Full-automatic intelligent archival repository accurate voice instruction recognition method based on WFST

Abstract

The invention provides a WFST-based full-automatic intelligent archival repository accurate voice instruction recognition method, relates to the technical field of voice recognition, and aims to solve the problems that in a mechanical automatic archival repository scene, unknown voice instructions or wrong voice instructions cannot be recognized, and comprehensive voice recognition cannot be performed. The method specifically comprises the steps of obtaining a preset voice text instruction of an archival repository; according to the voice text instruction, constructing a WFST decoding graph of an archival repository voice instruction, obtaining a to-be-recognized archival repository voice instruction, matching the to-be-recognized archival repository voice instruction with the trained WFST decoding graph, and recognizing a specific voice operation instruction of a user; and if the voice instruction issued by the user is not standard or an incorrect instruction is issued, performing secondary recognition by combining an ASR technology with an intention classification method, and recognizing a voice operation instruction. Accurate recognition of unknown voice instructions or wrongly recognized voice instructions and the like is achieved, and then accurate and comprehensive voice instruction recognition of a full-automatic intelligent archival repository is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of speech recognition technology, and in particular to a method for accurate speech command recognition in a fully automatic intelligent archive warehouse based on WFST. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] With the development of artificial intelligence technology, the construction and management of archive warehouses have also become more intelligent, with extremely high requirements for archive storage, search, borrowing and returning. As the number of archives increases, the operation time of archives becomes longer and longer. Traditional archive management cannot achieve fast storage, borrowing and returning of archives, especially in the scenario of mechanically automated archive warehouses.

[0004] Speech recognition technology is the most convenient and efficient way of human-computer interaction. Through speech recognition technology, data archives can be searched, etc. However, the existing archive warehouse speech recognition technology still has the following shortcomings:

[0005] (1) The scene is relatively fixed, and it can only recognize existing voice commands, but cannot recognize unknown voice commands. It does not process unknown or unclear voice commands;

[0006] (2) During the voice command recognition process, when an erroneous voice command is issued, the appropriate path cannot be found, and no effective method can be taken to find the appropriate path again, and the correct voice operation command cannot be recognized. Summary of the invention

[0007] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a method for accurate voice command recognition of a fully automatic intelligent archive warehouse based on WFST, which performs voice control on archive warehouse equipment through voice recognition technology to improve the intelligence level of the company's archive warehouse.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] The first aspect of the present invention provides a method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST, comprising:

[0010] Get the voice and text instructions preset in the archive warehouse;

[0011] According to the voice text instruction, a WFST decoding graph of the archive warehouse voice instruction is constructed. The specific process is as follows:

[0012] Use pronunciation dictionary, acoustic model, and language model to build WFST decoding graph;

[0013] According to the voice commands of the actual scene, the content of the WFST decoding graph is changed to obtain a new voice command WFST decoding graph;

[0014] According to the voice command of the actual scene, the new WFST decoding graph is trained to obtain a trained WFST decoding graph;

[0015] Obtain the archive warehouse voice command to be recognized, and match it with the trained WFST decoding graph to recognize the user's specific voice operation command;

[0016] If the voice command issued by the user is not standard or is an erroneous command, the ARS technology is combined with the intent classification method to determine the user's intention. Based on the user's intention classification, the user is guided to re-issue the voice command for secondary voice recognition.

[0017] As an implementation method, the specific process of obtaining the voice text instructions preset in the archive warehouse is as follows:

[0018] Get the preset voice commands in the archive warehouse;

[0019] Converting the preset voice command into a voice text command;

[0020] Construct a speech-to-text command dataset for the archive repository.

[0021] As an implementation method, the content in the WFST decoding graph is changed according to the voice command of the actual scene to obtain a new voice command WFST decoding graph. The specific process is:

[0022] Constructing archive name database:

[0023] According to the information of the archive name database, the content in the WFST decoding graph is replaced to obtain a new WFST decoding graph.

