A multi-language voice data processing method for real-time talkback communication of marine VHF equipment

Abstract

This invention relates to the field of data processing technology and discloses a multilingual voice data processing method for real-time intercom communication of VHF equipment at sea. This method addresses the problem that during voice data processing, intercom content may be presented in a simplified manner rather than with complete semantics. The method includes: determining the context dependency of the current intercom round corresponding to the correction text; when the correction text is determined to have a context dependency, determining semantic missingness to obtain a semantic missingness determination result; when the semantic missingness determination result indicates that the correction text has semantic missingness, extracting context text associated with the current intercom round from historical intercom text, and completing the missing communication object, location description, risk background, or operation object in the correction text based on the context text to obtain the final completed text. This effectively improves the accuracy, continuity, and practical reliability of multilingual voice data processing at sea.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a multilingual voice data processing method for real-time two-way communication of VHF equipment at sea. Background Technology

[0002] With the continuous development of maritime communication support systems, intelligent shipping management technologies, and multilingual voice processing technologies, the application of maritime VHF equipment in scenarios such as ship navigation command, collision avoidance coordination, waterway passage management, and emergency communication is becoming increasingly widespread. During navigation, ships continuously generate a large amount of real-time intercom voice data, channel communication data, transmit / receive identification data, and sea area location correlation data. This data is typically acquired by shipboard VHF equipment, communication management terminals, or navigation assistance systems, and then processed and analyzed uniformly through corresponding voice processing platforms, communication analysis platforms, or liaison assistance systems. In real-time maritime communication, the system usually needs to perform language recognition, text conversion, terminology correction, semantic extraction, and multilingual conversion processing on the intercom content based on the aforementioned voice data and communication correlation data, in order to output understandable communication content to communication recipients with different language backgrounds. Therefore, how to accurately process multilingual voice data and improve the completeness of communication semantic output results in continuous intercom communication scenarios at sea has become an important and continuously focused technical direction in the field of maritime intelligent communication processing.

[0003] In existing technologies, maritime communication processing systems based on real-time speech recognition and multilingual conversion have been gradually applied. These systems typically acquire real-time intercom voice data generated by maritime VHF equipment, preprocess the voice content, identify the language, convert it to text, and perform semantic analysis. Based on the analysis results, they generate corresponding output content in the target language. By utilizing processing technologies such as speech recognition, terminology correction, and multilingual conversion, these systems can automatically process maritime communication content in different language contexts, thereby improving the efficiency and intelligence of information transmission during maritime communication.

[0004] In existing maritime VHF real-time intercom voice processing mechanisms, the voice content generated in a single intercom session is typically treated as an independent processing object. Each corresponding voice segment undergoes segment-by-segment recognition, conversion, and semantic parsing, with the corresponding text content serving as the primary basis for subsequent semantic extraction and target language output. During multilingual conversion, the system generally generates the corresponding standard semantic content or target language expression directly based on the recognized text corresponding to the current voice segment, thus completing the real-time processing of a single intercom message. This approach is widely adopted in practice. Its basic idea is to maintain a clear voice processing chain by independently recognizing and converting each voice message, facilitating rapid processing and timely output of real-time voice data.

[0005] However, the above-mentioned technologies have at least the following technical problems: In the actual operational environment of VHF real-time intercom communication at sea, continuous communication between ships is often characterized by concise statements, compressed expression, strong contextual continuity, and clear semantic connections across rounds. Some intercom content does not fully convey the semantics of the communication target, positional relationship, risk background, or operational instructions in the current round. Instead, it builds upon previous intercom content, supplementing, confirming, correcting, or briefly expressing only partial information. For example, after the target vessel, relative bearing, or avoidance requirements have been clearly stated in the previous round, the subsequent round often only provides follow-up expressions such as "maintain original course," "continue to avoid," "turn slightly to starboard," or "execute as before." While this type of intercom content can be correctly understood by both parties in conjunction with the preceding context, the voice text corresponding to the current round itself usually lacks a complete and independent semantic structure.

[0006] In this situation, if the communication content is still processed independently based solely on the recognized text corresponding to the current single intercom segment, semantic parsing and multilingual conversion may fail to effectively complete the preceding communication objects, location descriptions, risk relationships, or operational context upon which the current intercom content depends. This results in a system output that, while completing recognition and translation at the literal level, still suffers from semantic issues such as missing objects, unclear referents, incomplete command context, or deviations in communication meaning. These problems are amplified, especially in real-time multilingual conversion scenarios, making it difficult for the receiving end to accurately reconstruct the complete communication intent during continuous intercom based solely on the current output. This leads to a discrepancy between the semantic output and the actual communication context during real-time maritime intercom communication, ultimately affecting the accuracy, continuity, and reliability of multilingual voice data processing at sea. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a multilingual voice data processing method for real-time intercom communication of marine VHF equipment, so as to solve the problems existing in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for processing multilingual voice data for real-time two-way communication using a maritime VHF device includes the following steps: Step 1: Acquire raw two-way voice data generated by the maritime VHF device and simultaneously acquire the corresponding channel information, including channel number, transmission time, transmitter identifier, receiver identifier, and sea area location data; construct a voice dataset to be processed based on the raw two-way voice data and corresponding communication information; Step 2: Preprocess the raw two-way voice data in the voice dataset to be processed to obtain purified voice data; and perform voice activity detection on the purified voice data to divide it into multiple effective voice segments; Step 3: Extract voice feature data from each effective voice segment, identify the corresponding target language based on the voice feature data, call the corresponding speech recognition model according to the target language, and perform text conversion processing on each effective voice segment to obtain speech recognition text; Step 4: Perform terminology correction processing on the speech recognition text to obtain corrected text; Step 5: Analyze the current two-way voice data corresponding to the corrected text. The process involves several steps: Step 6: Context dependency determination is performed on the communication round. If context dependency is found in the correction text, semantic missingness determination is performed to obtain a semantic missingness result. When the semantic missingness determination result indicates that the correction text has semantic missingness, context text related to the current communication round is extracted from historical communication texts. Based on the context text, missing communication objects, location descriptions, risk backgrounds, or operation objects in the correction text are completed to obtain the final completed text. Step 7: Semantic parsing is performed on the final completed text to extract communication objects, event types, location descriptions, risk descriptions, and operation instruction information. Standard semantic data is generated based on the extraction results. The corresponding target output language is determined based on the standard semantic data. Step 8: Multilingual conversion processing is performed on the standard semantic data according to the target output language to obtain the target output text and the corresponding target language voice data. The target output text, target language voice data, and standard semantic data are sent to the corresponding maritime VHF equipment and stored together.

