A method and device for transcribing audio and video of a court hearing

Abstract

The present application relates to a kind of court audio and video transcription method and device, including the slice processing of audio and video in court process, and initial text of slice is generated according to the order of slice;The initial text of slice is analyzed according to the specification compliance, and the compliance score of each predetermined specification is obtained;According to the compliance score corresponding to each predetermined specification, select multiple error correction tools from error correction tool set as target error correction tool;The selected target error correction tool is planned and combined, and the error correction tool combination is obtained;Initial text is corrected using error correction tool combination, and the error correction text of slice is obtained;According to the error correction text of each slice, generate court record text.The present application significantly improves the accuracy and specification of court audio and video transcription by intelligent error correction tool selection and combination, and can be widely applied to the automatic generation of court record.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of artificial intelligence technology, and in particular to a method and apparatus for transcribing court hearing audio and video. Background Technology

[0002] In the process of digitization, accurately transcribing the audio and video content of court hearings into written text is of great significance for case archiving, judges' review of case files, legal research, and information disclosure. However, current methods for transcribing audio and video from courtroom settings suffer from problems such as insufficient recognition accuracy, non-standardized court transcript texts, rigid error correction processes, and high reliance on manual intervention. Summary of the Invention

[0003] To address the problems existing in the prior art, embodiments of this specification provide a method and apparatus for transcribing court hearing audio and video. The method involves slicing the audio and video during the court hearing, generating initial text for each slice in sequence, performing a standard compliance analysis on the initial text to obtain a compliance score for each predetermined standard, selecting target error correction tools from a set of error correction tools based on the compliance scores, generating a ranking of the target error correction tools, obtaining a combination of error correction tools, and using this combination of error correction tools to correct errors in the initial text, ultimately obtaining the court hearing transcript text.

[0004] The specific technical solutions of the embodiments in this specification are as follows:

[0005] On the one hand, embodiments of this specification provide a method for transcribing court hearing audio and video, the method comprising:

[0006] The audio and video recordings during the court proceedings are sliced, and the initial text of the slices is generated in the order of the slices.

[0007] Perform a specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification;

[0008] Based on the compliance scores corresponding to each predetermined specification, multiple error correction tools are selected from the error correction tool set as target error correction tools;

[0009] The selected target error correction tools are planned and combined to obtain an error correction tool combination;

[0010] The initial text is corrected using the aforementioned error correction tool combination to obtain the corrected text of the slice;

[0011] The court transcript text is generated based on the error correction text of each slice.

[0012] Performing a specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification further includes:

[0013] For each predetermined specification, intent deconstruction processing is performed to obtain constraint intents. The constraint threshold, constraint weight, and constraint logic of the constraint intents are extracted. A constraint intent graph is constructed with the constraint intents as nodes and constraint dependencies as edges. The constraint intents are the core constraint objectives pointed to by the predetermined specifications. The constraint dependencies include logical or hierarchical dependencies between multiple constraint intents within the same predetermined specification, as well as collaborative dependencies, conflict dependencies, or pre-constraint dependencies between constraint intents of different predetermined specifications. The constraint logic represents the logical relationship between multiple constraint intents contained in the same predetermined specification used to limit the overall compliance with the judgment rules.

[0014] Semantic role labeling and event extraction are performed on the initial text to identify response entities that match the constraint intent graph;

[0015] An intention-response causal graph is constructed based on a structural causal model. The nodes of the intention-response causal graph are the response entities, and the edges are the causal relationships between the response entities and the constraint intentions. The satisfaction degree of a single constraint intention is calculated.

[0016] Based on the satisfaction of each constraint intent and the constraint logic, the compliance score of each predetermined specification is calculated.

[0017] Furthermore, calculating the satisfaction of a single constraint intention further includes:

[0018] Determine the type of constraint threshold corresponding to the constraint intent. The type of constraint threshold includes numerical threshold, interval threshold, text existence threshold, and semantic compliance threshold.

[0019] Based on the type of the constraint threshold, calculate the degree of deviation of the responding entity from the constraint intent;

[0020] The degree of satisfaction of the constraint intention is calculated based on the degree of deviation, the constraint weight corresponding to the constraint intention, and the scene adaptation coefficient corresponding to the scene to which the initial text belongs.

[0021] Furthermore, calculating the degree of deviation of the responding entity from the constraint intent, based on the type of the constraint threshold, further includes:

[0022] If the constraint threshold is a numerical threshold, then the degree of deviation = |numerical value corresponding to the response entity - constraint threshold| / constraint threshold;

[0023] If the type of the constraint threshold is an interval threshold, then the degree of deviation = the absolute value of the interval in which the value corresponding to the response entity exceeds the constraint threshold / the boundary value of the interval in which the value of the constraint threshold exceeds the constraint threshold.

[0024] If the type of the constraint threshold is a text existence threshold, then: when the response entity contains the text corresponding to the constraint threshold, the deviation degree is 0; when the response entity does not contain the text corresponding to the constraint threshold, the deviation degree is 1.

[0025] If the constraint threshold is a semantic compliance threshold, then the semantic similarity value between the response entity and the text corresponding to the constraint threshold is calculated using a pre-trained threshold similarity model, and the deviation degree = 1 - similarity value.

[0026] Furthermore, the constraint logic includes AND, OR, and NOT operations;

[0027] Based on the satisfaction of each constraint intent and the constraint logic, the compliance score for each predetermined specification is calculated, which further includes:

[0028] If the constraint logic is AND, then the compliance score of the predetermined specification is the product of the satisfaction of all constraint intentions under the predetermined specification;

[0029] If the constraint logic is OR, then the compliance score of the predetermined specification is 1 minus the product of the non-compliance scores of all constraint intentions under the predetermined specification, where non-compliance score = 1 - constraint intention satisfaction score;

[0030] If the constraint logic is NOT, then: if there is a prohibited feature in the initial text corresponding to the constraint intent, the satisfaction of the constraint intent is 0; otherwise, the satisfaction is 1. The compliance score of the predetermined specification is equal to the normalized result of the satisfaction of all constraint intents under the predetermined specification.

