices, ensuring comprehensive and complete collection of OID information and laying a solid data foundation for the general intent mapping table. Simultaneously, the OID information in the general intent mapping table accurately corresponds to user intent tags, establishing a direct association between OID items and user operation intents, effectively converting OID information into user intents, and significantly improving the accuracy of subsequent intent matching. Furthermore, this embodiment considers both manufacturer-specific and general node traversal logic, enabling the general intent mapping table to adapt to the OID systems of different printer brands, effectively improving the universality and cross-device adaptability of the method steps described in this embodiment.

[0083] In one embodiment of the present invention, the operation intent and supplementary information corresponding to each row of OID nodes in the general intent mapping table 111 are labeled, including:

[0084] Iterate through each row of the general intent map table to obtain each OID item contained in the general intent map table;

[0085] For each OID entry in the general intent mapping table, determine the corresponding operation intent and supplementary information for the general intent mapping table 111;

[0086] The operation intent and supplementary information corresponding to each OID item in the general intent mapping table are labeled.

[0087] The working principle of the above technical solution is as follows: First, traverse each row of data in the general intent mapping table and extract the single OID contained in each row; then, based on the functional attributes corresponding to the OID, the OID information and the preset intent association rules, determine the operation intent and supplementary information corresponding to the OID; finally, associate the determined operation intent and supplementary information with the corresponding OID to complete the labeling in the general intent mapping table, so that each row of OID in the table is bound to a clear intent and auxiliary information.

[0088] The technical effects of the above solution are as follows: In this embodiment, the above solution assigns clear operational intents and supplementary information to OIDs through annotation, thereby adding complete operational intents and supplementary information, as well as a complete association system between the general intent mapping table and OIDs, effectively enhancing the semantic expressive power of the mapping table. Simultaneously, the operational intents and user command intent tags accurately correspond, providing a direct basis for the rapid matching of subsequent user intents and OIDs, thus effectively improving the efficiency of intent recognition and command execution. Furthermore, it effectively avoids execution deviations caused by missing information, improving the operational stability and accuracy of subsequent intent determination.

[0089] In one embodiment of the present invention, a dedicated function program sequence corresponding to a dedicated function to be performed by voice-controlled printer is formed into a dedicated intent mapping table 112, and the corresponding operation intent and supplementary information are labeled, including:

[0090] Based on the current printer's attributes and the various queries and configuration operations that can be performed, determine the dedicated functions that need to be completed by voice control of printer 100;

[0091] A dedicated intent mapping table is formed based on the dedicated functions that can be performed by voice-controlled printers as needed. This dedicated intent mapping table contains a list of dedicated function programs corresponding to the dedicated functions, and the dedicated intent mapping table 112 is shown in Table 2.

[0092] Table 2

[0093]

[0094] For each dedicated function in the dedicated intent mapping table, determine the corresponding operation intent and supplementary information in the dedicated intent mapping table 112;

[0095] The operation intents and supplementary information corresponding to the dedicated functions contained in the dedicated intent mapping table are labeled.

[0096] The working principle of the above technical solution is as follows: First, based on the inherent attributes of the current printer and the various query and configuration operations it can perform, the above technical solution accurately defines the dedicated functions that the printer needs to complete. Then, based on the determined dedicated functions, a dedicated intent mapping table is constructed. For each dedicated function in the dedicated intent mapping table, its corresponding operation intent and supplementary information supporting the function execution are clarified. Finally, these operation intents and supplementary information are associated and labeled with the corresponding dedicated functions in the dedicated intent mapping table to form a complete association system. The implementation of the operation intents and corresponding functions in the dedicated intent mapping table needs to be done by embedding a mini-program (because these are not general functions of the printer, they cannot be queried through the OID column), forming a dedicated function program column. That is, the implementation of each dedicated function needs to be achieved through a corresponding dedicated function program.

[0097] The technical effects of the above solution are as follows: In this embodiment, dedicated functions are determined based on the printer's own attributes and executable operations, enabling the dedicated intent mapping table to accurately match the current printer's functional characteristics. This avoids redundancy in the general functions included in the dedicated intent mapping table, thereby improving the relevance and adaptability of the dedicated intent mapping table to the actual functions of the printer. Simultaneously, by annotating the operation intent and supplementary information, a direct association is established between the dedicated functions and user needs, providing necessary auxiliary data for function execution, thereby reducing the error rate in function comparison and further improving the execution efficiency of dedicated functions. Furthermore, by establishing a system of association between dedicated functions and their corresponding operation intents and supplementary information, the dedicated intent mapping table possesses clear semantic direction and execution basis, effectively improving the response speed of dedicated function determination in various scenarios, thereby enhancing the accuracy and efficiency of printer dedicated function invocation.

[0098] In one embodiment of the present invention, voice information sent by a user to a printer 100 is collected in real time, and the voice information is converted into text data, including:

[0099] The user sends a voice command to the printer 100, such as "Tell me the printer's device model". The voice acquisition and command text recognition module 101 captures the voice information and performs voice information noise processing to obtain the voice information after noise processing.