[0009] Preferably, the step of determining the context dependency of the current intercom round corresponding to the correction text is as follows: Obtain the sequence of text terms contained in the correction text, and identify the term type of each text term in the sequence to obtain a term type marker corresponding to each text term; wherein, the term type marker includes communication object terms, location description terms, risk description terms, operation object terms, pronoun terms, successor terms, and operation instruction terms; based on the term type markers corresponding to each text term, perform complete semantic item detection on the correction text to obtain a set of semantic items already contained in the correction text; wherein, the set of semantic items includes communication object, location description, risk background, and operation object; perform dependency trigger detection on pronoun terms, successor terms, and operation instruction terms in the correction text; when the correction text contains pronoun terms, successor terms, or operation instruction terms, determine that the current intercom round corresponding to the correction text has a context association trigger condition; when the correction text does not contain pronoun terms, successor terms, or operation instruction terms, determine that the current intercom round corresponding to the correction text does not have a context association trigger condition. The following steps relate to triggering conditions: When it is determined that the current intercom round has context-related triggering conditions, it is determined whether the semantic item set is missing at least one of the following: communication object, location description, risk background, or operation object; when at least one of the semantic item set is missing, it is determined that the current intercom round corresponding to the correction text has context dependency; when the semantic item set does not lack communication object, location description, risk background, or operation object, it is determined that the current intercom round corresponding to the correction text does not have context dependency due to semantic missingness; when it is determined that the current intercom round does not have context-related triggering conditions, it is determined whether the semantic item set contains communication object, location description, risk background, or operation object; when the semantic item set contains communication object, location description, risk background, or operation object, it is determined that the current intercom round corresponding to the correction text does not have context dependency; when the semantic item set is missing at least one of the following: communication object, location description, risk background, or operation object, it is determined that the current intercom round corresponding to the correction text is a round to be further dependently verified, and supplementary dependency verification is performed on the round to be further dependently verified.

[0010] Preferably, the step of identifying the term type of each text term in the text term sequence is as follows: Obtain each text term in the text term sequence, and match each text term with various term rules in a preset term rule library; wherein, the preset term rule library includes communication object term rules, location description term rules, risk description term rules, operation object term rules, pronoun term rules, successor term rules, and operation instruction term rules; when any text term matches a communication object term rule, the corresponding text term is marked as a communication object term; when any text term matches a location description term rule, the corresponding text term is marked as a location description ... When a term matches a risk description term rule or an operation object term rule, the corresponding text term is marked as a risk description term or an operation object term, respectively. When any text term matches a substitute term rule, a successor term rule, or an operation instruction term rule, the corresponding text term is marked as a substitute term, a successor term, or an operation instruction term, respectively. When any text term matches multiple term rules, the priority type of the corresponding text term is determined according to the order of communication object term rules, location description term rules, risk description term rules, operation object term rules, substitute term rules, successor term rules, and operation instruction term rules, so as to obtain a unique term type label for each text term.

[0011] Preferably, the step of performing complete semantic item detection on the correction text is as follows: obtaining the term type markers corresponding to each text term in the correction text, and classifying each text term according to the term type markers to obtain a communication object term set, a location description term set, a risk description term set, an operation object term set, a pronoun term set, a successor term set, and an operation instruction term set; performing object semantic extraction on the communication object term set to determine whether the correction text contains explicit communication object information; when there is at least one communication object term in the communication object term set, recording the corresponding communication object information into the communication object semantic item in the semantic item set; performing location semantic extraction on the location description term set to determine whether the correction text contains explicit location description information; when there is at least one location description term in the location description term set, recording the corresponding orientation information, distance information, channel location description information, or relative location description information into the location description semantic item in the semantic item set; performing risk semantic extraction on the risk description term set to determine whether the correction text contains explicit risk background. Information; when at least one risk descriptive term exists in the risk descriptive term set, the corresponding collision risk information, encounter risk information, interference risk information, or emergency risk information is recorded in the risk background semantic term in the semantic term set; correlation detection is performed on the operation object term set and the operation instruction term set to determine whether the correction text contains explicit operation object information; when at least one operation object term exists in the operation object term set, the corresponding heading adjustment object, speed adjustment object, or avoidance execution object is recorded in the operation object semantic term in the semantic term set; when there is an operation instruction term in the operation instruction term set and no operation object term in the operation object term set, the operation object semantic term in the semantic term set is recorded as a missing semantic term; based on the detection results of the communication object semantic term, position description semantic term, risk background semantic term, and operation object semantic term, a semantic term set corresponding to the correction text is generated; among them, the detected semantic terms are recorded as included semantic terms in the semantic term set, and the undetected semantic terms are recorded as missing semantic terms in the semantic term set for subsequent semantic missing determination.

[0012] Preferably, the step of supplementing dependency verification for the next dependency verification round is as follows: obtaining the correction text corresponding to the next dependency verification round and the semantic item set corresponding to the correction text, and determining the missing semantic item set corresponding to the next dependency verification round based on the semantic items not included in the semantic item set; performing abbreviation detection on each text term in the correction text to determine whether the correction text contains abbreviation terms; when it is determined that the correction text contains abbreviation terms, performing association determination on the correspondence between the abbreviation terms and the missing semantic item set; when the semantic reference of the abbreviation term cannot be independently determined in the current correction text, and the semantic reference corresponds to at least one item in the missing semantic item set, determining that the next dependency verification round has implicit contextual association conditions; when it is determined that the next dependency verification round has implicit contextual association conditions... When conditions are met, it is determined whether the abbreviated terms in the correction text require the context of the preceding intercom to determine the corresponding communication object, location description, risk background, or operation object. If the context of the preceding intercom is required, it is determined that the next verification round has a context dependency. If it is determined that there are no abbreviated terms in the correction text, or if there are abbreviated terms but their semantic meaning can be independently determined in the current correction text, it is determined that the next verification round does not have a context dependency. If it is determined that the next verification round has a context dependency, the correction text corresponding to the next verification round is used as the target text for subsequent context text extraction and processing. If it is determined that the next verification round does not have a context dependency, the correction text corresponding to the next verification round is used as an independent parsable text for subsequent semantic parsing processing.