[0031] Furthermore, based on the compliance scores corresponding to each predetermined specification, multiple error correction tools are selected from the error correction tool set as target error correction tools, including:

[0032] Specifications whose compliance scores are below the predetermined score threshold will be designated as specifications to be corrected.

[0033] Extract the constraint intentions in the specification to be corrected that have a satisfaction level lower than the corresponding satisfaction level threshold, and use them as the constraint intentions to be corrected.

[0034] The error correction tool corresponding to the constraint intent to be corrected is selected from the error correction tool set and used as the target error correction tool.

[0035] Furthermore, the selected target error correction tools are planned and combined to obtain an error correction tool combination that further includes:

[0036] The selected error correction tools are sorted according to their error correction priority to obtain the error correction tool combination.

[0037] Furthermore, the error-correcting text of the slice is further obtained by using the aforementioned error-correction tool combination to correct the initial text.

[0038] The initial text is processed sequentially according to the order of the target error correction tools in the error correction tool set to obtain the error-corrected text of the slice.

[0039] Furthermore, the generation of court transcript text based on the error-corrected text of each slice further includes:

[0040] The error correction text of each slice is concatenated in the order of the slices to obtain the concatenated text;

[0041] The court hearing record text is obtained based on the spliced ​​text.

[0042] On the other hand, embodiments of this specification also provide a transcription device for court hearing audio and video, the device comprising:

[0043] An initial text generation unit is used to slice the audio and video during the court hearing and generate the initial text of the slices in the order of the slices.

[0044] The specification compliance score calculation unit is used to perform specification compliance analysis on the initial text of the slice and obtain a compliance score for each predetermined specification.

[0045] The error correction tool selection unit is used to select multiple error correction tools from the error correction tool set based on the compliance score corresponding to each predetermined specification, and use them as target error correction tools.

[0046] The error correction tool planning and combination unit is used to plan and combine the selected target error correction tools to obtain an error correction tool combination.

[0047] The error correction unit is used to correct the initial text using the error correction tool combination to obtain the error-corrected text of the slice;

[0048] The court transcript text generation unit is used to generate court transcript text based on the error correction text of each segment;

[0049] Performing a specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification further includes:

[0050] For each predetermined specification, intent deconstruction processing is performed to obtain constraint intents. The constraint threshold, constraint weight, and constraint logic of the constraint intents are extracted. A constraint intent graph is constructed with the constraint intents as nodes and constraint dependencies as edges. The constraint intents are the core constraint objectives pointed to by the predetermined specifications. The constraint dependencies include logical or hierarchical dependencies between multiple constraint intents within the same predetermined specification, as well as collaborative dependencies, conflict dependencies, or pre-constraint dependencies between constraint intents of different predetermined specifications. The constraint logic represents the logical relationship between multiple constraint intents contained in the same predetermined specification used to limit the overall compliance with the judgment rules.

[0051] Semantic role labeling and event extraction are performed on the initial text to identify response entities that match the constraint intent graph;

[0052] An intention-response causal graph is constructed based on a structural causal model. The nodes of the intention-response causal graph are the response entities, and the edges are the causal relationships between the response entities and the constraint intentions. The satisfaction degree of a single constraint intention is calculated.

[0053] Based on the satisfaction of each constraint intent and the constraint logic, the compliance score of each predetermined specification is calculated.

[0054] On the other hand, embodiments of this specification also provide a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the above-described method.

[0055] On the other hand, embodiments of this specification also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0056] On the other hand, embodiments of this specification also provide a computer program product, which includes a computer program that, when executed by a processor, implements the above-described method.

[0057] Compared with the prior art, the examples in this specification have the following beneficial effects:

[0058] The beneficial effects of this invention are as follows: Compared with existing technologies, it supplements the shortcomings of pure audio recognition by using video multimodal recognition, and combines the deep understanding capabilities of legal knowledge bases and LLM to significantly improve the transcription accuracy and text standardization of proper nouns and complex sentences; it can diagnose text problems and dynamically select the most suitable processing tools, realizing a leap from a fixed pipeline to intelligent diagnosis and treatment, and has a stronger ability to handle complex errors; through multiple rounds of automated iterative error correction, most problems can be solved within the machine's closed loop, triggering human intervention only in difficult cases, greatly reducing the workload and cost of manual proofreading; it not only handles micro-level segmentation problems, but also ensures the quality of the entire text through the final global polishing LLM, forming a complete quality assurance system of segmentation processing-global optimization; the tool library and scheduling strategy can be continuously optimized with data accumulation, and can be easily integrated with new text processing tools or professional domain knowledge bases. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 The diagram shown is a flowchart illustrating a method for transcribing court hearing audio and video in an embodiment of this specification.

[0061] Figure 2 The diagram shown is a flowchart illustrating the process of performing a specification compliance analysis on the initial text of the slice in an embodiment of this specification to obtain a compliance score for each predetermined specification.

[0062] Figure 3 The diagram shown is a flowchart illustrating the process of calculating the satisfaction of a single constraint intent in an embodiment of this specification.

[0063] Figure 4 The diagram shown is a flowchart illustrating the process of selecting multiple error correction tools from the error correction tool set based on the compliance scores corresponding to each predetermined specification in an embodiment of this specification, which are then used as target error correction tools.

[0064] Figure 5 The diagram shown is a flowchart illustrating the process of generating court transcript text based on the error-corrected text of each slice in an embodiment of this specification.

[0065] Figure 6 The diagram shown is a structural schematic of a court hearing audio and video transcription device according to an embodiment of this specification.

[0066] Figure 7 The diagram shown is a structural schematic of the computer device in an embodiment of this specification.

[0067] [Explanation of Figure Markers]:

[0068] 601. Initial Text Generation Unit;

[0069] 602. Standard compliance scoring calculation unit;

[0070] 603. Error correction tool selection unit;

[0071] 604. Error correction tool planning combination unit;

[0072] 605. Error Correction Unit;

[0073] 702. Computer equipment;

[0074] 704. Processing equipment;

[0075] 706. Storage resources;

[0076] 708. Drive system;

[0077] 710. Input / Output Module;

[0078] 712. Input devices;

[0079] 714. Output devices;

[0080] 716. Presentation equipment;

[0081] 718. Graphical User Interface;

[0082] 720. Network interface;

[0083] 722. Communication link;

[0084] 724. Communication bus. Detailed Implementation

[0085] The technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the embodiments of this specification.