[0100] The noise-processed speech information is converted into text data, for example, "I want to check the printer model," and the text data is then subjected to confidence verification. Specifically:

[0101] Retrieve the text data corresponding to the converted speech information;

[0102] Scan the text data and extract the keywords contained within it, such as "query" or "printer model".

[0103] Retrieve the voice information processing history of the printing device from the cloud platform corresponding to the printing device, and retrieve the frequency of keyword occurrence and the accuracy of voice command execution corresponding to the keyword from the voice information processing history.

[0104] When there are multiple keywords, the correlation coefficient between each pair of keywords in the multiple related words is determined based on the speech information processing history. Where C represents the correlation coefficient between any two keywords; P represents the percentage of occurrence of any two keywords in the speech information processing history; S represents the speech execution success rate corresponding to the occurrence of any two keywords in the speech information processing history; and L represents the percentage of occurrence of any speech command swapped when any two keywords in the speech information processing history are associated.

[0105] The confidence score of the text data is obtained by combining the frequency of occurrence of the keywords, the accuracy of voice command execution corresponding to the keywords, and the correlation coefficient between each pair of keywords. Where Q represents the confidence score of the text data; n represents the total number of pairs of keywords in the text data; C i f represents the correlation coefficient corresponding to the i-th keyword group; 01i and f 02i f represents the frequency of the two keywords contained in the i-th keyword group in the speech information processing history; ref Indicates the preset frequency reference value; S 01i and S 02i These represent the voice command execution accuracy rates of the two keywords contained in the i-th group of keywords in the voice information processing history.

[0106] The confidence score of the text data is compared with a preset confidence threshold.

[0107] When the confidence level of the text data is not lower than the preset confidence threshold, the text data is determined to have passed verification, and subsequent steps are executed.

[0108] When the confidence level of the text data is lower than the preset confidence threshold, the system will issue a voice broadcast to the user to read the text data content and confirm the instruction with the user again.

[0109] The working principle of the above technical solution is as follows: the voice acquisition and command text recognition module 101 captures the voice commands sent by the user to the printer in real time, first performs noise processing on the acquired raw voice information to remove or weaken interference signals and obtain voice information with higher purity; then, based on speech-to-text technology, the noise-processed voice information is converted into structured text data, completing the conversion of voice signal to text information.

[0110] On the other hand, after the text data is determined, its feasibility is verified. Specifically, the converted text data is scanned to identify and retrieve printing-related keywords, such as "query" and "printer model." The historical data of voice information processing is synchronously retrieved from the cloud platform corresponding to the printer, extracting key parameters, namely, keyword frequency, accuracy of voice command execution corresponding to the keywords, the percentage of each pair of keywords appearing together in multi-keyword scenarios, the success rate of voice execution when they appear together, and the percentage of voice command swapping when they appear together. For the multiple keywords extracted from the text, based on the retrieved historical data, the correlation coefficient between each pair of keywords is calculated using three dimensions: correlation ratio, execution success rate, and swapping ratio, quantifying the rationality and relevance of the keyword combination. By integrating the keyword combination correlation coefficient with the frequency of each keyword, the preset frequency reference value, and the accuracy of each voice command execution, the overall confidence value of the text data is obtained through integrated processing, thereby comprehensively evaluating the matching degree and executability of the text data with the user's true intent.

[0111] The effects of the above technical solution are as follows: Noise processing weakens interference signals, improves the purity of speech information, provides a high-quality signal foundation for subsequent speech-to-text conversion, effectively reduces the impact of noise on speech recognition, improves the accuracy of speech information converted into text data, and thus reduces the probability of subsequent intent reasoning deviations due to recognition errors. Simultaneously, it converts unstructured speech signals into structured text data, providing a parsable data carrier for the subsequent instruction text intent recognition module 102 to infer operational intent and supplementary parameters. Furthermore, by using a multi-dimensional confidence acquisition method combining keyword attributes with keyword combination attributes, it comprehensively considers the rationality and executability of text semantics, significantly reducing the risk of misjudgment caused by isolated keyword recognition and semantic ambiguity, ensuring that verified text data corresponds to effective device operations, reducing the impact of invalid commands on printer operating efficiency, and thus improving the reliability and stability of voice interaction. In addition, compared to general speech-to-text solutions, the above semantic validity verification method is more adaptable to the usage requirements of printing equipment, maximizing the adaptability between recognition accuracy and printing scenarios.

[0112] In one embodiment of the present invention, the voice acquisition and command text recognition module 101 captures voice information and performs voice information noise processing, including:

[0113] Automatically initiate noise sampling according to the noise sampling window, collect current ambient noise, and extract baseline features:

[0114] Environmental noise is classified based on baseline characteristics to determine noise types, including steady-state noise, transient noise, and mixed noise.