[0013] Preferably, the step of determining semantic missingness in the corrected text and obtaining the semantic missingness determination result is as follows: obtaining a set of semantic items corresponding to the corrected text; wherein, the set of semantic items includes communication objects, location descriptions, risk backgrounds, and operation objects; reading the detection results corresponding to the semantic items of communication objects, location descriptions, risk backgrounds, and operation objects in the set of semantic items respectively; when the detection result corresponding to any semantic item is a missing semantic item, recording the corresponding semantic item into the set of missing semantic items; determining whether the set of missing semantic items is empty; when the set of missing semantic items is empty, determining that the corrected text does not have semantic missingness; when the set of missing semantic items is not empty, determining that the corrected text has semantic missingness; when determining that the corrected text has semantic missingness, generating a semantic missingness determination result corresponding to the corrected text based on the semantic item categories contained in the set of missing semantic items.

[0014] Preferably, the step of extracting context text associated with the current intercom round from historical intercom text includes: obtaining channel information, correction text, and semantic term set corresponding to the current intercom round; extracting historical intercom texts from historical intercom text whose transmission time is earlier than the transmission time corresponding to the current intercom round and within a preset time range, based on the transmission time corresponding to the current intercom round, to obtain an initial candidate text set; filtering historical intercom texts with the same channel number corresponding to the current intercom round from the initial candidate text set to obtain a channel candidate text set; extracting association features from each historical intercom text in the channel candidate text set to obtain the historical association features corresponding to each historical intercom text; wherein, historical association features are extracted from historical intercom texts. The associated features include at least one of the following: transmitter identifier, receiver identifier, communication object, location description, risk background, and operation object corresponding to the historical intercom text. The historical associated features corresponding to each historical intercom text are compared with the channel information and semantic item set corresponding to the current intercom round to obtain the associated feature matching results for each historical intercom text. From the associated feature matching results for each historical intercom text, historical intercom texts that meet preset association conditions are extracted as target historical intercom texts, and these target historical intercom texts are arranged in order of transmission time from most recent to oldest. The text content corresponding to each arranged target historical intercom text is determined as the context text corresponding to the current intercom round.

[0015] Preferably, the steps for obtaining the final completed text are as follows: obtaining the set of missing semantic items corresponding to the corrected text, the context text corresponding to the current intercom round, and the set of semantic items already contained in the corrected text; extracting semantic items from the context text to obtain the set of context semantic items corresponding to the context text; extracting context semantic items from the context semantic item set that are consistent with the category of each missing semantic item in the set of missing semantic items to obtain a set of candidate completed semantic items; performing consistency verification between each candidate completed semantic item in the set of candidate completed semantic items and the set of semantic items already contained in the corrected text, and selecting target completed semantic items that are consistent with the semantic content of the current intercom round; inserting each target completed semantic item into the corresponding missing semantic item position in the corrected text, and retaining the original included semantic items in the corrected text unchanged, to obtain the completed text; when there is no target completed semantic item corresponding to any missing semantic item category in the set of candidate completed semantic items, retaining the corresponding missing semantic item position unchanged, and generating completed text based on the remaining target completed semantic items that have been inserted; performing validity verification on the completed text, and recording the completed text that passes the validity verification as the final completed text.

[0016] Preferably, the step of validating the completed text and recording the completed text that passes the validity check as the final completed text comprises: obtaining the completed text and the target completed semantic item corresponding to the completed text; performing a complete semantic item detection on the completed text to obtain the complete semantic item set corresponding to the completed text; determining whether the complete semantic item set contains communication objects, location descriptions, risk backgrounds, and operation objects; when the complete semantic item set contains communication objects, location descriptions, risk backgrounds, and operation objects, determining that the completed text meets the integrity check conditions; performing a consistency check between the target completed semantic item and the original included semantic items in the completed text; when there is no semantic conflict between the target completed semantic item and the original included semantic items in the completed text, determining that the completed text meets the consistency check conditions; when it is determined that the completed text simultaneously meets the integrity check conditions and the consistency check conditions, determining that the completed text passes the validity check, and recording the completed text that passes the validity check as the final completed text; when it is determined that the completed text does not simultaneously meet the integrity check conditions and the consistency check conditions, not recording the completed text as the final completed text.

[0017] The technical effects and advantages of this invention are as follows: The system performs context dependency determination on the current intercom round corresponding to the correction text. When context dependency is determined, semantic missingness determination is performed on the correction text to obtain semantic missingness determination results. When the semantic missingness determination results indicate that the correction text has semantic missingness, the system extracts the context text associated with the current intercom round from the historical intercom text, and completes the missing communication object, location description, risk background or operation object in the correction text based on the context text to obtain the final completed text. This effectively improves the accuracy, continuity and reliability of the maritime multilingual voice data processing process. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, provided as an embodiment of this application. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The multilingual voice data processing method for real-time intercom communication of marine VHF equipment involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention provides a multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, such as... Figure 1 As shown, it includes the following steps: Step 1: Acquire the raw intercom voice data generated by the marine VHF equipment and simultaneously acquire the corresponding channel information, including channel number, transmission time, transmitter identifier, receiver identifier, and sea area location data; construct the voice dataset to be processed based on the raw intercom voice data and the corresponding communication information; Step 2: Preprocess the raw intercom voice data in the voice dataset to be processed to obtain cleaned voice data. The preprocessing includes silence removal, environmental noise suppression, and voice enhancement. Then, perform voice activity detection on the cleaned voice data to divide it into multiple effective voice segments. It should be noted that in this embodiment, the detection of speech activity in the purified speech data and the division into multiple valid speech segments can be achieved using existing speech activity detection methods. Specifically, based on speech energy features, time-frequency features, speech pause features, or endpoint detection results in the purified speech data, speech segments and non-speech segments can be distinguished, and continuous speech segments can be divided into corresponding valid speech segments. The above-mentioned speech activity detection and speech segment division methods are common basic technologies for speech preprocessing in this field, and their specific implementation methods can be selected according to the actual application scenario. This application does not impose specific limitations on them.

