A unit-oriented remote video mediation system and method

Abstract

The application provides a unit user-oriented remote video mediation system and method, which can support remote video mediation of traffic management departments and multiple participants including unit users, can effectively improve mediation efficiency, shorten the period, and fill the gap of the prior art in the unit user scenario. It carries out business through audio and video visual interaction mode, restores the communication content, evidence process and confirmation of all parties of the mediation scene completely and objectively. The whole process of mediation is recorded and recorded synchronously, and the download function of the video mediation video is provided after the completion of the video mediation, which reduces the probability of secondary disputes, objections and wrangling between the parties after the dispute, and guarantees the compliance and effectiveness of the mediation result.

Description

Technical Field

[0001] This invention relates to the field of computer information processing technology, specifically to a remote video mediation system and method for unit users. Background Technology

[0002] In recent years, with the continuous development of my country's social economy, the number of motor vehicles has been increasing year by year, and traffic activities such as road passenger and freight transport, official business travel, and commercial vehicle operations have become increasingly frequent. Against this backdrop, disputes arising from road traffic accidents and traffic violations, including damage compensation, liability determination, and conflicts, have continued to grow. Many of these cases involve logistics and transportation companies, passenger transport companies, enterprises, and operating platforms. These disputes are no longer limited to simple settlements between ordinary individuals; they often involve multiple parties, including drivers, their employers, insurance companies, and legal departments. Traditional handling methods primarily rely on face-to-face mediation between the parties. However, offline mediation has significant shortcomings: firstly, it is constrained by time and space; disputes involving entities in different locations and with dispersed personnel often lead to high costs and long processing times. Secondly, it suffers from low collaboration efficiency; entity users involve multiple complex processes such as internal approvals, insurance coordination, and legal communication, making it difficult for traditional offline mediation to synchronize the available time slots of all parties. Thirdly, the system for managing and delivering mediation documents has shortcomings, lacking online delivery channels and digital management methods. Therefore, there is an urgent need for a remote video mediation technology for entity users. Summary of the Invention

[0003] To address the problems of high cost, long cycle, and low collaboration efficiency in existing offline mediation methods involving multiple parties, this invention provides a remote video mediation system for corporate users. This system supports remote video mediation between traffic management departments and multiple participants, including corporate users, effectively improving mediation efficiency, shortening the cycle, and filling the gap in existing technologies for corporate user scenarios. This application also discloses a remote video mediation method for corporate users.

[0004] The technical solution of the present invention is as follows: a remote video mediation system for unit users, characterized in that it includes: a video mediation participant information maintenance module, a video mediation meeting information reservation module, a mediation information collection module, a mediation document confirmation module, and a client; The video mediation participant information maintenance module is used to receive participant information maintenance requests submitted by mediators; the video mediation participant information maintenance module sets up a participant role dictionary, which supports custom participant type fields; through the custom participant type fields, mediators can customize the roles of participants; The video mediation meeting information reservation module is used to receive meeting reservation requests submitted by mediators. The meeting reservation request includes meeting information, including: start time, end time, participants, and meeting title. After successful reservation, a unique meeting ID and a temporary access verification code are automatically generated and pushed to all participants via an encrypted SMS channel. The mediation information collection module supports mediators in collecting mediation compensation information during multi-party video conferences, enabling real-time creation of mediation records. The mediation compensation information includes: the basis for mediation, basic facts and losses, compensation items and amounts, compensation liability and proportion, and compensation method and time limit. The mediation information collection module includes: a voice information collection module, a screen information collection module, and a document production module; The voice information acquisition module uses voice recognition technology to translate the conversation between the mediator and the mediator in a video mediation conference in real time; it also allows the mediator to select conversation content to add to the mediation record. The screen information acquisition module supports the mediator's screen sharing function, allowing the mediator to present evidence and related materials online and exchange information with the mediator, replacing the offline direct display of original documents. After information collection is completed, the document creation module allows mediators to select the type of document to generate mediation documents based on the mediation situation, and then generate an electronic link online and push it to the party being mediated. The types of mediation documents include: damage compensation mediation agreement and damage compensation mediation termination agreement. The mediation document confirmation module is used to respond to the mediation party's operation of receiving the mediation document on the client. The mediation document confirmation module receives the mediation document after the mediation party signs and confirms it, generates the mediation document content with signature information and pushes it to the mediator; the mediation document confirmation module stores the mediation document with the mediation party's signature as an official document. The client includes: mediators and those being mediated participating in remote video mediation conferences based on the client.