[0115] For steady-state noise, a high-frequency retention filtering strategy is adopted to deeply suppress the noise peak frequency band of steady-state noise; wherein, the deep suppression refers to the suppression of the noise peak frequency band by an attenuation of 7dB-10dB;

[0116] For transient noise, only the transient noise portion of the speech information whose noise duration is less than a preset specific duration t is moderately suppressed. The moderate suppression refers to the suppression of the noise peak frequency band by an attenuation of 4dB-6dB. The preset specific duration t ranges from 180ms to 220ms.

[0117] For mixed noise, noise suppression is performed using the noise energy parameter E corresponding to the time period of the mixed noise.

[0118] The working principle of the above technical solution is as follows: Accurate processing of environmental noise when the equipment is not running is achieved through a hierarchical logic of noise sampling, feature extraction, type classification, and differentiated suppression. First, when the equipment is not running, a noise sampling window of 300ms-500ms is automatically activated to collect the current environmental noise and extract baseline features. Second, based on the baseline features, the environmental noise is classified into steady-state noise, transient noise, or mixed noise. Finally, a differentiated suppression strategy is implemented for different noise types: for steady-state noise, high-frequency retention filtering is used, with deep suppression of 7dB-10dB in its peak frequency band; for transient noise, only the portion lasting 180ms-220ms is moderately suppressed by 4dB-6dB; and for mixed noise, suppression is implemented based on the noise energy parameter E of its corresponding time period, achieving targeted processing of complex environmental noise.

[0119] The effects of the above technical solution are as follows: This embodiment achieves accurate differentiation of steady-state, transient, and mixed noise by extracting multi-dimensional baseline features in the time and frequency domains, providing a reliable basis for subsequent differential suppression and avoiding misjudgment of noise type caused by traditional single-feature classification. Simultaneously, this embodiment designs differential suppression strategies for different noise characteristics. Deep suppression of steady-state noise can effectively offset continuous interference, while high-frequency retention filtering reduces damage to high-frequency details of speech. Short-time limiting processing of transient noise avoids false suppression of effective short-time components in the speech signal, and moderate attenuation balances noise suppression and speech protection. Energy parameter-based processing of mixed noise takes into account both noise characteristics, solving the problem of poor performance of a single strategy under mixed interference. Furthermore, through refined processing strategies such as high-frequency retention and short-time limiting, this embodiment effectively suppresses noise while maximizing the preservation of the time-domain waveform and frequency-domain features of the speech command signal, effectively avoiding speech distortion caused by excessive noise reduction and providing high-quality signal input for subsequent command recognition. Meanwhile, this embodiment, through dynamic sampling windows and multi-feature classification, can handle complex noise scenarios in various office environments, such as mixed interference from air conditioning and conversation, as well as transient interference from sudden keyboard tapping, thereby maximizing the stability of voice interaction under different non-operating conditions.

[0120] One embodiment of the present invention, for mixed noise, utilizes the noise energy parameter E of the corresponding time period of the mixed noise for noise suppression, including:

[0121] Extract each frame corresponding to the mixed noise time period, and obtain the noise energy parameters using the frame energy and energy change rate of each frame contained in the mixed noise time period. Where E represents the noise energy parameter; m represents the total number of frames contained in the time period corresponding to the mixed noise; E i E represents the frame energy corresponding to the i-th frame; base Indicates the baseline energy of ambient noise when the printer is not running; ΔE i ΔE represents the rate of change of noise energy between the i-th frame and the previous frame. ref The reference energy change rate is taken as the average energy change rate during the pure steady-state noise period multiplied by 2, serving as a normalization benchmark for the degree of fluctuation to avoid numerical overflow; B represents the energy correction coefficient, and... , where ΔE max-ref ΔE represents the absolute difference between the rate of energy change corresponding to the frame with the maximum energy in the mixed noise and the reference rate of energy change; max-min This represents the absolute difference between the rate of energy change corresponding to the frame with the highest energy and the rate of energy change corresponding to the frame with the lowest energy in the mixed noise.

[0122] When the noise energy parameter exceeds the preset energy parameter threshold, the noise peak frequency band corresponding to the mixed noise is suppressed by an attenuation of 8dB-12dB.

[0123] When the noise energy parameter does not exceed the preset energy parameter threshold, the noise peak frequency band corresponding to the mixed noise is suppressed by an attenuation of 5dB-7dB.

[0124] The working principle of the above technical solution is as follows: the noise suppression strategy for mixed noise is to perform precise noise suppression by combining frame feature extraction, energy parameter quantization and dynamic attenuation control. First, each frame corresponding to the mixed noise time period is extracted to obtain the frame energy and energy change rate of each frame. Second, based on the total number of frames, the baseline energy of the ambient noise, the reference energy change rate, and the energy correction coefficient, the noise energy parameter E is calculated. Finally, based on the comparison result between the noise energy parameter E and the preset energy parameter threshold, the peak frequency band of the mixed noise is dynamically attenuated. That is, when the noise energy parameter E exceeds the threshold, a strong attenuation of 8dB-12dB is used, and when it does not exceed the threshold, a moderate attenuation of 5dB-7dB is used to achieve targeted suppression of mixed noise.