[0024] As an implementation mode, the voice command of the archive warehouse to be recognized is obtained, and matched with the trained WFST decoding graph to recognize the specific voice operation command of the user. The specific process is as follows:

[0025] Obtain the voice commands of the archive warehouse to be recognized in real time;

[0026] Converting the voice command of the archive warehouse to be identified into a voice text command;

[0027] The voice text instructions are compared with the words in the WFST decoding graph. If the corresponding voice instruction words or semantically consistent content are identified, the most suitable path is matched in the WFST decoding graph to identify the specific voice operation instructions of the voice instruction.

[0028] As an implementation method, the speech secondary recognition specifically includes the following steps:

[0029] The pre-trained model BERT text classification model is used to input the preset voice instructions of the archive warehouse into the pre-trained model BERT text classification model to train the voice text instruction intention of the archive warehouse;

[0030] Obtain the voice command to be recognized, and convert the voice to be recognized into text command using ASR technology;

[0031] The voice text command to be recognized is input into the trained BERT text classification model to match the corresponding voice text command intention. The semantics of the voice command issued by the user is consistent with the meaning of the voice text command. Figure 1 To;

[0032] Guide the user to re-issue the voice command to be recognized;

[0033] The voice text instruction to be recognized is matched to the path with the largest weight and the shortest length in the WFST decoding graph to identify the user's specific voice operation instruction.

[0034] As an implementation method, the WFST decoding graph of the archive warehouse voice command is constructed based on the TLG graph.

[0035] The second aspect of the present invention provides a fully automatic intelligent archive warehouse accurate voice command recognition system based on WFST, comprising:

[0036] The voice text instruction acquisition module is used to acquire the preset voice instructions of the archive warehouse, convert the preset voice instructions into voice text instructions, and construct a preset voice text instruction data set;

[0037] The WFST decoding graph construction module of the voice command is used to construct the WFST decoding graph using the TLG model according to the preset voice text command data set, update the WFST decoding graph according to the actual scenario, and train the WFST decoding graph to obtain the trained WFST decoding graph;

[0038] The first voice recognition module is used to match the voice command to be recognized with the trained WFST decoding graph to identify the specific voice operation command.

[0039] The second voice recognition module is used to determine the user's intention through ARS technology combined with the intention classification method if the voice command issued by the user is not standard or the command issued is wrong. According to the user's intention classification, it guides the user to re-issue the voice command for re-recognition.

[0040] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method as described in the first aspect of the present invention.

[0041] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method as described in the first aspect of the present invention are implemented.

[0042] A fifth aspect of the present invention provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the methods and functions involved in any of the above-mentioned embodiments.

[0043] One or more of the above technical solutions have the following beneficial effects:

[0044] In this embodiment, by constructing the WFTS decoding diagram of the archive warehouse voice command, the voice operation command is accurately identified. At the same time, the ARG technology and intent classification method are used to achieve accurate recognition of unknown voice commands or erroneous voice commands.

[0045] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0047] Figure 1 Flow chart of the WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method in this embodiment;

[0048] Figure 2 Schematic diagram of a WFST decoding diagram of a fully automatic intelligent archive warehouse accurate voice command recognition method based on WFST in this embodiment;

[0049] Figure 3 A schematic diagram of name replacement in the fully automatic intelligent archive warehouse accurate voice command recognition method based on WFST in this embodiment;

[0050] Figure 4 Schematic diagram of the replaced WFST decoding diagram of the fully automatic intelligent archive warehouse accurate voice command recognition method based on WFST in this embodiment. DETAILED DESCRIPTION

[0051] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0053] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0054] Glossary:

[0055] WFST: Weighted Finite State Transition Machine. A WFST graph has a series of states, each state must participate in the operation, there are connecting arcs between states, and each connecting arc has an input label, an output label and a weight.

[0056] The overall idea proposed by the present invention is as follows: first, the preset voice instructions of the mechanical automation archive warehouse are obtained, the audio is converted into text using the ASR technology, a WFST decoding graph of the archive warehouse is constructed, and the WFST decoding graph is changed according to the actual scenario to realize the recognition of more voice instructions, and the language instructions to be recognized are matched with the constructed WFST decoding graph to identify the user's specific voice operation instructions. If the match is unsuccessful, the ASR technology is combined with the intent classification algorithm for secondary recognition to obtain accurate recognition results.