[0005] Its further features are: The voice information acquisition module includes: an acoustic noise reduction module, a weighted search decoding module, and a traffic accident terminology knowledge base module; The acoustic noise reduction module is used to eliminate ambient noise, audio and video echoes, and other speech interference at the mediation site. The weighted search decoding module receives the spectral feature stream generated sequence from the acoustic noise reduction module and calls the traffic accident terminology knowledge base in real time to perform path weighting and scoring, thereby increasing the priority of candidate paths for traffic accident terminology and systematically improving the accuracy of traffic accident terminology recognition. The traffic accident terminology knowledge base module provides hierarchical professional terminology, weighted score constraints, colloquial synonym mapping, and business field binding data services to the weighted search decoding modular output module. The original mixed speech stream input from the participants' terminals is used as the system's speech input; it is then fed into the acoustic noise reduction module, which performs multi-channel splitting, echo cancellation, and layered frequency domain noise reduction, outputting standard spectral features; the spectral features output by the acoustic noise reduction module are input into the weighted search decoding module, which initiates a real-time call request during the decoding process to read the terminology sequence, matching weights, and synonym mapping rules stored in the traffic accident terminology knowledge base module; the weighted search decoding module scores and filters candidate word paths based on the knowledge base weights, and then outputs standardized traffic accident mediation text for archiving mediation records; The acoustic noise reduction module includes: a multi-channel audio splitting subunit, an echo cancellation subunit, and a layered frequency domain noise reduction subunit; the multi-channel audio splitting subunit segments the multi-channel mixed audio according to the unique identifier of the conference terminal device or the participant ID of the participant, decouples the original audio, and isolates the voice of each participant; The echo cancellation subunit uses a normalized least mean square adaptive filtering algorithm to synchronously pull the audio played at the far end of the device's speaker as a reference signal and dynamically update the filtering weights to cancel the playback echo collected by the microphone. w n+1 =w n +(1 / ||x n+1 || 2 )·e(n)x n ; In the formula, w n Let x be the filter coefficient at time n, e(n) be the error signal, and x be the filter coefficient at time n. n For audio input; ||·|| 2 To find the square of the second norm; The hierarchical frequency domain noise reduction subunit performs the following operations: a1: Receives the speech slices output by the echo cancellation subunit, performs short-time Fourier transform, splits the amplitude spectrum from the original phase, and performs frequency domain segmentation; the segmented frequency domain includes: a low-frequency noise layer of 0-2kHz, a mid-voice layer of 2-4kHz, and a high-frequency technical terminology layer of 4-8kHz. And complete the three-layer frequency domain segmentation according to 0-2kHz, 2-4kHz, and 4-8kHz; a2: Each frequency band independently updates the noise floor in real time using the improved minimum control recursive average IMCRA algorithm, and calculates the a priori and post-affirmative signal-to-noise ratios and the probability of speech presence p(k,l) in layers, where k is the frequency index and l is the time index; a3: Minimum gain G based on the differences in each frequency band minSolve for the optimal frequency domain gain of OM-LSA, correct the amplitude spectrum with gain and preserve the original phase; G(k,l)=[G H_1 (k,l)] p(k,l) ·G min ((1-p(k,l)) ; In the formula, [G H_1 [(k,l)] represents the optimal gain when speech is present, G min This is the minimum gain limit for this frequency band; Different G values ​​are set for the low-frequency noise layer, the mid-frequency human voice layer, and the high-frequency technical terminology layer. min value; a4: Outputs the corrected amplitude spectrum and the original phase; The acoustic noise reduction module ultimately outputs the meeting IDs of the mediator and the mediator, as well as the noise-reduced spectral feature stream, respectively. The weighted search decoding module performs the following operations: b1: The weighted search decoding module has a pre-trained acoustic model based on a convolutional neural network and a pre-trained general language model built in. The weighted search decoding module receives the spectral feature stream output by the acoustic noise reduction module, outputs a temporal phoneme sequence through the acoustic model and sends it into the general speech model. The general language model generates multiple candidate word paths based on phonemes, and the basic score of the general words is uniformly set to 1.0. b2: Traverse the phoneme segments of each path, retrieve the professional terms from the traffic accident terminology knowledge base, and if a match is found, add a special weight of 3.5~5.0 to increase the path score; if no match is found, the score remains unchanged. b3: Compare the total scores after all path corrections, select the optimal word sequence, and output the final text; The traffic accident terminology knowledge base module includes: a terminology collection and entry sub-unit, a hierarchical terminology storage unit, a dual-index hybrid retrieval unit, and a dynamic incremental update unit; The terminology collection and database sub-unit extracts standard professional terms from multiple terminology sources to form a raw terminology material pool; the terminology sources include: official legal provisions, standard mediation documents, road coding specifications, and operation and management regulations. The hierarchical terminology storage unit is equipped with a professional terminology database, including: an insurance compensation core sub-database, a unit user main terminology sub-database, a road traffic accident liability sub-database, and a spoken language mapping sub-database; The core insurance claims sub-database is used to store core claims terms and is used to prioritize the identification of key terms related to claims. The unit user subject terminology sub-database is used to store enterprise-specific mediation terminology; The road traffic accident liability sub-database is used for liability determination terminology; The colloquial mapping sub-library is used to store the one-to-one correspondence between colloquial