[0125] The above technical solution achieves the following effects: In this embodiment, the noise energy parameter E integrates frame energy and energy change rate. The frame energy reflects noise intensity, and the energy change rate reflects noise fluctuation characteristics. The energy correction coefficient B dynamically compensates for the difference in the proportion of steady-state and transient components in the mixed noise, effectively improving the accuracy of quantifying the degree of mixed noise interference and avoiding the problem of large deviations in evaluation results caused by relying solely on energy intensity in traditional technologies. Differential attenuation is applied based on the comparison result between the noise energy parameter E and the threshold. This effectively suppresses mixed noise with strong attenuation when the noise interference is strong, and balances noise suppression and speech protection with moderate attenuation when the interference is weak. This solves the contradiction between the need for strong suppression of strong interference and the need to protect speech when the interference is weak in mixed noise. Simultaneously, by fusing multi-frame features and dynamic correction, it can simultaneously adapt to the combined interference of steady-state components (such as continuous background noise) and transient components (such as sudden pulses) in mixed noise, avoiding the problem of insufficient suppression of certain components in mixed noise by a single strategy, and significantly improving the signal-to-noise ratio in mixed noise scenarios. Differential attenuation is applied only to the peak noise frequency band, and excessive attenuation is avoided by quantifying the energy parameters. This maximizes the preservation of signal characteristics in the core speech frequency band, reduces speech distortion during the noise reduction process, and provides high-quality input for subsequent speech command recognition.

[0126] In one embodiment of the present invention, the operation intent and supplementary parameters corresponding to the text data are obtained by reasoning through the intent recognition model, including:

[0127] The text data is input to the instruction text intent recognition module 102; wherein, the instruction text intent recognition module 102 is equipped with an intent recognition model.

[0128] The intent recognition model deployed in the instruction text intent recognition module is used to reason about the text data and extract the operation intent (such as "query information") and supplementary parameters (such as "device model") from the text data.

[0129] The working principle of the above technical solution is as follows: The instruction text intent recognition module 102 receives the converted text data, analyzes the core semantics of the text through the built-in semantic reasoning logic, separates and extracts the operation intent that represents the user's core needs from the text, and clarifies the supplementary parameters of the execution object of the intent, thus completing the conversion from unstructured text to structured intent and parameters.

[0130] The intent recognition model in this embodiment is built on a pruned BERT architecture, and the principle and structure of the intent recognition model are as follows: Figure 4 As shown:

[0131] The basic architecture of an intent recognition model consists of an input layer, an embedding layer, a Transformer layer, and an output layer, where:

[0132] Input layer: Used for inputting text data;

[0133] Embedding layer: The BertEmbeddings module converts user-input natural language sentences into vector representations.

[0134] The sentence is split into word pieces, and each word is mapped to a word vector using Word Embeddings;

[0135] By adding Position Embeddings (position vectors) and Token Type Embeddings (segment vectors), a fused vector containing word, position, and paragraph information is finally obtained.

[0136] Transformer layers (4 BertLayer layers): used to capture long-range dependencies and contextual information in data through a self-attention mechanism. This invention reduces the existing conventional 12-layer Transformer layer structure to 4 layers, compressing the parameter scale from 100 million to 60 million. Each layer includes BertAttention (including Query / Key / Value transformation and attention calculation), BertIntermediate (intermediate linear transformation and activation), and BertOutput (output linear transformation, Dropout, and LayerNorm), realizing feature extraction and transformation layer by layer. This reduction in the number of Transformer layers not only meets the capacity requirements of existing printer systems but also improves work efficiency.

[0137] Output layer: Adapts to the operation intent and supplementary parameters to output labels, supporting intent recognition and parameter extraction tasks.

[0138] For the intent recognition model, the training samples constructed using the operation intents and supplementary information listed in the general intent mapping table 111 and the special intent mapping table 112 are used to train the intent recognition model. The intent recognition model corresponding to the completed training and testing is then deployed to the printer 100. The specific training, testing, and deployment process includes:

[0139] Constructing training samples for the intent recognition model, training the intent recognition model using the training samples, and deploying the completed training and testing intent recognition model to printer 100, including:

[0140] Construct training samples, which include user command text and operation intent labels and / or supplementary parameter labels corresponding to the user command text. The operation intent labels and / or supplementary parameter labels are matched one-to-one with the operation intents and / or supplementary parameters in the general intent mapping table 111 and the special intent mapping table 112.

[0141] The intent recognition model is trained based on training samples;

[0142] Construct test samples, test the intent recognition model based on the test samples, and calculate the label inference probability of each operation intent label and / or supplementary parameter label. When all label inference probabilities are less than 60%, it is determined that the text intent is not highly relevant to all existing labels, and the message "Sorry, I can only understand print settings related instructions" is output.

[0143] The intent recognition model is trained and tested multiple times based on training and test samples until the intent accuracy and supplementary parameter accuracy reach the preset qualified standards. Then the training of the intent recognition model is considered complete.