[0057] Embodiment 1

[0058] join Figure 1 As shown, this embodiment discloses a method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST, including:

[0059] S1: Obtain the voice text instructions preset in the archive warehouse;

[0060] S2: According to the voice text instruction, a WFST decoding graph of the archive warehouse voice instruction is constructed, and the specific process is as follows:

[0061] Use pronunciation dictionary, acoustic model, and language model to build WFST decoding graph;

[0062] According to the voice commands of the actual scene, the content of the WFST decoding graph is changed to obtain a new voice command WFST decoding graph;

[0063] According to the voice command of the actual scene, the new WFST decoding graph is trained to obtain a trained WFST decoding graph;

[0064] S3: Obtain the archive warehouse voice command to be recognized, and match it with the trained WFST decoding graph to identify the user's specific voice operation command;

[0065] S4: If the voice command issued by the user is not standard or an erroneous command is issued, the ARS technology is combined with the intent classification method to determine the user's intent. Based on the user's intent classification, the user is guided to re-issue the voice command for secondary voice recognition.

[0066] In step S1, the voice text instructions preset in the archive warehouse are obtained, and the specific process is as follows;

[0067] S101: Obtaining a voice command preset in the archive warehouse;

[0068] Specifically, according to the scenario of the mechanically automated archive warehouse, common voice commands of the archive warehouse are issued in sequence through a microphone or other pickup devices and received by the filing cabinet robot as preset voice commands, including: "Please open the door", "Please close the door", "Please take out XXX files", "Please take out XXX and XXX files", "Please take out XXX, XXX and XXX files", "Please take out XXX, XXX, XXX files", "Please take out XXX, XXX, XXX files", "Start retrieving files", "Start returning files", etc.

[0069] In this embodiment, if there are already commonly used voice commands for the archive warehouse, there is no need to reset them.

[0070] S102: converting the preset voice command into a voice text command, specifically: using ASR recognition technology to recognize the preset voice command as a voice text command;

[0071] After the above steps, the conversion of archive warehouse voice commands into voice text commands is realized, which is convenient for subsequent feature extraction.

[0072] S103: Construct a voice text instruction dataset in a mechanical automation archive warehouse scenario;

[0073] Specifically, a preset voice-text instruction dataset is constructed, and the preset voice-text instruction dataset is: ("Please open the door", "Please close the door", "Please take out XXX file", "Please take out XXX and XXX files", "Please take out XXX, XXX and XXX files", "Please take out XXX, XXX, XXX files", "Start taking out files", "Start returning files"), where XXX represents the name of the file holder. The dataset also includes description instructions with similar semantics.

[0074] In this embodiment, when an unknown or erroneous voice instruction occurs, the user is guided to issue it again, or is reminded of the voice instruction format, and the voice text instruction is retrieved again.

[0075] After the above steps, the voice text instructions in the mechanical automation archive warehouse scenario can be obtained, which provides a data source for constructing the WFST generation graph of the archive warehouse voice instructions.

[0076] In step S2, according to the voice text instruction, a method for constructing a WFST decoding graph of the archive warehouse voice instruction is specifically as follows:

[0077] S201: Construct a WFST decoding graph using the pronunciation dictionary, acoustic model, and language model;

[0078] Through TLG's Compose, the acoustic modeling unit, dictionary, and language model are combined to generate a static decoding network. A search strategy is used during the decoding process to obtain the optimal decoding result for the input speech.

[0079] Specifically, take "Please take out Li Ming's file" as an example to construct the TLG graph of WFST, that is, the sentence to be recognized is "Please take out Li Ming's file".

[0080] Among them, T inputs the probability of the Chinese character consonants and finals of the voice command, L is the dictionary, which inputs the Chinese character pinyin of the voice command, and G is the voice model, which inputs / outputs the Chinese characters of the voice command to construct the G graph.

[0081] like Figure 2 As shown, in this embodiment, state nodes 0, 1, 2, 3, 4, 5, and 6 are set, and connection edges 0-1, 1-2, 2-3, 3-4, 4-5, and 5-6 are established between the state nodes in sequence, and the input / output texts are set on the connection edges in sequence as: please, take, out, Li Ming, file, and case, and the corresponding weight values ​​1, 0.6, 0.6, 0.6, 1, and 0.6 are set respectively.