terms and standard professional terms; The hierarchical term storage unit hierarchizes terms according to their business importance and configures differentiated weights; The dual-index hybrid retrieval unit constructs two independent indexes to adapt to the two types of tasks: real-time speech decoding and text standardization. The first is a phoneme prefix index, which performs fuzzy prefix matching on real-time phoneme segments during the decoding process, matches professional terms, and adds weights. The second is a text hash index, which calls the content in the spoken language mapping sub-library to replace spoken expressions with official standard terms in batches. The dynamic incremental update unit updates and iterates the content in the hierarchical terminology storage unit of the knowledge base; it extracts professional terms from documents and records using NLP algorithms and outputs them to the hierarchical terminology storage unit for storage according to terminology type. The clients include: a web page client for mediators and a mobile app for the parties being mediated; The mediation information collection module also includes the following operations: Before the participants' clients officially enter the video mediation meeting, the system will forcibly intercept them to confirm whether each client has enabled camera and microphone permissions. If not, the client will be prompted to enable permissions via message. If permissions are enabled, the meeting will proceed to the next stage. The mediation information collection module allows mediators to use a web client to read the camera and microphone permission status via the navigator.permissions.query method; if the returned result is denied or prompt, it indicates that the permissions are not enabled. The mediation information collection module will prompt that permission is not enabled if the ContextCompat.checkSelfPermission method returned by the mediator using a mobile app is not PackageManager.PERMISSION_GRANTED. The mediation document confirmation module specifically includes the following operations: c1: Build an iReport hierarchical template library based on the classification of unit users. The iReport hierarchical templates include various types such as ordinary party templates, corporate templates, and traffic accident templates. c2: Preset user-specific field fill positions and fixed layer area for electronic signature information in the template; c3: After the mediator confirms that the signatures of all parties and the documents are correct, the data is transmitted to the intranet in an encrypted manner; The electronic signature layer and document content are encrypted using the SM4 algorithm in fragmentation. c4: Design an algorithm for converting data between internal and external networks, separating and storing electronic signature information in JSON format from external networks and other text information, and automatically translating it into report parameters that iReport can recognize; c5: Based on the document type identifier carried by the internal and external network data conversion algorithm, retrieve the corresponding iReport template. The template has two preset independent layer areas: the main text rendering layer and the electronic signature layer reserved layer. c6: Pre-render the main text; Calculate the boundary of the area occupied by the complete text content in memory, and obtain the set of coordinates of the remaining blank area of ​​the page. The region boundary includes: page boundary coordinates, total occupied height, and page break position; c7: Read the layer parameters of the signature image calculated in the internal and external network data conversion algorithm, combine them with the data of the area boundary, traverse the blank area of ​​the page, and select the optimal coordinates for placing the signature image; The layer parameters include: pixel width and height; c8: Dynamic layer rendering and document compositing; Based on the coordinate information of the optimal coordinates, the independently encapsulated electronic signature layer is rendered to the target position, and the main text rendering layer and the reserved electronic signature layer are merged to generate a complete PDF of the mediation agreement; c9: Send the mediation document with signature information to the mediator; c10: The mediation document confirmation module stores mediation documents with the signature of the mediated party as official documents; During storage, SM4 encryption is used, and the national standard SM3 hash operation is called synchronously to generate fixed-length ciphertext. The original SM3 ciphertext is stored in the database to determine whether the document has been tampered with. For mediators of different user types, each is assigned their own SM2 private key to digitally sign the original SM3 ciphertext stored in the database. The internal / external network data conversion algorithm includes the following operations: d1: Receive complete JSON data transmitted over the internet based on a predefined entity bean; d2: Split the JSON data into two independent data streams; The independent data streams include: text business data stream and electronic signature data stream; The text business data stream includes structured data such as: mediation basis, basic facts and losses, compensation items and amounts, compensation liability and proportion, compensation methods and time limits; The electronic signature data stream includes: a Base64 string that is uniformly serialized during transmission of the original binary image of the electronic signature; d3: Decode the Base64 string to restore the original byte array of the digital signature; simultaneously parse the header image information of the byte array to obtain the pixel width and height of the digital signature image, and generate iReport layer parameters for subsequent PDF rendering; The mediation document confirmation module also includes the following operations: e1: A separate business queue is set up for large-volume unstructured data; the large-volume unstructured data includes: the electronic signature information of the mediated party and mediation documents; e2: An asynchronous message queue transmission unit is installed in the middle service layer, and the push of large-volume unstructured data messages is completed through the queue for cross-network exchange; e3: Configure an automatic retry policy for each cross-network exchange message: When queue data consumption fails, re-push according to the preset time interval and number of pushes (fre); if consumption still fails after fre retries, write it to the system error log, and manually resend the electronic signature information and mediation document message according to the log.