[0144] Finally, the completed intent recognition model is deployed to printer 100.

[0145] The effects of the above technical solution are as follows: After training and testing, the intent recognition model can accurately extract the core semantics of text data, clarify the core needs and execution objects of user interaction, and provide a standardized matching basis for subsequent general / dedicated intent mapping table queries; reduce the understanding threshold of text semantics, simplify the logical complexity of subsequent instruction execution scheduling, and improve the efficiency of intent matching; ensure the consistency between operation intent and supplementary parameters, reduce deviations in subsequent function calls caused by semantic ambiguity, and improve the reliability of instruction execution; quickly complete the conversion of text into structured information, ensure efficient connection of the entire voice interaction process, and enhance the smoothness of user interaction.

[0146] On the other hand, for the Transformer layer of the intent recognition model, the original 12-layer BERT model, which occupied approximately 1GB of memory, was compressed to 4 layers through layer reduction. Furthermore, the number of parameters was reduced from 100 million to 60 million, making the memory consumption suitable for the lightweight, megabyte-level resource environments of embedded devices such as printers. This significantly improved the model's inference speed and its compatibility with printer devices, enabling real-time inference on the device side without relying on cloud computing power. This avoids issues such as network latency and bandwidth limitations, while meeting the low-resource, high-response instruction processing requirements of the printer's local end. In scenarios involving printing settings with specific instructions, the false positive rate of intent recognition is effectively reduced. It can accurately distinguish various operational intents in printing scenarios and effectively reduce the error rate of supplementary parameter extraction. This allows for accurate capture of printing parameter details contained in the instructions, effectively avoiding printing operation errors caused by intent deviations or parameter errors, and ensuring the accuracy and reliability of instruction parsing and execution. Meanwhile, by using the judgment logic of "probability of all labels < 60%", intelligent boundary recognition of non-printing setting commands is achieved, avoiding invalid parsing of irrelevant commands and reducing meaningless erroneous output. At the same time, a unified and clear feedback message, such as "Sorry, I can only understand commands related to printing settings", can quickly convey the interaction boundary to the user, greatly improving the intuitiveness and smoothness of human-computer interaction.

[0147] In one embodiment of the present invention, voice information is queried and operation intent is determined by looking up a general intent mapping table 111 and a dedicated intent mapping table 112, and the query results and operation intent are broadcast and fed back via voice, including:

[0148] The instruction execution scheduling module independently queries the general intent mapping table and the special intent mapping table respectively;

[0149] If an OID item corresponding to the operation intent is found in the general intent mapping table, the control OID management module executes the corresponding operation using the standard get / set command of the SNMP protocol (e.g., snmpget 1.3.6.1.2.1.1.1.0) based on the matched OID item in the general intent mapping table, and transmits the query and set operation results to the speech synthesis module; the speech synthesis module synthesizes speech content based on the matching results and performs speech playback.

[0150] If no OID item corresponding to the operation intent is found in the general intent mapping table, the instruction execution scheduling module returns "Unable to execute your instruction" to the speech synthesis module, and the speech synthesis module then reads the content aloud.

[0151] If a matching item corresponding to the operation intent is found in the dedicated intent mapping table, the instruction execution scheduling module calls the program execution content corresponding to the matching item in the dedicated execution module set to complete the query and setting operation results, and passes the query and setting operation results to the speech synthesis module; the speech synthesis module synthesizes speech broadcast content based on the matching results and broadcasts it.

[0152] If no matching item corresponding to the operation intent is found in the dedicated intent mapping table, the instruction execution scheduling module returns "Unable to execute your instruction" to the speech synthesis module, which then reads the content aloud.

[0153] The working principle of the above technical solution is as follows: The instruction execution scheduling module 103 follows independent query logic for general and specific purposes, searches the general intent mapping table 111, and if a matching OID is found, it triggers the OID management module 104 to execute the operation via the SNMP protocol; if no match is found, it returns feedback that the instruction cannot be executed. At the same time, it searches the specific intent mapping table 112. If a matching item is found, it executes the corresponding operation; if no match is found, it returns feedback that the instruction cannot be executed. All operation results or feedback information are transmitted to the speech synthesis module 106 and broadcast to the user in speech form, forming a complete closed loop from query to execution to feedback.

[0154] The effects of the above technical solution are as follows: A query strategy prioritizing general tables ensures rapid matching and execution of printer general functions, effectively improving command response efficiency; a dedicated intent mapping table 112 is used as a supplementary lookup to achieve full-scenario coverage of both general and proprietary functions, avoiding command execution failures due to differences in function types; simultaneously, the above technical solution clearly defines the calling logic of the SNMP standard commands and dedicated programs corresponding to the OID, effectively improving the accuracy and reliability of operation execution; and it provides corresponding feedback for different matching results, with both operation results and execution failure prompts clearly announced via voice, enhancing the certainty and user experience of user interaction.