[0086] It should be noted that the terms "first," "second," etc., in the description, claims, and accompanying drawings of the embodiments herein are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0087] It should be noted that the acquisition, storage, use, and processing of data in the technical solutions of the embodiments of this specification all comply with the relevant provisions of national laws and regulations.

[0088] It should be noted that in the embodiments of this specification, certain software, components, models, and other existing solutions in the industry may be mentioned. These should be considered as exemplary, and their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but does not mean that the applicant has already used or necessarily used such a solution. Similarly, the descriptions of prediction algorithms, table organization forms, and hardware structures are also exemplary.

[0089] To address the problems existing in the prior art, embodiments of this specification provide a method and apparatus for transcribing court hearing audio and video. The method involves slicing the audio and video during the court hearing, generating initial text for each slice in sequence, performing a norm compliance analysis on the initial text to obtain a compliance score for each predetermined norm, selecting target error correction tools from a set of error correction tools based on the compliance scores, generating a ranking of the target error correction tools, obtaining a combination of error correction tools, and using this combination of error correction tools to correct errors in the initial text, ultimately obtaining the court hearing transcript text. Figure 1 The diagram illustrates a flowchart of a method for transcribing court hearing audio and video according to an embodiment of this specification. The diagram depicts the process of transcribing court hearing audio and video into a court hearing transcript. The order of steps listed in the embodiment is merely one possible execution order among many and does not represent the only possible order. In actual system or device products, the methods shown in the embodiment or the accompanying drawings can be executed sequentially or in parallel. Specifically, as shown... Figure 1 As shown, the method may include:

[0090] Step 101: Segment the audio and video during the court hearing and generate the initial text of the segments according to the segment order;

[0091] Step 102: Perform a specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification;

[0092] Step 103: Select multiple error correction tools from the error correction tool set based on the compliance scores corresponding to each predetermined specification, and use them as target error correction tools;

[0093] Step 104: Plan and combine the selected target error correction tools to obtain an error correction tool combination;

[0094] Step 105: Use the error correction tool combination to correct the initial text to obtain the corrected text of the slice;

[0095] Step 106: Generate the court transcript text based on the error correction text of each slice.

[0096] In the embodiments of this specification, the type of audio and video during the court hearing is first determined to be either pure audio or video (recording or video). Subsequently, the input audio and video are sliced ​​according to a preset time length or semantic pause point (silence detection, scene switching) to obtain a series of audio and video slice segments.

[0097] If it is an audio slice, a high-precision ASR engine is called to convert speech to text, and the initial text is obtained.

[0098] If it's a video clip, the Omni multimodal recognition engine is invoked. This engine simultaneously processes the speech information from the audio track, converting the speech to text to obtain the initial text. Additionally, it can process visual information from the video track (such as text in documents displayed on the screen, text in evidence images, speaker labels, etc.), using this visual information as a basis for subsequent error correction of the initial text.

[0099] Then, a specification compliance analysis is performed on the initial text of the slice to obtain a compliance score for each predetermined specification. Specifically, in the embodiments of this specification, such as Figure 2 The diagram illustrates the process of performing a specification compliance analysis on the initial text of the slice in an embodiment of this specification to obtain a compliance score for each predetermined specification. The process may include the following steps:

[0100] Step 201: Perform intent deconstruction processing on each of the predetermined specifications to obtain the constraint intent, and extract the constraint threshold, constraint weight and constraint logic of the constraint intent. Construct a constraint intent graph with the constraint intent as nodes and the constraint dependency relationship as edges.

[0101] In this step, the predetermined specifications can be set by staff according to actual business needs, and this embodiment of the specification does not impose any limitations. For example, the predetermined specifications may include: format conformity, grammatical correctness, punctuation accuracy, terminology accuracy, and logical coherence.

[0102] The constraint intent refers to the core constraint purpose pointed to by the predetermined norm. It refers to the single-dimensional substantive constraint conditions (such as at least one of format constraints, content integrity constraints, semantic compliance constraints, and expression accuracy constraints) obtained after semantically deconstructing the predetermined norm and used to limit the text to meet. A predetermined norm can be broken down into one or more constraint intents. For example, the constraint intent of format normativity includes the structure of legal documents, paragraph format, and heading hierarchy; the constraint intent of grammatical correctness includes sentence integrity, subject-verb agreement, and correct tense; the constraint intent of punctuation accuracy includes the special punctuation usage in legal documents and the pairing of quotation marks and brackets; the constraint intent of terminology accuracy includes legal professional terms, institution names, and legal provision citations; and logical coherence includes argumentation logic, the order of factual statements, and causal relationships.

[0103] The constraint dependencies include logical or hierarchical dependencies between multiple constraint intentions within the same predetermined specification, as well as collaborative, conflicting, or pre-constraint dependencies between constraint intentions of different predetermined specifications.

[0104] The constraint logic represents the logical relationship between multiple constraint intentions contained in the same predetermined specification, used to limit the overall compliance determination rule, including but not limited to:

[0105] Conjunction logic (AND): All constraint intentions must be satisfied for the specification to be considered as a whole compliant;

[0106] Disjunction logic (OR): at least one constraint must be satisfied;

[0107] Negation logic (NOT): The text must not contain a certain type of content / feature;

[0108] Hierarchical priority logic: Some intentions are judged before other intentions.

[0109] For example, the predefined specification is that the title must contain "brand" and the number of characters ≤ 20, which is broken down into two constraint intentions: (1) contain the brand field; (2) the number of characters ≤ 20; the constraint logic is conjunction (AND), that is, both constraint intentions must be satisfied at the same time for the specification to be considered compliant.

[0110] The pre-defined contact information can be either a mobile phone number or an email address, which can be broken down into two constraints: (1) fill in the mobile phone number; (2) fill in the email address; the constraint logic is disjunction (OR), that is, either one needs to be satisfied.