[0021] Step 3: Extract speech feature data from each valid speech segment, identify the corresponding target language based on the speech feature data, call the corresponding speech recognition model according to the target language, and perform text conversion processing on each valid speech segment to obtain speech recognition text; It should be noted that in this embodiment, the extraction of speech feature data from each valid speech segment, the identification of the corresponding target language based on the speech feature data, and the invocation of the corresponding speech recognition model to perform text conversion processing on each valid speech segment according to the target language can all be achieved using existing speech recognition processing methods. Specifically, the corresponding language can be identified based on the spectral features, temporal features, speech rate features, pronunciation features, or acoustic representation features in the valid speech segments, and then the speech recognition model corresponding to the recognition result can be invoked to convert the valid speech segments into corresponding speech recognition text. The above-mentioned speech feature extraction, language identification, and text conversion processing methods are common basic speech processing techniques in this field, and their specific implementation methods can be selected according to the actual application scenario. This application does not impose specific limitations on them.

[0022] Step 4: Perform terminology correction processing on the speech recognition text to obtain corrected text; among which, terminology correction processing is used to uniformly correct ship names, heading terms, speed terms, collision avoidance terms, distress terms, and port terms. It should be noted that, in this embodiment, the terminology correction processing of the speech recognition text can be achieved using existing terminology standardization methods. Specifically, based on pre-built ship name terminology, course terminology, speed terminology, collision avoidance terminology, distress terminology, and port terminology databases for maritime VHF real-time intercom communication scenarios, non-standard expressions, synonyms, abbreviations, or expressions with recognition errors appearing in the speech recognition text can be matched and corrected to obtain the corresponding corrected text. The above-mentioned terminology correction processing method belongs to the basic text standardization technology commonly used in this field, and its specific implementation method can be selected according to the actual application scenario; this application does not specifically limit it in this regard.

[0023] Step 5: Analyze the context dependency of the current intercom round corresponding to the correction text. If the correction text is determined to have a context dependency, perform semantic missing judgment on the correction text to obtain the semantic missing judgment result. When the semantic missing determination result indicates that the correction text has semantic missing, the context text associated with the current intercom round is extracted from the historical intercom text, and the missing communication object, location description, risk background or operation object in the correction text is filled in based on the context text to obtain the final filled text. It should be noted that, in this embodiment, the current intercom round refers to one intercom processing unit corresponding to the correction text currently undergoing semantic processing.

[0024] In this embodiment, it should be specifically explained that the step of determining the context dependency of the current intercom round corresponding to the correction text is as follows: The text term sequence contained in the correction text is obtained, and the term type of each text term in the text term sequence is identified to obtain the term type label corresponding to each text term; among which, the term type label includes communication object terms, location description terms, risk description terms, operation object terms, pronoun terms, successor terms, and operation instruction terms, etc. It should be noted that, in this embodiment, the term "substitute" refers to the denotation used in the current intercom round to replace the communication object, location object, or risk object that has appeared in the previous one, such as "your ship," "my ship," "this ship," "the aforementioned location," or "this area," etc.; the term "continuing" refers to the term used to indicate that there is a relationship of continuation, confirmation, supplementation, or correction between the current intercom content and the previous intercom content, such as "continue," "still," "maintain," "in the original manner," or "according to the aforementioned content," etc.; the term "operational instruction" refers to the term used to indicate that there is an action instruction meaning such as heading adjustment, speed adjustment, avoidance execution, berthing and unberthing operation, or communication response in the current intercom round, such as "turn right," "decelerate," "avoid," "maintain heading," or "respond immediately," etc.

[0025] Based on the term type tags corresponding to each text term, a complete semantic term detection is performed on the correction text to obtain the set of semantic terms already contained in the correction text; among which, the set of semantic terms includes communication objects, location descriptions, risk backgrounds, and operation objects; Dependency trigger detection is performed on the pronouns, successor terms, and operation instruction terms in the correction text; when the correction text contains pronouns, successor terms, or operation instruction terms, it is determined that the current intercom round corresponding to the correction text has a context-related trigger condition; when the correction text does not contain pronouns, successor terms, or operation instruction terms, it is determined that the current intercom round corresponding to the correction text does not have a context-related trigger condition. When it is determined that the current intercom round has a context-related triggering condition, it is determined whether the semantic item set is missing at least one of the following: communication object, location description, risk background, or operation object; when the semantic item set is missing at least one, it is determined that the current intercom round corresponding to the correction text has a context dependency; when the semantic item set is not missing the communication object, location description, risk background, and operation object, it is determined that the current intercom round corresponding to the correction text does not have a context dependency caused by semantic missingness. When it is determined that there is no contextual triggering condition for the current intercom round, it is checked whether the semantic item set contains communication object, location description, risk background, and operation object. When the semantic item set contains communication object, location description, risk background, and operation object, it is determined that there is no contextual dependency for the current intercom round corresponding to the correction text. When the semantic item set is missing at least one of communication object, location description, risk background, or operation object, it is determined that the current intercom round corresponding to the correction text is a round to be further dependently verified, and supplementary dependency verification is performed on the round to be further dependently verified.

[0026] In this embodiment, it should be specifically explained that the steps for identifying the term type of each text term in the text term sequence are as follows: The system retrieves each text term from the text term sequence and matches each text term with various term rules in a preset term rule library. The preset term rule library includes rules for communication objects, location descriptions, risk descriptions, operation objects, pronouns, successors, and operation instructions. It should be noted that, in this embodiment, the preset term rule base can be obtained using existing domain terminology construction methods. Specifically, based on common communication object names, location descriptions, risk warnings, alternative expressions, continuation expressions, and operation command expressions in maritime VHF real-time intercom communication scenarios, different types of text terms can be pre-organized, classified, and configured with rules to form a set of rules for term matching and recognition. The aforementioned construction method of the preset term rule base belongs to the basic natural language processing techniques commonly used in this field. For example, it can be implemented using existing terminology dictionary construction methods, keyword classification and maintenance methods, or domain rule configuration methods. This application does not specifically limit it in this regard.