[0155] This invention proposes a system for implementing voice interaction in a printer, the system comprising:

[0156] The general intent mapping table forming module is used to traverse the various OID nodes corresponding to the printer, select the target OID node, construct a general intent mapping table for the target OID node, and form the content of the OID column of the general intent mapping table;

[0157] The standard module of the general intent mapping table is used to annotate the operation intent and supplementary information corresponding to each OID node in the general intent mapping table;

[0158] The dedicated intent mapping table forming module is used to form a dedicated function program list of the dedicated intent mapping table 112 as needed for dedicated functions to be completed by voice control of the printer, and to mark the corresponding operation intent and supplementary information.

[0159] The dedicated intent mapping table standard module is used to construct training samples for the intent recognition model based on the operation intents and supplementary information listed in the general intent mapping table and the dedicated intent mapping table, and to use the training samples to train and test the intent recognition model, and then deploy the intent recognition model corresponding to the completed training and testing to the printer.

[0160] The voice acquisition and voice semantic data processing module is used to acquire voice information sent by the user to the printer in real time, convert the voice information into text data, and obtain the operation intention and / or supplementary parameters corresponding to the text data by reasoning through the intent recognition model.

[0161] The query and voice feedback control module is used to query the corresponding voice information and determine the operation intent by looking up the general intent mapping table and the special intent mapping table, and to broadcast and provide feedback on the query results and operation intent via voice.

[0162] The working principle of the above technical solution is as follows: Specifically, as shown in the figure below. Figure 3The device principle corresponding to the system in this embodiment shown includes a voice acquisition and command text recognition module 101, a command text intent recognition module 102, a command execution scheduling module 103, an OID management module 104, a dedicated execution module set 105, a voice synthesis module 106, a general intent mapping table 111, a dedicated intent mapping table 112, and an OID traversal module 201; wherein, the voice acquisition and command text recognition module 101 and the command text intent recognition module 102 establish a data connection relationship; the command text intent recognition module 102 and the command execution scheduling module 103 establish a data connection relationship; the command execution scheduling... Module 103 establishes data connection relationships with the OID management module 104, the dedicated execution module set 105, and the speech synthesis module 106 respectively; the OID traversal module 201 establishes a data connection relationship with the OID management module 104; at the same time, the instruction execution scheduling module 103 queries the general intent mapping table 111 and the dedicated intent mapping table 112, and the voice acquisition and instruction text recognition module 101, instruction text intent recognition module 102, instruction execution scheduling module 103, OID management module 104, dedicated execution module set 105, and speech synthesis module 106 are embedded in the corresponding control system of the printer 100.

[0163] First, a general-purpose program is developed to handle common printer queries and settings. Printer-specific attributes and operations are handled through dedicated programs. Since the number of printer-specific attributes and operations is typically limited, the number of dedicated programs developed can be controlled. The core of the general-purpose program is to first obtain the printer-supported attribute and configuration items. By using a dedicated tool to iterate through the printer-supported OID items, a general list of OID items can be obtained, including the OID item, its corresponding description, and other information.

[0164] Based on this, each OID is labeled with its operational intent and supplementary parameters, forming a general intent mapping table. When a user sends a voice command like "Tell me the printer's device model," the intent recognition model first infers the user's operational intent as "query information," with the supplementary parameter being "device model." Then, it retrieves the corresponding OID through the general intent mapping table, calls the OID management program, and uses the SNMP protocol's get / set standard commands to obtain or set the OID's value. Finally, the returned result is returned to the user via speech synthesis. If no operable OID is found in the list, the corresponding result is broadcast aloud. Furthermore, while searching the general intent mapping table, a custom-defined dedicated intent mapping table is also used to retrieve and call the dedicated program module corresponding to the operational intent to obtain the corresponding query information or execute a dedicated configuration operation.

[0165] The above technical solution achieves the following results: by querying and setting numerous printer operation intentions and implementing them through standard SNMP protocol commands, the amount of programming required to implement printer operation commands is reduced. On one hand, by annotating the operation intentions and supplementary information associated with OIDs in the general intent mapping table, a precise mapping mechanism from voice commands and operation intentions to device OIDs is established. On the other hand, by combining text data reasoning to obtain operation intentions and supplementary parameters, the core needs of user voice commands can be effectively identified, reducing misunderstandings caused by differences or ambiguities in command expression. Simultaneously, through a dual query mechanism combining priority querying of the general table with matching of the dedicated table, it is ensured that user commands can quickly locate the corresponding execution path, i.e., OID operation or dedicated program call, effectively reducing the probability of command execution failure and thus improving the reliability of voice interaction.