[0082] Use fstcompi le in the OpenFST tool to compile the constructed G graph of "Please take out Li Ming's file" into the g_with_s l ot.fst file, the content of which is as follows:

[0083]

[0084] Use fsttablecompose in the openfst tool to combine the constructed G graph with T and L to build the WFST decoding graph of "Please take out Li Ming's file".

[0085] The same process is used to construct WFST decoding graphs for commonly used voice commands in other scenarios of mechanical automation.

[0086] S202: According to the voice command of the actual scene, the content of the WFST decoding graph is changed to obtain a new voice command WFST decoding graph;

[0087] The voice commands in the archive warehouse contain entity information such as the names of the archive members, and different entity information libraries need to be built. For example, when it is necessary to operate archives with different names, it is necessary to replace different archive names.

[0088] like Figure 2 , Figure 3 , Figure 4 As shown, in this embodiment, if the voice command is "Please take out Zhang Wei's file", it is necessary to replace the input / output "Li Ming" of the 4-5 connecting edge in the G graph with "Zhang Wei".

[0089] (1) Build the archive name database and use fstcompile in the OpenFST tool to compile it into the name_slot.fst file, the content of which is as follows:

[0090]

[0091] In this embodiment, if there is other entity information in the voice command of the archive warehouse, the required entity information library is constructed according to the actual scenario.

[0092] (2) Replace the content of the corresponding connecting edge in the initial G graph, that is, replace "Li Ming" with "Zhang Wei", and use fstrepa l ce in openfst to replace name_s l ot in g_with_s l ot.fst to form the final G graph g.fst. The specific commands are as follows:

[0093] fstreplace--epsilon_on_replace$lang / g_with_slot.fst$root_label\

[0094] $lang / name_slot.fst$address_slot_label$lang / g.fst;

[0095] (3) Use fsttablecompose in the openfst tool to combine the replaced G graph with the predefined T and L graphs, as shown in the following command:

[0096] fsttablecompose$lang / L.fst$lang / g.fst|fstdeterminizestar

[0097] --use-log=true|\fstminimizeencoded|fstarcsort--sort_type=ilabel>

[0098] $lang / lg.fst||exit 1;

[0099] fsttablecompose$lang / T.fst$lang / lg.fst>$lang / tlg.fst||exit 1;

[0100] The final constructed TLG.fst graph, that is, the WFST decoding graph of the archive warehouse voice command is completed.

[0101] In this embodiment, the WFST decoding graph can be constructed or changed according to different voice instructions based on the actual scenario of the mechanical automation archive warehouse.

[0102] After the above steps, the WFST decoding graph of the archive warehouse voice command is obtained, and by changing the content of the WFST decoding graph, a more comprehensive WFST decoding graph is obtained, which provides a basis for the recognition of subsequent voice commands and makes the recognized user's specific voice operation instructions more accurate.

[0103] S203: training the constructed WFST decoding graph according to the voice instructions of the actual scene;

[0104] In order to improve the accuracy and robustness of the WFST decoding graph, the constructed WFST decoding graph is trained.

[0105] Specifically, the voice text instructions with the name of the person in the archive in the preset voice text instruction data set are used as training samples of the WFST decoding graph, and are input into the WFST decoding graph for matching. The user's specific voice operation instructions are identified in turn, and the path with the maximum weight and the shortest length is matched to obtain an accurate recognition result.

[0106] After the above steps, a trained WFST decoding graph is obtained, which improves the accuracy of WFST decoding graph recognition and makes WFST decoding graph recognition more accurate.

[0107] like Figure 1 As shown, in step S3, the archive warehouse voice command to be recognized is obtained and matched with the trained WFST decoding graph to recognize the user's specific voice operation command. The specific process is as follows:

[0108] S301: Acquire the voice command of the archive warehouse to be recognized in real time;

[0109] Specifically, the user issues voice commands corresponding to the scenario based on the actual scenario.

[0110] S302: Converting the voice command of the archive warehouse to be recognized into a voice text command;

[0111] S303: Compare the voice text information with the words in the WFST decoding graph, and if the corresponding voice instruction words or semantically consistent content are identified, match the most suitable path in the WFST decoding graph to identify the specific voice operation instructions of the voice instruction, thereby obtaining a more accurate recognition result.