[0027] When any text term matches the communication object term rule, the corresponding text term is marked as a communication object term; the communication object term is used to represent the corresponding ship name, ship call sign, communication subject name or communication target name in the correction text; When any text term matches a location description term rule, the corresponding text term is marked as a location description term; location description terms are used to characterize the corresponding orientation information, distance information, channel location description information, or relative location description information in the correction text; When any text term matches the risk description term rule or the operation object term rule, the corresponding text term will be marked as a risk description term or an operation object term, respectively. Among them, the risk description term is used to characterize the collision risk, encounter risk, interference risk or emergency risk description information in the correction text, and the operation object term is used to characterize the heading adjustment object, speed adjustment object or avoidance execution object in the correction text. When any text term matches the substitution term rule, successor term rule, or operation instruction term rule, the corresponding text term is marked as a substitution term, successor term, or operation instruction term, respectively. Among them, the substitution term is used to represent the alternative reference to the previous communication object or the previous position object, the successor term is used to represent the continuation relationship of the previous intercom content, and the operation instruction term is used to represent the action instruction content in the current intercom round. When any text term matches multiple term rules simultaneously, the priority type of the corresponding text term is determined according to the order of communication object term rules, location description term rules, risk description term rules, operation object term rules, alias term rules, successor term rules, and operation instruction term rules, so as to obtain a unique term type label for each text term.

[0028] In this embodiment, it should be specifically explained that the steps for performing complete semantic item detection on the corrected text are as follows: Obtain the term type tag corresponding to each text term in the correction text, and classify each text term according to the term type tag to obtain the communication object term set, the location description term set, the risk description term set, the operation object term set, the proxy term set, the successor term set, and the operation instruction term set; The semantic extraction of the communication object term set is performed to determine whether the corrected text contains explicit communication object information; when there is at least one communication object term in the communication object term set, the corresponding communication object information is recorded in the communication object semantic term in the semantic term set; Location semantics are extracted from the set of location descriptive terms to determine whether the corrected text contains explicit location descriptive information. When there is at least one location descriptive term in the set of location descriptive terms, the corresponding orientation information, distance information, channel location descriptive information, or relative location descriptive information is recorded as a location descriptive semantic term in the semantic term set. Risk semantic extraction is performed on the risk descriptive term set to determine whether the corrected text contains explicit risk background information; when there is at least one risk descriptive term in the risk descriptive term set, the corresponding collision risk information, encounter risk information, interference risk information or emergency risk information is recorded in the risk background semantic term in the semantic term set; The set of operation object terms and the set of operation instruction terms are correlated to determine whether the correction text contains explicit operation object information. When there is at least one operation object term in the set of operation object terms, the corresponding heading adjustment object, speed adjustment object, or avoidance execution object is recorded as an operation object semantic term in the semantic term set. When there is an operation instruction term in the set of operation instruction terms and no operation object term in the set of operation object terms, the operation object semantic term in the semantic term set is recorded as a missing semantic term. Based on the detection results of semantic items of communication objects, location descriptions, risk backgrounds, and operation objects, a set of semantic items corresponding to the corrected text is generated. Among them, the detected semantic items are recorded as included semantic items in the semantic item set, and the undetected semantic items are recorded as missing semantic items in the semantic item set for subsequent semantic missing determination.

[0029] It should be noted that in this embodiment, object semantic extraction, location semantic extraction, risk semantic extraction, and operation object semantic extraction can all be implemented using existing text information extraction methods. Specifically, based on the term type tags corresponding to each text term in the correction text, existing named entity recognition methods, keyword matching methods, rule parsing methods, semantic slot extraction methods, or field mapping methods can be used to extract, classify, and structure the corresponding communication object information, location description information, risk background information, and operation object information in the correction text. The above-mentioned semantic extraction methods are common basic natural language processing techniques in this field, and their specific underlying implementation methods can be selected according to the actual application scenario; this application does not impose specific limitations on them.

[0030] In this embodiment, it should be specifically explained that the steps for supplementing dependency checks in further dependency check rounds are as follows: Obtain the correction text corresponding to the further dependency verification round and the set of semantic items corresponding to the correction text, and determine the set of missing semantic items corresponding to the further dependency verification round based on the semantic items not included in the set of semantic items; The text terms in the correction text are subjected to abbreviation detection to determine whether the correction text contains abbreviation terms; among them, abbreviation terms include text terms that only represent action content, text terms that only represent state maintenance, text terms that only represent confirmation response, and text terms that only represent local correction. It should be noted that, in this embodiment, the abbreviated expression terms refer to text terms that only express local actions, local states, confirmation meanings, or correction meanings in the current intercom round, without fully providing the corresponding communication object, location description, risk background, or operation object, such as "keep," "continue," "received," "turn right," "decelerate," or "do as instructed."

[0031] When it is determined that the correction text contains abbreviated terms, the correspondence between the abbreviated terms and the set of missing semantic terms is determined; when the semantic reference of the abbreviated terms cannot be independently determined in the current correction text, and the semantic reference corresponds to at least one item in the set of missing semantic terms, it is determined that there is an implicit contextual association condition in the next round of dependent verification. It should be noted that, in this embodiment, the correlation determination between abbreviated expression terms and the set of missing semantic terms can be achieved using existing semantic association analysis methods. Specifically, based on the action meaning, state meaning, confirmation meaning, or correction meaning represented by the abbreviated expression terms, combined with the missing communication objects, location descriptions, risk backgrounds, and operation objects in the set of missing semantic terms, existing keyword matching methods, rule parsing methods, dependency relationship analysis methods, semantic slot mapping methods, or contextual semantic association methods can be used to determine whether the semantic reference of the abbreviated expression terms corresponds to at least one item in the set of missing semantic terms. The above-mentioned correlation determination method belongs to the basic natural language processing technology commonly used in this field, and its specific underlying implementation method can be selected according to the actual application scenario. This application does not specifically limit it in this regard.

[0032] When it is determined that there is an implicit contextual relationship in the verification round to be further dependent on, it is judged whether the abbreviated terms in the correction text need to be combined with the previous intercom content to determine the corresponding communication object, location description, risk background or operation object; when it needs to be combined with the previous intercom content to determine, it is determined that there is a contextual dependency in the verification round to be further dependent on. When it is determined that there are no abbreviated terms in the correction text, or that although there are abbreviated terms, their semantic references can be independently determined in the current correction text, it is determined that there is no contextual dependency in the next round of dependency verification. When it is determined that there is a contextual dependency in the verification round to be further dependent, the corrected text corresponding to the verification round to be further dependent is used as the target text for subsequent contextual text extraction and processing; when it is determined that there is no contextual dependency in the verification round to be further dependent, the corrected text corresponding to the verification round to be further dependent is used as independent parsable text for subsequent semantic parsing processing.