[0166] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for implementing voice interaction in a printer, characterized in that, The method includes: S1. Traverse the various OID nodes corresponding to the printer (100), select the target OID node, construct a general intent mapping table (111) for the target OID node, and form the content of the OID column of the general intent mapping table (111); S2. Label the operation intent and supplementary information corresponding to each OID node in the general intent mapping table (111); S3. As needed, special functions that can be completed by voice control of the printer are formed into a special intent mapping table (112) of special function program list, and the corresponding operation intent and supplementary information are marked. S4. Based on the operation intent and supplementary information listed in the general intent mapping table (111) and the special intent mapping table (112), construct training samples for the intent recognition model, train the intent recognition model using the training samples, and deploy the intent recognition model corresponding to the completed training and testing to the printer (100). S5. Real-time acquisition of voice information sent by the user to the printer (100), conversion of the voice information into text data, and obtaining the operation intent and / or supplementary parameters corresponding to the text data by reasoning through the intent recognition model; S6. By looking up the general intent mapping table (111) and the special intent mapping table (112), determine the OID operation item or special procedure, execute the OID operation or execute the special procedure, obtain the result text of the OID operation or special procedure execution, and convert the result text into voice broadcast and feedback to the user. Step S3 includes: determining the special functions that need to be completed by voice control of the printer (100) based on the attributes of the current printer (100) and the various query and configuration operations that can be performed; A dedicated intent mapping table (112) is formed for the dedicated functions that need to be completed by voice control of the printer (100). The dedicated intent mapping table (112) contains a dedicated function program list corresponding to the dedicated functions. For the dedicated functions in the dedicated intent mapping table (112), determine the operation intent and supplementary information corresponding to the dedicated intent mapping table (112); and mark the operation intent and supplementary information corresponding to the dedicated functions contained in the dedicated intent mapping table (112).

2. The method for implementing voice interaction in a printer according to claim 1, characterized in that, Traverse the various OID nodes corresponding to the printer (100), select the target OID node, construct a general intent mapping table (111) for the target OID node, and form the content of the OID column of the general intent mapping table (111). Mark the operation intent and supplementary information corresponding to each row of OID nodes in the general intent mapping table (111), including: The OID traversal module (201) performs traversal operations on various OID nodes, sequentially traversing the printer manufacturer's private nodes, printer-related general nodes, and device-related general nodes, and selecting OID nodes suitable for printer voice control as target OID nodes. By collecting the OID information of the target OID node, a general intent mapping table (111) is formed; wherein the general intent mapping table (111) contains OID items and corresponding OID information.

3. The method for implementing voice interaction in a printer according to claim 1, characterized in that, The operation intent and supplementary information corresponding to each OID node in the general intent mapping table (111) are labeled, including: Iterate through each row of the general intent mapping table (111) and obtain each OID item contained in the general intent mapping table (111); For each OID item in the general intent mapping table (111), determine the corresponding operation intent and supplementary information of the general intent mapping table (111); The operation intent and supplementary information corresponding to each OID item in the general intent mapping table (111) are labeled.

4. The method for implementing voice interaction in a printer according to claim 1, characterized in that, Constructing training samples for the intent recognition model, training the intent recognition model using the training samples, and deploying the completed training and testing intent recognition model to the printer (100), including: Construct training samples, which include user instruction text and operation intent labels and / or supplementary parameter labels corresponding to the user instruction text. The operation intent labels and / or supplementary parameter labels are matched one-to-one with the operation intents and / or supplementary parameters in the general intent mapping table (111) and the special intent mapping table (112). The intent recognition model is trained based on training samples; Construct test samples, test the intent recognition model based on the test samples, and calculate the label inference probability of each operation intent label and / or supplementary parameter label. When all label inference probabilities are less than 60%, it is determined that the text intent is not highly relevant to all existing labels, and the message "Sorry, I can only understand print settings related instructions" is output. The intent recognition model is trained and tested multiple times based on training and test samples until the intent accuracy and supplementary parameter accuracy reach the preset qualified standards. Then the training of the intent recognition model is considered complete. Deploy the completed intent recognition model to the printer (100).

5. The method for implementing voice interaction in a printer according to claim 1, characterized in that, Real-time acquisition of voice information sent by the user to the printer (100), and conversion of the voice information into text data, including: The user sends a voice command to the printer (100), the voice acquisition and command text recognition module (101) captures the voice information and performs voice information noise processing to obtain the voice information after noise processing; and converts the voice information after noise processing into text data. The voice acquisition and command text recognition module (101) captures voice information and performs voice information noise processing, including: Automatically initiate noise sampling according to the noise sampling window, collect current ambient noise, and extract baseline features: Environmental noise is classified based on baseline characteristics to determine noise types, including steady-state noise, transient noise, and mixed noise. For steady-state noise, a high-frequency retention filtering strategy is adopted to deeply suppress the noise peak frequency band of steady-state noise; wherein, the deep suppression refers to the suppression of the noise peak frequency band by an attenuation of 7dB-10dB; For transient noise, only the transient noise portion of the speech information whose noise duration is less than a preset specific duration t is moderately suppressed. The moderate suppression refers to the suppression of the noise peak frequency band by an attenuation of 4dB-6dB. The preset specific duration t ranges from 180ms to 220ms. For mixed noise, noise suppression is performed using the noise energy parameter E corresponding to the time period of the mixed noise.