[0112] In this embodiment, for example, when the voice instruction to be recognized is "Please take out Li Si's file", the ASR technology is used to convert the voice instruction to be recognized into a voice text instruction. In the constructed voice instruction WFST decoding graph, the name library is called to recognize the corresponding words, and the most suitable path is searched. By finding the path with the largest weight and the shortest length, the specific voice operation of the voice instruction "Please take out Li Si's file" can be recognized and sent to the corresponding mechanical equipment in the archive warehouse to execute the instruction operation.

[0113] After the above steps, the specific voice operation instructions of the user's voice commands are identified, providing accurate data support for the subsequent issuance of instructions to the archive warehouse mechanical equipment to complete the operation.

[0114] like Figure 1 As shown, in this embodiment, in step S4, since different people have different pronunciation organs, accents and speaking styles, the voice instructions issued may be non-standard or erroneous, resulting in failure to find a suitable path for the voice instruction to be recognized in the WFST decoding diagram of the archive warehouse voice instruction, and failure to recognize the user's specific voice operation instructions.

[0115] If the voice command issued by the user is not standard or an erroneous command is issued, the user's intention is judged through the intention classification method, and the user is guided to re-issue the voice command based on the user's intention classification for secondary recognition.

[0116] The specific implementation process is:

[0117] (1) Using the pre-trained model BERT text classification model, the voice instructions preset by the archive warehouse are input into the pre-trained model BERT text classification model to train the voice text instruction intention of the archive warehouse;

[0118] (2) Obtaining the voice command to be recognized;

[0119] For example, the user issues a voice command of "I want to create a file named Li Xiaohong".

[0120] (3) Use ASR technology to convert the speech to be recognized into text;

[0121] Convert the user's voice command "I want to name the file Li Xiaohong" into a text command.

[0122] (4) Input the voice text command to be recognized into the trained BERT text classification model to match the corresponding voice text command intention, that is, the semantics of the voice command issued by the user and the meaning of the voice text command. Figure 1 To;

[0123] Input the text instruction "I want to name the file named Li Xiaohong" into the trained bert text classification model, and find the text instruction "I want to name the file named Li Xiaohong" Figure 1 The voice text command is "Please take out Li Xiaohong's file".

[0124] (5) guiding the user to re-issue the voice command to be recognized;

[0125] The user re-issued the voice command: "Please retrieve Li Xiaohong's file."

[0126] (6) The voice text instruction to be recognized is matched to the path with the largest weight and the shortest length in the WFST decoding graph, that is, the user's specific voice operation instruction is recognized.

[0127] The recognized word "Please take out Li Xiaohong's file" is matched to the most suitable path in the path of TLG.fst in the WFST decoding graph, thus obtaining a more accurate recognition result.

[0128] After the above steps, accurate recognition of "unknown voice commands and erroneous commands" is achieved, more accurate specific voice operation instructions are obtained, and accurate recognition of voice commands in mechanical automation archive warehouses is realized.

[0129] Embodiment 2

[0130] like Figure 1-4 As shown, this embodiment provides a fully automatic intelligent archive warehouse accurate voice command recognition system based on WFST, including:

[0131] The voice text instruction acquisition module is used to acquire the preset voice instructions of the archive warehouse, convert the preset voice instructions into voice text instructions, and construct a preset voice text instruction data set;

[0132] The WFST decoding graph construction module of the voice command is used to construct the WFST decoding graph using the TLG model according to the preset voice text command data set, update the WFST decoding graph according to the actual scenario, and train the WFST decoding graph to obtain the trained WFST decoding graph;

[0133] The first voice recognition module is used to match the voice command to be recognized with the trained WFST decoding graph to identify the specific voice operation command.

[0134] The second voice recognition module is used to determine the user's intention through ARS technology combined with the intention classification method if the voice command issued by the user is not standard or the command issued is wrong. According to the user's intention classification, it guides the user to re-issue the voice command for secondary recognition.

[0135] Embodiment 3

[0136] This embodiment provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the steps in the WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method as described in the first aspect of the present invention are implemented.

[0137] Embodiment 4

[0138] This embodiment provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in the WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method as described in the first aspect of the present invention are implemented.