[0111] The predetermined specification is that the text must not contain sensitive words. This is broken down into one constraint intention: (1) There are no sensitive words. The constraint logic is negation (NOT). If a sensitive word appears, it does not meet the requirements.

[0112] Step 202: Perform semantic role labeling and event extraction on the initial text to identify response entities that match the constrained intent graph;

[0113] In the embodiments of this specification, the constraint intent refers to the requirements or constraints imposed on the text by the specification (such as "character count ≤ 20" or "must contain a brand"). The response entity refers to the specific content, value, fragment, or feature extracted from the initial text to "respond / match" the constraint intent. The response entity can be extracted from the initial text according to the constraint intent using a Large Language Model (LLM). In essence, the response entity refers to the text fragment, semantic unit, numerical index, feature attribute, or expression result obtained from the initial text through semantic parsing and feature extraction, which corresponds to the constraint intent. It is the objective carrier in the initial text used to determine whether the corresponding constraint intent is met.

[0114] For example, if the constraint intent is that the number of title characters is ≤20, the response entity is the number of title characters calculated from the text (e.g., 25 or 18); if the constraint intent is that the text contains a brand name, the response entity is the specific fragment "XX brand" extracted from the text; if the constraint intent is that it does not contain any non-compliant expressions, the response entity is the feature of any non-compliant word fragments / words detected in the text.

[0115] Step 203: Construct an intention-response causal graph based on the structural causal model. The nodes of the intention-response causal graph are the response entities, and the edges are the causal relationships between the response entities and the constraint intentions. Calculate the satisfaction degree of a single constraint intention.

[0116] In the embodiments of this specification, when constructing an intent-response causal graph within the Structural Causal Model (SCM) framework, the process of a response entity querying a constraint intent graph is the core association process of performing multi-dimensional matching, precise binding, attribute verification, and completion of associated intents between the standardized response entity extracted from the text side and the constraint intent graph pre-constructed on the specification side. Essentially, it allows the unconstrained textual objective carrier (response entity) to find constraint intent nodes in the constraint intent graph that it can directly respond to or indirectly associate with, preliminarily determines the causal relationship type between the two, and finally outputs the precise mapping relationship between the response entity and the constraint intent, providing a unique basis for the subsequent construction of causal edges and node associations in the intent-response causal graph.

[0117] Specifically, firstly, the raw response entities extracted from the initial text are standardized and transformed into a structured feature set that can be directly matched with constraint intent graph nodes, eliminating the ambiguity and diversity of textual descriptions and laying the foundation for subsequent queries. Then, based on the standardized response entity features, potential matching candidate constraint intent nodes are screened in the constraint intent graph, narrowing the scope of precise matching. Simultaneously, the constraint dependency edges of the graph are combined to include upstream and downstream related nodes, avoiding omissions of indirect connections. Next, deep semantic matching is performed between candidate nodes and standardized response entities, excluding semantically similar but inconsistent intent nodes, and determining the constraint intent nodes that are uniquely or precisely bound to the response entity. This is the core step in the query process, determining whether the correspondence between the entity and the intent is accurate. Finally, based on the precisely bound constraint intent nodes… The process involves extracting predefined constraint thresholds, constraint logic, and constraint weights, performing compliance checks on the response entities, and initially determining the causal relationship type between them. This step directly provides the core basis for defining the causal edges of the subsequent intent-response causal graph (the edges of the intent-response causal graph represent the causal relationship between the response entity and the constraint intent). Then, combined with the constraint dependency edges of the constraint intent graph, a second check is performed on the upstream and downstream dependency intents of the precisely bound nodes to complete the binding and causal relationships between the response entities and indirectly related constraint intents. This ensures the completeness of the causal associations in the intent-response causal graph and avoids the loss of causal relationships due to the omission of dependency intents. Finally, the results of all the above steps are integrated into a structured mapping table as the final product of the query process, directly providing the core input for the subsequent construction of the intent-response causal graph.

[0118] In the embodiments of this specification, a structural causal model (SCM) is introduced to construct an intent-response causal graph to quantify the degree to which the text satisfies the specified intent. Nodes in the intent-response causal graph represent response entities in the text (e.g., a missing brand field in a title with 25 characters); edges represent causal relationships (e.g., a character count of 25 violates the character count constraint). The calculation logic for a single constraint intent is: Single intent satisfaction = 1 – (deviation degree × constraint weight × scenario adaptation coefficient). For example, a product title with 25 characters (5 characters over the limit) has a deviation degree of 5 / 20 = 0.25, thus the satisfaction degree can be calculated as 1 - (0.25 × 0.5 × 1.1) = 0.86 (where the constraint weight is 0.5 and the scenario adaptation coefficient is 1.1).

[0119] Step 204: Based on the satisfaction of each constraint intent and the constraint logic, calculate the compliance score of each predetermined specification.

[0120] In the embodiments described in this specification, such as Figure 3 The diagram shown illustrates the process for calculating the satisfaction of a single constraint intent in an embodiment of this specification, which may include the following steps:

[0121] Step 301: Determine the type of constraint threshold corresponding to the constraint intent. The type of constraint threshold includes numerical threshold, interval threshold, text existence threshold, and semantic compliance threshold.

[0122] Step 302: Calculate the degree of deviation of the response entity from the constraint intent based on the type of the constraint threshold;

[0123] In this step, if the type of the constraint threshold is a numerical threshold, then the degree of deviation = |the numerical value corresponding to the response entity - the constraint threshold| / the constraint threshold;

[0124] If the type of the constraint threshold is an interval threshold, then the degree of deviation = the absolute value of the interval in which the value corresponding to the response entity exceeds the constraint threshold / the boundary value of the interval in which the value of the constraint threshold exceeds the constraint threshold.

[0125] If the type of the constraint threshold is a text existence threshold, then: when the response entity contains the text corresponding to the constraint threshold, the deviation degree is 0; when the response entity does not contain the text corresponding to the constraint threshold, the deviation degree is 1.

[0126] If the constraint threshold is a semantic compliance threshold, then the semantic similarity value between the response entity and the text corresponding to the constraint threshold is calculated using a pre-trained threshold similarity model, and the deviation degree = 1 - similarity value.