[0033] In this embodiment, it is necessary to specifically explain the steps for determining semantic missing information in the corrected text and obtaining the semantic missing information determination result: Obtain the set of semantic terms corresponding to the corrected text; wherein, the set of semantic terms includes the communication object, location description, risk background, and operation object; Read the detection results corresponding to the communication object semantic item, location description semantic item, risk background semantic item, and operation object semantic item in the semantic item set respectively; When the detection result corresponding to any semantic item is a missing semantic item, the corresponding semantic item is recorded in the missing semantic item set; Determine if the set of missing semantic items is empty; if the set of missing semantic items is empty, determine that the corrected text does not have semantic missing items; if the set of missing semantic items is not empty, determine that the corrected text has semantic missing items. When it is determined that the corrected text has semantic missing information, a semantic missing information determination result is generated based on the semantic item categories contained in the set of missing semantic items.

[0034] In this embodiment, it should be specifically explained that the step of extracting the context text associated with the current intercom round from the historical intercom text is as follows: Obtain the channel information, correction text, and semantic item set corresponding to the current intercom round; Based on the transmission time corresponding to the current intercom round, extract historical intercom texts whose transmission time is earlier than the transmission time corresponding to the current intercom round and are within a preset time range from the historical intercom texts to obtain an initial candidate text set; It should be noted that the preset time range can be obtained by using the time window setting method in the existing technology, and can be adjusted according to the actual communication scenario to limit the time interval for tracing back the historical intercom text from the current intercom round.

[0035] The initial candidate text set is used to select historical intercom texts with the same channel number as the current intercom round to obtain the channel candidate text set; The association features of each historical intercom text in the channel candidate text set are extracted to obtain the historical association features corresponding to each historical intercom text; wherein, the historical association features include at least one of the following: transmitter identifier, receiver identifier, communication object, location description, risk background, and operation object corresponding to the historical intercom text; The historical association features corresponding to each historical intercom text are compared with the channel information and semantic item set corresponding to the current intercom round to obtain the association feature matching results corresponding to each historical intercom text. In the matching results of the associated features corresponding to each historical intercom text, the historical intercom text that meets the preset association conditions is extracted as the target historical intercom text, and each target historical intercom text is arranged in order of transmission time from recent to distant. It should be noted that the preset association conditions can be obtained by using the association matching rule setting method in the existing technology, and can be adjusted according to the actual communication scenario, in order to determine whether there is a contextual relationship between the historical intercom text and the current intercom round.

[0036] The text content corresponding to each target's historical intercom text after being arranged is determined as the context text corresponding to the current intercom round.

[0037] In this embodiment, it should be specifically explained that the final steps for obtaining the completed text are as follows: Obtain the set of missing semantic items corresponding to the correction text, the context text corresponding to the current intercom round, and the set of semantic items already contained in the correction text; Semantic terms are extracted from the context text to obtain a set of context semantic terms corresponding to the context text; Extract contextual semantic items from the contextual semantic item set that are consistent with the category of each missing semantic item in the missing semantic item set to obtain a candidate completion semantic item set; The consistency of each candidate semantic item in the candidate semantic item set with the set of semantic items already contained in the correction text is checked, and the target semantic item that is consistent with the semantic content of the current intercom round is selected. Each target completion semantic item is inserted into the corresponding missing semantic item position in the correction text, while the original included semantic items in the correction text are retained unchanged, thus obtaining the completed text; When there is no target completion semantic item in the candidate completion semantic item set that corresponds to any missing semantic item category, the position of the corresponding missing semantic item is retained unchanged, and the completion text is generated based on the remaining target completion semantic items that have been completed. Perform validity checks on the completed text, and record the completed text that passes the validity check as the final completed text.

[0038] In this embodiment, it is necessary to specifically explain the steps of validating the completed text and recording the completed text that passes the validity check as the final completed text: Retrieve the completed text and the corresponding target completion semantic item; Perform complete semantic item detection on the completed text to obtain the complete semantic item set corresponding to the completed text; Determine whether the set of completed semantic items contains communication objects, location descriptions, risk backgrounds, and operation objects; if the set of completed semantic items contains communication objects, location descriptions, risk backgrounds, and operation objects, determine that the completed text meets the integrity verification conditions; Perform a consistency check between the target completion semantic item and the original included semantic item in the completion text; when there is no semantic conflict between the target completion semantic item and the original included semantic item in the completion text, determine that the completion text meets the consistency check condition; When it is determined that the completed text satisfies both the integrity check condition and the consistency check condition, the completed text is deemed to have passed the validity check, and the completed text that has passed the validity check is recorded as the final completed text. If it is determined that the completed text does not simultaneously meet the integrity check conditions and the consistency check conditions, the completed text will not be recorded as the final completed text.

[0039] Step 6: Perform semantic parsing on the final completed text to extract communication objects, event types, location descriptions, risk descriptions, and operation instruction information, and generate standard semantic data based on the extraction results; determine the corresponding target output language based on the standard semantic data; It should be noted that in this embodiment, semantic parsing of the final completed text, extraction of communication objects, event types, location descriptions, risk descriptions, and operation instruction information, generation of standard semantic data based on the extraction results, and determination of the corresponding target output language based on the standard semantic data can all be achieved using existing semantic analysis processing methods. The specific implementation method can be selected according to the actual application scenario, and this application does not impose any specific limitations on it.

[0040] Step 7: Based on the target output language, perform multilingual conversion processing on the standard semantic data to obtain the target output text and the corresponding target language speech data; send the target output text, target language speech data and standard semantic data to the corresponding maritime VHF equipment and store them together.

[0041] It should be noted that in this embodiment, the multilingual conversion processing of standard semantic data to obtain the target output text and the corresponding target language speech data can be achieved by using existing multilingual conversion processing methods. The specific implementation method can be selected according to the actual application scenario, and this application does not make any specific limitations on it.