[0139] Embodiment 5

[0140] The purpose of this embodiment is to provide a computer program product containing instructions, which, when running on a computer, enables the computer to execute the methods and functions involved in any of the above embodiments.

[0141] The steps involved in the apparatus of the above embodiment correspond to the method embodiment 1, and the specific implementation method can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0142] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0143] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST, characterized in that: include: Get the voice and text instructions preset in the archive warehouse; According to the voice text instruction, a WFST decoding graph of the archive warehouse voice instruction is constructed. The specific process is as follows: Use pronunciation dictionary, acoustic model, and language model to build WFST decoding graph; According to the voice commands of the actual scene, the content of the WFST decoding graph is changed to obtain a new voice command WFST decoding graph; According to the voice command of the actual scene, the new WFST decoding graph is trained to obtain a trained WFST decoding graph; Obtain the archive warehouse voice command to be recognized, and match it with the trained WFST decoding graph to recognize the user's specific voice operation command; If the voice command issued by the user is not standard or is an erroneous command, the ARS technology is combined with the intent classification method to determine the user's intention. Based on the user's intention classification, the user is guided to re-issue the voice command for secondary voice recognition.

2. A method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST as claimed in claim 1, characterized in that: The specific process of obtaining the preset voice text instructions of the archive warehouse is as follows: Get the preset voice commands in the archive warehouse; Converting the preset voice command into a voice text command; Construct a speech-to-text command dataset for the archive repository.

3. A method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST as claimed in claim 1, characterized in that: The voice command according to the actual scene is used to change the content of the WFST decoding graph to obtain a new voice command WFST decoding graph. The specific process is as follows: Constructing archive name database: According to the information of the archive name database, the content in the WFST decoding graph is replaced to obtain a new WFST decoding graph.

4. A method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST as claimed in claim 1, characterized in that: The process of obtaining the voice command of the archive warehouse to be identified and matching it with the trained WFST decoding graph to identify the specific voice operation command of the user is as follows: Obtain the voice commands of the archive warehouse to be recognized in real time; Convert the voice command of the archive warehouse to be recognized into voice text command; The voice text instructions are compared with the words in the WFST decoding graph. If the corresponding voice instruction words or semantically consistent content are identified, the most suitable path is matched in the WFST decoding graph to identify the user's specific voice operation instructions.

5. The method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST as claimed in claim 1, characterized in that: The specific process of the secondary speech recognition is as follows: The pre-trained model BERT text classification model is used to input the preset voice instructions of the archive warehouse into the pre-trained model BERT text classification model to train the voice text instruction intention of the archive warehouse; Obtain the voice command to be recognized, and convert the voice to be recognized into text command using ASR technology; Input the voice text command to be recognized into the trained BERT text classification model to match the corresponding voice text command intent. The semantics of the voice command issued by the user is consistent with the voice text command intent. Guide the user to re-issue the voice command to be recognized; The voice text instruction to be recognized is matched to the path with the largest weight and the shortest length in the WFST decoding graph to identify the user's specific voice operation instruction.

6. A method for accurate voice command recognition in a fully automatic intelligent archive warehouse based on WFST as claimed in claim 1, characterized in that: The WFST decoding graph of the archive warehouse voice command is constructed based on the TLG graph.

7. A fully automatic intelligent archive warehouse accurate voice command recognition system based on WFST, characterized in that: The WFST-based fully automatic intelligent archive warehouse accurate voice command recognition method according to any one of claims 1 to 6 comprises: The voice text instruction acquisition module is used to acquire the preset voice instructions of the archive warehouse, convert the preset voice instructions into voice text instructions, and construct a preset voice text instruction data set; The WFST decoding graph construction module of the voice command is used to construct the WFST decoding graph using the TLG model according to the preset voice text command data set, update the WFST decoding graph according to the actual scenario, and train the WFST decoding graph to obtain the trained WFST decoding graph; The first voice recognition module is used to match the voice command to be recognized with the trained WFST decoding graph to identify the specific voice operation command. The second voice recognition module is used to determine the user's intention through ARS technology combined with the intention classification method if the voice command issued by the user is not standard or the command issued is wrong. According to the user's intention classification, it guides the user to re-issue the voice command for secondary recognition.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method described in any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 6 are performed.

Images