[0127] Step 303: Calculate the satisfaction degree of the constraint intention based on the degree of deviation, the constraint weight corresponding to the constraint intention, and the scene adaptation coefficient corresponding to the scene to which the initial text belongs.

[0128] In this step, the constraint weights can be preset values, and the scene adaptation coefficients can be set by staff based on experience. The adaptation coefficients for each scene are calculated based on the satisfaction level. This specification does not impose any restrictions on the embodiments.

[0129] In the embodiments described in this specification, constraint logic includes AND, OR, and NOT. AND logic indicates that all constraint intentions must be satisfied simultaneously, OR logic indicates that any constraint intention only needs to be satisfied, and NOT logic indicates that the constraint intention is a prohibitive requirement and the corresponding feature must not appear.

[0130] Furthermore, if the constraint logic is AND, then the compliance score of the predetermined specification is the product of the satisfaction of all constraint intentions under the predetermined specification;

[0131] If the constraint logic is OR, then the compliance score of the predetermined specification is 1 minus the product of the non-compliance scores of all constraint intentions under the predetermined specification, where non-compliance score = 1 - constraint intention satisfaction score;

[0132] If the constraint logic is NOT, then: if there is a prohibited feature in the initial text corresponding to the constraint intent, the satisfaction of the constraint intent is 0; otherwise, the satisfaction is 1. The compliance score of the predetermined specification is equal to the normalized result of the satisfaction of all constraint intents under the predetermined specification, that is: sum the satisfaction of all constraint intents, and then calculate the average satisfaction to obtain the normalized result, which is the compliance score of the predetermined specification.

[0133] This specification's embodiment demonstrates a method for performing norm compliance analysis on the initial text of a slice to obtain a compliance score for each predetermined norm. This method breaks through the traditional technical framework of text similarity matching, upgrading from surface-level text matching to norm intent deconstruction and causal reasoning analysis. It solves the technical pain point of uninterpretable scoring in existing surface-level matching schemes for text-domain normative text. This specification's embodiment achieves interpretable compliance analysis through hierarchical scoring and deviation tracing, and introduces dynamic intent graphs and scene adaptation coefficients, solving industry problems such as difficulty in norm iteration adaptation and poor adaptation to scene differences.

[0134] Then, according to the conformity score corresponding to each predetermined specification, the embodiments of this specification select multiple error correction tools from the error correction tool set as target error correction tools.

[0135] In the embodiments of this specification, the error correction toolset includes, but is not limited to, various dedicated error correction modules such as grammar correction tools, spell checking tools, punctuation correction tools, sentence completion tools, hot word extraction tools, legal knowledge base query tools, prosecution text auxiliary generation tools, and visual information-assisted error correction tools. Among them, the grammar correction tool detects and corrects grammatical errors based on dependency parsing and sequence labeling models (BiLSTM-CRF); the spell checking tool combines edit distance algorithms and legal terminology dictionaries for spell correction; the punctuation correction tool completes and corrects punctuation based on rule engines and BERT punctuation prediction models; the sentence completion tool uses a legal text pre-trained language model to repair sentence integrity; the legal knowledge base query tool connects to a legal knowledge graph, including a legal and regulatory database, a case database, and a proper noun dictionary, to query relevant legal knowledge, which is used as the basis for the large language model to correct the initial text; the prosecution text auxiliary generation tool generates standard format text based on templates and legal document specifications; and the visual information-assisted error correction tool uses text information extracted from video frame OCR to disambiguate homophones.

[0136] In the embodiments of this specification, a compliance score threshold of 70 points can be set. When the compliance score of a predetermined specification is lower than this threshold, the corresponding error correction tool is automatically selected and added to the target error correction tool list. An intelligent matching algorithm ensures that the selected error correction tool accurately corresponds to the identified specification issues.

[0137] Subsequently, the selected target error correction tools are planned and combined to obtain an error correction tool combination. The embodiments in this specification can formulate a reasonable execution order and combination strategy based on the functional characteristics and processing priorities of different error correction tools. Basic error correction tools such as spell checking are executed first, followed by grammar correction, and finally terminology standardization and format adjustment. A coordination mechanism is established between tools to avoid conflicts between different tools, ensuring the stability of the error correction process and the optimization of its effectiveness.

[0138] Next, a combination of error correction tools is used to correct the initial text, resulting in the corrected text of the slice. In the embodiments of this specification, each error correction tool can be called sequentially to process the initial text according to a predetermined execution order. After each error correction tool completes its processing, it passes the result to the next tool, forming a pipeline-like error correction process. Modification records are retained during the error correction process to facilitate subsequent quality checks and audits. After complete error correction processing, a sliced ​​corrected text with significantly improved accuracy and standardization is generated.

[0139] According to one embodiment of this specification, in order to improve the error correction effect, when setting the constraint intent and the error correction tool, the relationship between the two can be considered, thereby pre-setting the constraint intent and the corresponding error correction tool. When selecting the error correction tool, the corresponding error correction tool can be selected according to the constraint intent. Specifically, as follows... Figure 4 The diagram shown is a flowchart illustrating the process of selecting multiple error correction tools from a set of error correction tools based on the compliance scores corresponding to each predetermined specification, as the target error correction tool, in an embodiment of this specification. The process may include the following steps:

[0140] Step 401: Designate specifications whose compliance scores are lower than the predetermined score threshold as specifications to be corrected;

[0141] Step 402: Extract the constraint intentions in the specification to be corrected that have a satisfaction level lower than the corresponding satisfaction level threshold, and use them as the constraint intentions to be corrected;

[0142] Step 403: Select the error correction tool corresponding to the constraint intent to be corrected from the error correction tool set, and use it as the target error correction tool.

[0143] In the embodiments of this specification, the lower the compliance score corresponding to a certain predetermined specification, the less the initial text conforms to the predetermined specification. Therefore, in the embodiments of this specification, the staff can pre-set a predetermined score threshold based on experience. If the compliance score corresponding to a certain predetermined specification of the initial text is lower than the predetermined score threshold corresponding to the predetermined specification, it means that the initial text needs to be corrected according to the predetermined specification.