6. The method for implementing voice interaction in a printer according to claim 5, characterized in that, Converting the noise-processed speech information into text data includes: Retrieve the text data corresponding to the converted speech information; Scan the text data and extract the keywords contained within it; Retrieve the voice information processing history of the printing device from the cloud platform corresponding to the printing device, and retrieve the frequency of keyword occurrence and the accuracy of voice command execution corresponding to the keyword from the voice information processing history. When there are multiple keywords, the correlation coefficient between each pair of keywords in the multiple related words is determined according to the speech information processing history. The confidence score of the text data is obtained by combining the frequency of occurrence of the keywords, the accuracy of voice command execution corresponding to the keywords, and the correlation coefficient between each pair of keywords. The confidence score of the text data is compared with a preset confidence threshold. When the confidence level of the text data is not lower than the preset confidence threshold, the text data is determined to have passed verification, and subsequent steps are executed. When the confidence level of the text data is lower than the preset confidence threshold, the system will issue a voice broadcast to the user to read the text data content and confirm the instruction with the user again.

7. The method for implementing voice interaction in a printer according to claim 1, characterized in that, The intention and supplementary parameters corresponding to the text data are obtained by reasoning through the intention recognition model, including: The text data is input to the instruction text intent recognition module (102); wherein, the instruction text intent recognition module (102) is equipped with an intent recognition model; The intention recognition model deployed in the instruction text intention recognition module (102) is used to reason about the text data and extract the operation intention and supplementary parameters in the text data.

8. The method for implementing voice interaction in a printer according to claim 1, characterized in that, The system performs voice information matching and operation intent determination by searching the general intent mapping table (111) and the special intent mapping table (112), and provides voice broadcast and feedback on the query results and operation intent, including: The instruction execution scheduling module (103) independently queries the general intent mapping table (111) and the special intent mapping table (112). If an OID item corresponding to the operation intent is found in the general intent mapping table (111), the OID management module (104) controls the OID management module (104) to perform the corresponding operation through the get / set standard command of the SNMP protocol according to the OID item matched in the general intent mapping table (111), and passes the query and setting operation results to the speech synthesis module (106); the speech synthesis module (106) synthesizes speech broadcast content based on the matching results and broadcasts it. If no OID item corresponding to the operation intent is found in the general intent mapping table (111), the instruction execution scheduling module (103) returns "Unable to execute your instruction" to the speech synthesis module (106), and the speech synthesis module (106) broadcasts the content. If a matching item corresponding to the operation intent is found in the dedicated intent mapping table (112), the instruction execution scheduling module (103) calls the program execution content corresponding to the matching item in the dedicated execution module set (105) to complete the query and setting operation results, and passes the query and setting operation results to the speech synthesis module (106); the speech synthesis module (106) synthesizes speech broadcast content based on the matching results and performs speech broadcast; If no matching item corresponding to the operation intent is found in the dedicated intent mapping table (112), the instruction execution scheduling module (103) returns "Unable to execute your instruction" to the speech synthesis module (106), and the speech synthesis module (106) broadcasts the content.

9. A system for implementing voice interaction in a printer, characterized in that, The system includes: The general intent mapping table forming module is used to traverse the various OID nodes corresponding to the printer (100), select the target OID node, construct a general intent mapping table (111) for the target OID node, and form the content of the OID column of the general intent mapping table (111); The standard module of the general intent mapping table is used to annotate the operation intent and supplementary information corresponding to each OID node in the general intent mapping table (111); The dedicated intent mapping table forming module is used to form a dedicated function program list of the dedicated intent mapping table (112) according to the dedicated functions to be completed by voice control of the printer as needed, and to complete the annotation of its corresponding operation intent and supplementary information. The dedicated intent mapping table standard module is used to construct training samples for the intent recognition model based on the operation intent and supplementary information listed in the general intent mapping table (111) and the dedicated intent mapping table (112), and to train and test the intent recognition model using the training samples, and to deploy the intent recognition model corresponding to the completed training and testing to the printer (100). The voice acquisition and voice semantic data processing module is used to acquire voice information sent by the user to the printer (100) in real time, convert the voice information into text data, and obtain the operation intention and / or supplementary parameters corresponding to the text data by reasoning through the intent recognition model. The query and voice feedback control module is used to query the corresponding voice information and determine the operation intent by looking up the general intent mapping table (111) and the special intent mapping table (112), and to broadcast and provide feedback on the query results and operation intent. The dedicated intent mapping table forming module is used to determine the dedicated functions that need to be completed by voice control of the printer (100) based on the attributes of the current printer (100) and the various query and configuration operations that can be performed. A dedicated intent mapping table (112) is formed for the dedicated functions that need to be completed by voice control of the printer (100). The dedicated intent mapping table (112) contains a dedicated function program list corresponding to the dedicated functions. For the dedicated functions in the dedicated intent mapping table (112), determine the operation intent and supplementary information corresponding to the dedicated intent mapping table (112); and mark the operation intent and supplementary information corresponding to the dedicated functions contained in the dedicated intent mapping table (112).

Images