[0042] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0043] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for processing multilingual voice data for real-time two-way communication of VHF equipment at sea, characterized in that, Includes the following steps: Step 1: Acquire the raw intercom voice data generated by the marine VHF equipment and simultaneously acquire the corresponding channel information, including channel number, transmission time, transmitter identifier, receiver identifier, and sea area location data; construct the voice dataset to be processed based on the raw intercom voice data and the corresponding communication information; Step 2: Preprocess the raw intercom voice data in the voice dataset to be processed to obtain cleaned voice data; and perform voice activity detection on the cleaned voice data to divide it into multiple valid voice segments; Step 3: Extract speech feature data from each valid speech segment, identify the corresponding target language based on the speech feature data, call the corresponding speech recognition model according to the target language, and perform text conversion processing on each valid speech segment to obtain speech recognition text; Step 4: Perform term correction processing on the speech recognition text to obtain the corrected text; Step 5: Analyze the context dependency of the current intercom round corresponding to the correction text. If the correction text is determined to have a context dependency, perform semantic missing judgment on the correction text to obtain the semantic missing judgment result. When the semantic missing determination result indicates that the correction text has semantic missing, the context text associated with the current intercom round is extracted from the historical intercom text, and the missing communication object, location description, risk background or operation object in the correction text is filled in based on the context text to obtain the final filled text. Step 6: Perform semantic parsing on the final completed text to extract communication objects, event types, location descriptions, risk descriptions, and operation instruction information, and generate standard semantic data based on the extraction results; determine the corresponding target output language based on the standard semantic data; Step 7: Based on the target output language, perform multilingual conversion processing on the standard semantic data to obtain the target output text and the corresponding target language speech data; The target output text, target language voice data, and standard semantic data are sent to the corresponding maritime VHF equipment and stored together.

2. The multilingual voice data processing method for real-time two-way communication of VHF equipment at sea according to claim 1, characterized in that: The step of determining the context dependency of the current intercom round corresponding to the corrected text is as follows: The text term sequence contained in the correction text is obtained, and the term type of each text term in the text term sequence is identified to obtain the term type label corresponding to each text term; among which, the term type label includes communication object terms, location description terms, risk description terms, operation object terms, pronoun terms, successor terms, and operation instruction terms; Based on the term type tags corresponding to each text term, a complete semantic term detection is performed on the correction text to obtain the set of semantic terms already contained in the correction text; among which, the set of semantic terms includes communication objects, location descriptions, risk backgrounds, and operation objects; Dependency trigger detection is performed on the pronouns, successor terms, and operation instruction terms in the correction text; when the correction text contains pronouns, successor terms, or operation instruction terms, it is determined that the current intercom round corresponding to the correction text has a context-related trigger condition; when the correction text does not contain pronouns, successor terms, or operation instruction terms, it is determined that the current intercom round corresponding to the correction text does not have a context-related trigger condition. When it is determined that the current intercom round has a context-related triggering condition, it is determined whether the semantic item set is missing at least one of the following: communication object, location description, risk background, or operation object; when the semantic item set is missing at least one, it is determined that the current intercom round corresponding to the correction text has a context dependency; when the semantic item set is not missing the communication object, location description, risk background, and operation object, it is determined that the current intercom round corresponding to the correction text does not have a context dependency caused by semantic missingness. When it is determined that there is no contextual triggering condition for the current intercom round, it is checked whether the semantic item set contains communication object, location description, risk background, and operation object. When the semantic item set contains communication object, location description, risk background, and operation object, it is determined that there is no contextual dependency for the current intercom round corresponding to the correction text. When the semantic item set is missing at least one of communication object, location description, risk background, or operation object, it is determined that the current intercom round corresponding to the correction text is a round to be further dependently verified, and supplementary dependency verification is performed on the round to be further dependently verified.

3. A multilingual voice data processing method for real-time two-way communication of marine VHF equipment according to claim 2, characterized in that, The steps for identifying the term type of each text term in the text term sequence are as follows: The system retrieves each text term from the text term sequence and matches each text term with various term rules in a preset term rule library. The preset term rule library includes rules for communication objects, location descriptions, risk descriptions, operation objects, pronouns, successors, and operation instructions. When any text term matches the communication object term rule, the corresponding text term is marked as a communication object term; When any text term matches a location descriptor term rule, the corresponding text term is marked as a location descriptor term; When any text term matches a risk description term rule or an operation object term rule, the corresponding text term will be marked as a risk description term or an operation object term, respectively. When any text term matches the substitution term rule, successor term rule, or operation instruction term rule, the corresponding text term will be marked as a substitution term, successor term, or operation instruction term, respectively. When any text term matches multiple term rules simultaneously, the priority type of the corresponding text term is determined according to the order of communication object term rules, location description term rules, risk description term rules, operation object term rules, alias term rules, successor term rules, and operation instruction term rules, so as to obtain a unique term type label for each text term.

4. A multilingual voice data processing method for real-time two-way communication of marine VHF equipment according to claim 2, characterized in that, The steps for performing complete semantic item detection on the corrected text are as follows: Obtain the term type tag corresponding to each text term in the correction text, and classify each text term according to the term type tag to obtain the communication object term set, the location description term set, the risk description term set, the operation object term set, the proxy term set, the successor term set, and the operation instruction term set; The semantic extraction of the communication object term set is performed to determine whether the corrected text contains explicit communication object information; when there is at least one communication object term in the communication object term set, the corresponding communication object information is recorded in the communication object semantic term in the semantic term set; Location semantics are extracted from the set of location descriptive terms to determine whether the corrected text contains explicit location descriptive information. When there is at least one location descriptive term in the set of location descriptive terms, the corresponding orientation information, distance information, channel location descriptive information, or relative location descriptive information is recorded as a location descriptive semantic term in the semantic term set. Risk semantic extraction is performed on the risk descriptive term set to determine whether the corrected text contains explicit risk background information; when there is at least one risk descriptive term in the risk descriptive term set, the corresponding collision risk information, encounter risk information, interference risk information or emergency risk information is recorded in the risk background semantic term in the semantic term set; The set of operation object terms and the set of operation instruction terms are correlated to determine whether the correction text contains explicit operation object information. When there is at least one operation object term in the set of operation object terms, the corresponding heading adjustment object, speed adjustment object, or avoidance execution object is recorded as an operation object semantic term in the semantic term set. When there is an operation instruction term in the set of operation instruction terms and no operation object term in the set of operation object terms, the operation object semantic term in the semantic term set is recorded as a missing semantic term. Based on the detection results of semantic items of communication objects, location descriptions, risk backgrounds, and operation objects, a set of semantic items corresponding to the corrected text is generated. Among them, the detected semantic items are recorded as included semantic items in the semantic item set, and the undetected semantic items are recorded as missing semantic items in the semantic item set for subsequent semantic missing determination.