[0144] In the embodiments of this specification, since the constraint intent of the predetermined specification is set according to the association relationship of the error correction tool, the embodiments of this specification can select constraint intents with a satisfaction level lower than the corresponding satisfaction threshold from the identified constraint intents of the specification to be corrected, as the constraint intents to be corrected. It can be understood that if the satisfaction level of a certain constraint intent is not lower than the satisfaction threshold corresponding to the constraint intent, it means that the response entity corresponding to the constraint intent of the initial text does not need to be corrected; conversely, if the satisfaction level of a certain constraint intent is lower than the satisfaction threshold corresponding to the constraint intent, it means that the response entity corresponding to the constraint intent of the initial text needs to be corrected.

[0145] Then, the error correction tool corresponding to the constraint intent to be corrected is selected from the error correction tool set and used as the target error correction tool.

[0146] Then, all selected target error correction tools are sorted according to their error correction priority to obtain the error correction tool combination.

[0147] In the embodiments of this specification, the error correction priority of the error correction tools can be set by the staff based on experience. For example, the priority of error correction tools for formatting errors > the priority of error correction tools for terminology errors > the priority of error correction tools for grammatical errors > the priority of error correction tools for punctuation errors. The target error correction tools are sorted in descending order of priority to obtain the error correction tool combination.

[0148] Finally, the initial text is processed sequentially according to the order of the target error correction tools in the error correction tool suite to obtain the corrected text of the slice. Specifically, the initial text is first processed by the high-priority target error correction tool to obtain the corrected text, and then the low-priority target error correction tool is used to correct the text corresponding to the high-priority target error correction tool. After all target error correction tools in the error correction tool suite have been executed, the corrected text is obtained.

[0149] It should be noted that the visual information-assisted error correction tool uses text information extracted from video frame OCR to disambiguate homophones. Therefore, in processing video slices, this embodiment of the specification obtains not only the audio information in the video slices but also the visual information of the video frames, such as legal provisions, institution names, and / or personnel names appearing in court hearing videos. The audio information of the video slices will also contain the audio of legal provisions, institution names, and / or personnel names. After processing the audio, an initial text containing legal provisions, institution names, and / or personnel names is obtained. Because different characters may have the same pronunciation, the characters in the transcribed initial text may differ from the characters of legal provisions, institution names, and / or personnel names in the visual information. Therefore, this embodiment of the specification uses a visual information-assisted error correction tool to correct homophones in the initial text. During error correction, the visual information-assisted error correction tool uses the visual information as a reference.

[0150] Furthermore, in the embodiments of this specification, when obtaining the visual information of the slice, the content in the visual information is extracted as the constraint threshold of the corresponding constraint intention. The type of the constraint threshold is a text existence threshold. It is determined whether the corresponding response entity in the initial text contains the constraint threshold, thereby realizing the identification of homonym transcription errors in the initial text.

[0151] According to one embodiment of this specification, such as Figure 5 The diagram shown illustrates the process of generating court transcript text from the error-corrected text of each slice in an embodiment of this specification, which may include the following steps:

[0152] Step 501: Concatenate the error correction text of each slice in the order of each slice to obtain the concatenated text;

[0153] Step 502: Obtain the court hearing record text based on the spliced ​​text.

[0154] During the splicing process in the embodiments of this specification, it is necessary to ensure the semantic coherence and logical consistency between the slices, and to handle any incomplete sentences that may exist at the slice boundaries. A comprehensive check should be performed on the spliced ​​text, such as verifying the accuracy of time stamps, the identification of speakers, and the completeness of court proceedings.

[0155] Furthermore, according to one embodiment of this specification, during the error correction process of a slice, a loop judgment and iteration method can be introduced to improve the accuracy of slice error correction. Specifically, after correcting the initial text of the slice using a combination of error correction tools to obtain the corrected text of the slice, a specification compliance analysis can be performed on the corrected text to obtain a compliance score for each predetermined specification. If the compliance scores of all predetermined specifications meet the standard, the initial text correction of the slice is completed, and the initial text of the next slice is corrected. If there are compliance scores that do not meet the standard, multiple error correction tools can be selected based on the compliance scores of each predetermined specification of the current corrected text and planned and combined to obtain the error correction tool combination for the current corrected text, and the current corrected text is corrected again. In addition, the error correction tool combination ranked after the error correction tool combination used in the previous iteration can also be used to correct the current corrected text again, and this embodiment of the specification does not impose any limitations.

[0156] In the embodiments of this specification, an iteration limit (e.g., 3 times) can be set. If the original text of a certain slice still fails to meet the standard compliance score after a specified number of iterations, manual correction can be performed by staff.

[0157] The method described in this specification supplements the shortcomings of pure audio recognition with video multimodal recognition, and combines the deep understanding capabilities of legal knowledge bases and LLM (Limited Language Management) to significantly improve the transcription accuracy and text standardization of proper nouns and complex sentences. It can diagnose text problems and dynamically select the most suitable processing tools, achieving a leap from a fixed pipeline to intelligent diagnosis and treatment, with a stronger ability to handle complex errors. Through multiple rounds of automated iterative error correction, most problems can be solved within a closed machine loop, triggering human intervention only in difficult cases, greatly reducing the workload and cost of manual proofreading. It not only handles micro-level segmentation problems but also ensures the overall quality of the text through final global polishing LLM, forming a complete quality assurance system of segmentation processing-global optimization. The tool library and scheduling strategies can be continuously optimized with data accumulation and can be easily integrated with new text processing tools or professional domain knowledge bases. Based on the spliced ​​text, a court transcript is obtained, which conforms to the standard format requirements of court records and has good readability and legal validity.

[0158] Based on the same inventive concept, embodiments of this specification also provide a device for transcribing court hearing audio and video, such as... Figure 6 As shown, the device includes:

[0159] The initial text generation unit 601 is used to slice the audio and video during the court hearing and generate the initial text of the slices in the order of the slices.

[0160] The specification compliance score calculation unit 602 is used to perform specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification.

[0161] The error correction tool selection unit 603 is used to select multiple error correction tools from the error correction tool set according to the compliance score corresponding to each predetermined specification, and use them as target error correction tools.

[0162] The error correction tool planning and combination unit 604 is used to plan and combine the selected target error correction tools to obtain an error correction tool combination.

[0163] The error correction unit 605 is used to correct the initial text using the error correction tool combination to obtain the error-corrected text of the slice;

[0164] The court hearing record text generation unit 606 is used to generate court hearing record text based on the error correction text of each slice.

[0165] The beneficial effects obtained by the above-described device are the same as those obtained by the above-described method, and will not be described in detail in the embodiments of this specification.

[0166] like Figure 7 The diagram shown is a structural schematic of a computer device according to an embodiment of this specification. The methods described in this specification can be applied to the computer device of this embodiment.

[0167] like Figure 7 The diagram illustrates the structure of a computer device according to an embodiment of this specification. The computer device in this embodiment is capable of executing the methods described in the embodiments of this specification. The computer device 702 may include one or more processing devices 704, such as one or more central processing units (CPUs), each of which may implement one or more hardware threads. The computer device 702 may also include any storage resource 706 for storing information of any kind, such as code, settings, data, etc. Without limitation, for example, the storage resource 706 may include any type of RAM, any type of ROM, flash memory, hard disk, optical disk, etc. More generally, any storage resource can use any technology to store information. Furthermore, any storage resource may provide volatile or non-volatile retention of information. Furthermore, any storage resource may represent a fixed or removable component of the computer device 702. In one case, when the processing device 704 executes associated instructions stored in any storage resource or combination of storage resources, the computer device 702 may perform any operation of the associated instructions. The computer device 702 also includes one or more drive systems 708 for interacting with any storage resources, such as hard disk drive systems, optical disk drive systems, etc.

[0168] Computer device 702 may also include an input / output module 710 (I / O) for receiving various inputs (via input device 712) and providing various outputs (via output device 714). A specific output mechanism may include a presentation device 716 and an associated graphical user interface (GUI) 718. In other embodiments, the input / output module 710 (I / O), input device 712, and output device 714 may be omitted, and the device may function solely as a computer device within a network. Computer device 702 may also include one or more network interfaces 720 for exchanging data with other devices via one or more communication links 722. One or more communication buses 724 couple the components described above together.

[0169] Communication link 722 can be implemented in any way, such as via a local area network, a wide area network (e.g., the Internet), a point-to-point connection, or any combination thereof. Communication link 722 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc., governed by any protocol or combination of protocols.

[0170] This specification also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0171] This specification also provides computer-readable instructions, wherein when a processor executes the instructions, the program therein causes the processor to perform the above-described method.

[0172] It should be understood that in the various embodiments of this specification, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this specification.

[0173] It should also be understood that, in the embodiments of this specification, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the embodiments of this specification, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0174] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the embodiments in this specification.

[0175] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0176] In the embodiments provided in this specification, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0177] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments described in this specification, depending on actual needs.

[0178] Furthermore, the functional units in the various embodiments of this specification can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0179] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this specification, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this specification. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0180] This specification describes the principles and implementation methods of the embodiments using specific examples. The above descriptions of the embodiments are only for the purpose of helping to understand the methods and core ideas of the embodiments in this specification. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the embodiments in this specification. Therefore, the content of this specification should not be construed as a limitation on the embodiments in this specification.

Claims

1. A method for transcribing court hearing audio and video, characterized in that, The method includes: The audio and video recordings during the court proceedings are sliced, and the initial text of the slices is generated in the order of the slices. Perform a specification compliance analysis on the initial text of the slice to obtain a compliance score for each predetermined specification; Based on the compliance scores corresponding to each predetermined specification, multiple error correction tools are selected from the error correction tool set as target error correction tools; The selected target error correction tools are planned and combined to obtain an error correction tool combination; The initial text is corrected using the aforementioned error correction tool combination to obtain the corrected text of the slice; Generate court transcript text based on the error correction text of each slice; The initial text of the slice is subjected to a specification compliance analysis to obtain a compliance score for each predetermined specification, including: For each predetermined specification, intent deconstruction processing is performed to obtain constraint intents. The constraint threshold, constraint weight, and constraint logic of the constraint intents are extracted. A constraint intent graph is constructed with the constraint intents as nodes and constraint dependencies as edges. The constraint intents are the core constraint objectives pointed to by the predetermined specifications. The constraint dependencies include logical or hierarchical dependencies between multiple constraint intents within the same predetermined specification, as well as collaborative dependencies, conflict dependencies, or pre-constraint dependencies between constraint intents of different predetermined specifications. The constraint logic represents the logical relationship between multiple constraint intents contained in the same predetermined specification used to limit the overall compliance with the judgment rules. Semantic role labeling and event extraction are performed on the initial text to identify response entities that match the constraint intent graph; An intent-response causal graph is constructed based on a structural causal model. The nodes of the intent-response causal graph are the response entities in the initial text, and the edges in the intent-response causal graph represent the causal relationships between the response entities and the constraint intents. The satisfaction of a single constraint intent is calculated. The original response entities extracted from the initial text are standardized and transformed into a structured feature set that matches the nodes of the constraint intent graph. Based on the features of the standardized response entities, matching candidate constraint intent nodes are selected in the constraint intent graph. Semantic matching is performed on the candidate constraint intent nodes and the standardized response entities to determine the constraint intent nodes bound to the standardized response entities. Based on the bound constraint intent nodes, their predefined constraint thresholds, constraint logic, and constraint weights are extracted. The response entities are verified for compliance, and the causal relationship type between them is determined. Combined with the edges of the constraint intent graph, a secondary verification is performed on the upstream and downstream dependent intents of the bound constraint intent nodes to complete the binding relationship and causal relationship between the response entities and the indirectly related constraint intents. Based on the satisfaction of each constraint intent and the constraint logic, the compliance score of each predetermined specification is calculated.