5. A multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, as described in claim 2, characterized in that: The steps for supplementing dependency checks in subsequent dependency check rounds are as follows: Obtain the correction text corresponding to the further dependency verification round and the set of semantic items corresponding to the correction text, and determine the set of missing semantic items corresponding to the further dependency verification round based on the semantic items not included in the set of semantic items; Abbreviation detection is performed on each text term in the correction text to determine whether the correction text contains abbreviation terms; When it is determined that the correction text contains abbreviated terms, the correspondence between the abbreviated terms and the set of missing semantic terms is determined; when the semantic reference of the abbreviated terms cannot be independently determined in the current correction text, and the semantic reference corresponds to at least one item in the set of missing semantic terms, it is determined that there is an implicit contextual association condition in the next round of dependent verification. When it is determined that there is an implicit contextual relationship in the verification round to be further dependent on, it is judged whether the abbreviated terms in the correction text need to be combined with the previous intercom content to determine the corresponding communication object, location description, risk background or operation object; when it needs to be combined with the previous intercom content to determine, it is determined that there is a contextual dependency in the verification round to be further dependent on. When it is determined that there are no abbreviated terms in the correction text, or that although there are abbreviated terms, their semantic references can be independently determined in the current correction text, it is determined that there is no contextual dependency in the next round of dependency verification. When it is determined that there is a contextual dependency in the verification round to be further dependent, the corrected text corresponding to the verification round to be further dependent is used as the target text for subsequent contextual text extraction and processing; when it is determined that there is no contextual dependency in the verification round to be further dependent, the corrected text corresponding to the verification round to be further dependent is used as independent parsable text for subsequent semantic parsing processing.

6. A multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, as described in claim 1, characterized in that: The steps for determining semantic missing information in the corrected text and obtaining the semantic missing information determination result are as follows: Obtain the set of semantic terms corresponding to the corrected text; wherein, the set of semantic terms includes the communication object, location description, risk background, and operation object; Read the detection results corresponding to the communication object semantic item, location description semantic item, risk background semantic item, and operation object semantic item in the semantic item set respectively; When the detection result corresponding to any semantic item is a missing semantic item, the corresponding semantic item is recorded in the missing semantic item set; Determine if the set of missing semantic items is empty; if the set of missing semantic items is empty, determine that the corrected text does not have semantic missing items; if the set of missing semantic items is not empty, determine that the corrected text has semantic missing items. When it is determined that the corrected text has semantic missing information, a semantic missing information determination result is generated based on the semantic item categories contained in the set of missing semantic items.

7. A multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, as described in claim 1, characterized in that: The step of extracting the context text associated with the current intercom round from the historical intercom text is as follows: Obtain the channel information, correction text, and semantic item set corresponding to the current intercom round; Based on the transmission time corresponding to the current intercom round, extract historical intercom texts whose transmission time is earlier than the transmission time corresponding to the current intercom round and are within a preset time range from the historical intercom texts to obtain an initial candidate text set; The initial candidate text set is used to select historical intercom texts with the same channel number as the current intercom round to obtain the channel candidate text set; The association features of each historical intercom text in the channel candidate text set are extracted to obtain the historical association features corresponding to each historical intercom text; wherein, the historical association features include at least one of the following: transmitter identifier, receiver identifier, communication object, location description, risk background, and operation object corresponding to the historical intercom text; The historical association features corresponding to each historical intercom text are compared with the channel information and semantic item set corresponding to the current intercom round to obtain the association feature matching results corresponding to each historical intercom text. In the matching results of the associated features corresponding to each historical intercom text, the historical intercom text that meets the preset association conditions is extracted as the target historical intercom text, and each target historical intercom text is arranged in order of transmission time from recent to distant. The text content corresponding to each target's historical intercom text after being arranged is determined as the context text corresponding to the current intercom round.

8. A multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, as described in claim 1, characterized in that: The steps for obtaining the final completed text are as follows: Obtain the set of missing semantic items corresponding to the correction text, the context text corresponding to the current intercom round, and the set of semantic items already contained in the correction text; Semantic terms are extracted from the context text to obtain a set of context semantic terms corresponding to the context text; Extract contextual semantic items from the contextual semantic item set that are consistent with the category of each missing semantic item in the missing semantic item set to obtain a candidate completion semantic item set; The consistency of each candidate semantic item in the candidate semantic item set with the set of semantic items already contained in the correction text is checked, and the target semantic item that is consistent with the semantic content of the current intercom round is selected. Each target completion semantic item is inserted into the corresponding missing semantic item position in the correction text, while the original included semantic items in the correction text are retained unchanged, thus obtaining the completed text; When there is no target completion semantic item in the candidate completion semantic item set that corresponds to any missing semantic item category, the position of the corresponding missing semantic item is retained unchanged, and the completion text is generated based on the remaining target completion semantic items that have been completed. Perform validity checks on the completed text, and record the completed text that passes the validity check as the final completed text.

9. A multilingual voice data processing method for real-time two-way communication of VHF equipment at sea, as described in claim 8, characterized in that: The step of validating the completed text and recording the completed text that passes the validity check as the final completed text is as follows: Retrieve the completed text and the corresponding target completion semantic item; Perform complete semantic item detection on the completed text to obtain the complete semantic item set corresponding to the completed text; Determine whether the set of completed semantic items contains communication objects, location descriptions, risk backgrounds, and operation objects; if the set of completed semantic items contains communication objects, location descriptions, risk backgrounds, and operation objects, determine that the completed text meets the integrity verification conditions; Perform a consistency check between the target completion semantic item and the original included semantic item in the completion text; when there is no semantic conflict between the target completion semantic item and the original included semantic item in the completion text, determine that the completion text meets the consistency check condition; When it is determined that the completed text satisfies both the integrity check condition and the consistency check condition, the completed text is deemed to have passed the validity check, and the completed text that has passed the validity check is recorded as the final completed text. If it is determined that the completed text does not simultaneously meet the integrity check conditions and the consistency check conditions, the completed text will not be recorded as the final completed text.

Images