2. The method according to claim 1, characterized in that, Calculating the satisfaction of a single constraint intent includes: Determine the type of constraint threshold corresponding to the constraint intent. The type of constraint threshold includes numerical threshold, interval threshold, text existence threshold, and semantic compliance threshold. Based on the type of the constraint threshold, calculate the degree of deviation of the responding entity from the constraint intent; The degree of satisfaction of the constraint intent is calculated based on the degree of deviation, the constraint weight corresponding to the constraint intent, and the scene adaptation coefficient corresponding to the scene to which the initial text belongs.

3. The method according to claim 2, characterized in that, Calculating the degree of deviation of the response entity from the constraint intent, based on the type of the constraint threshold, includes: If the constraint threshold is a numerical threshold, then the degree of deviation = |numerical value corresponding to the response entity - constraint threshold| / constraint threshold; If the type of the constraint threshold is an interval threshold, then the degree of deviation = the absolute value of the interval in which the value corresponding to the response entity exceeds the constraint threshold / the boundary value of the interval in which the value of the constraint threshold exceeds the constraint threshold. If the type of the constraint threshold is a text existence threshold, then: when the response entity contains the text corresponding to the constraint threshold, the deviation degree is 0; when the response entity does not contain the text corresponding to the constraint threshold, the deviation degree is 1. If the constraint threshold is a semantic compliance threshold, then the semantic similarity value between the response entity and the text corresponding to the constraint threshold is calculated using a pre-trained threshold similarity model, and the deviation degree = 1 - similarity value.

4. The method according to claim 1, characterized in that, The constraint logic includes AND, OR, and NOT; Based on the satisfaction of each constraint intent and the constraint logic, the compliance score for each predetermined specification is calculated, which further includes: If the constraint logic is AND, then the compliance score of the predetermined specification is the product of the satisfaction of all constraint intentions under the predetermined specification; If the constraint logic is OR, then the compliance score of the predetermined specification is 1 minus the product of the non-compliance scores of all constraint intentions under the predetermined specification, where non-compliance score = 1 - constraint intention satisfaction score; If the constraint logic is NOT, then: if there is a prohibited feature in the initial text corresponding to the constraint intent, the satisfaction of the constraint intent is 0; otherwise, the satisfaction is 1. The compliance score of the predetermined specification is equal to the normalized result of the satisfaction of all constraint intents under the predetermined specification.

5. The method according to claim 1, characterized in that, Based on the compliance scores corresponding to each predetermined specification, multiple error correction tools are selected from the error correction tool set as target error correction tools, including: Specifications whose compliance scores are below the predetermined score threshold will be designated as specifications to be corrected. Extract the constraint intentions in the specification to be corrected that have a satisfaction level lower than the corresponding satisfaction level threshold, and use them as the constraint intentions to be corrected. The error correction tool corresponding to the constraint intent to be corrected is selected from the error correction tool set and used as the target error correction tool.

6. The method according to claim 5, characterized in that, The selected target error correction tools are planned and combined to obtain an error correction tool combination including: The selected error correction tools are sorted according to their error correction priority to obtain the error correction tool combination.

7. The method according to claim 6, characterized in that, The initial text is corrected using the aforementioned error correction tool combination to obtain the corrected text of the slice, which includes: The initial text is processed sequentially according to the order of the target error correction tools in the error correction tool set to obtain the error-corrected text of the slice.

8. The method according to claim 1, characterized in that, The court transcript text generated from the error-corrected text of each slice includes: The error correction text of each slice is concatenated in the order of the slices to obtain the concatenated text; The court hearing record text is obtained based on the spliced ​​text.

9. A transcribing device for court hearing audio and video, characterized in that, The device includes: An initial text generation unit is used to slice the audio and video during the court hearing and generate the initial text of the slices in the order of the slices. The specification compliance score calculation unit is used to perform specification compliance analysis on the initial text of the slice and obtain a compliance score for each predetermined specification. The error correction tool selection unit is used to select multiple error correction tools from the error correction tool set based on the compliance score corresponding to each predetermined specification, and use them as target error correction tools. The error correction tool planning and combination unit is used to plan and combine the selected target error correction tools to obtain an error correction tool combination. The error correction unit is used to correct the initial text using the error correction tool combination to obtain the error-corrected text of the slice; The court transcript text generation unit is used to generate court transcript text based on the error correction text of each segment; The initial text of the slice is subjected to a specification compliance analysis to obtain a compliance score for each predetermined specification, including: For each predetermined specification, intent deconstruction processing is performed to obtain constraint intents. The constraint threshold, constraint weight, and constraint logic of the constraint intents are extracted. A constraint intent graph is constructed with the constraint intents as nodes and constraint dependencies as edges. The constraint intents are the core constraint objectives pointed to by the predetermined specifications. The constraint dependencies include logical or hierarchical dependencies between multiple constraint intents within the same predetermined specification, as well as collaborative dependencies, conflict dependencies, or pre-constraint dependencies between constraint intents of different predetermined specifications. The constraint logic represents the logical relationship between multiple constraint intents contained in the same predetermined specification used to limit the overall compliance with the judgment rules. Semantic role labeling and event extraction are performed on the initial text to identify response entities that match the constraint intent graph; An intent-response causal graph is constructed based on a structural causal model. The nodes of the intent-response causal graph are the response entities in the initial text, and the edges in the intent-response causal graph represent the causal relationships between the response entities and the constraint intents. The satisfaction of a single constraint intent is calculated. The original response entities extracted from the initial text are standardized and transformed into a structured feature set that matches the nodes of the constraint intent graph. Based on the features of the standardized response entities, matching candidate constraint intent nodes are selected in the constraint intent graph. Semantic matching is performed on the candidate constraint intent nodes and the standardized response entities to determine the constraint intent nodes bound to the standardized response entities. Based on the bound constraint intent nodes, their predefined constraint thresholds, constraint logic, and constraint weights are extracted. The response entities are verified for compliance, and the causal relationship type between them is determined. Combined with the edges of the constraint intent graph, a secondary verification is performed on the upstream and downstream dependent intents of the bound constraint intent nodes to complete the binding relationship and causal relationship between the response entities and the indirectly related constraint intents. Based on the satisfaction of each constraint intent and the constraint logic, the compliance score of each predetermined specification is calculated.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1 to 8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

12. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images