[0006] A remote video mediation method for unit users, characterized by comprising the following steps: S1: Dictionary of personnel roles for mediators; The personnel role dictionary includes: a custom participant type field; the custom participant type field allows mediators to customize the roles of participants. S2: The mediator submits a meeting reservation request for a remote video mediation meeting and controls the meeting information of the video mediation meeting; S3: Based on the meeting information, complete the reservation for the video mediation meeting. After successful reservation, a unique meeting ID and temporary access verification code are automatically generated and pushed to all participants via an encrypted SMS channel. S4: Before the meeting start time arrives, based on the meeting information, the mediators and mediators enter the video mediation meeting; S5: During video mediation meetings, mediators collect information on mediation and compensation, enabling real-time creation of mediation records. S6: After completing the collection of mediation information, the mediator selects the type of document to be prepared according to the mediation situation, generates the mediation document from the transcript, and sends the electronic link of the mediation document to the mediator online. If a settlement is reached through mediation, a damages settlement agreement shall be prepared. If no agreement can be reached, a settlement agreement for damages shall be prepared. S7: The respondent receives the mediation document on the mobile app, and after confirming that the content of the mediation document is correct and signing it, generates the mediation document content with the signature information and pushes it to the mediator. S8: The mediator receives the mediation document content with signature information on the web terminal, and after confirming that the content of the mediation document and the signature information are correct, generates a mediation document with the signature information of the mediated party and stores it as an official document; S9: The entire mediation process is recorded on video and audio simultaneously, and a complete record is kept. After the video mediation is completed, the function of downloading the video mediation video is provided.

[0007] This application provides a remote video mediation system and method for organizational users. Through a video mediation participant information maintenance module, it supports mediators in customizing multiple participant roles and manages participation permissions based on participant type classification. This system is suitable for the business models of enterprises, transportation companies, and insurance institutions handling dispute mediation on behalf of others, overcoming the limitations of existing remote mediation technologies that are only designed for individual parties and do not support organizational users participating on behalf of others. The system conducts business through audio-visual visual interaction, completely and objectively reproducing the communication content, evidence presentation process, and confirmation status of all parties at the mediation site. Based on the mediation information collection module, the entire mediation process is simultaneously recorded and archived. After the video mediation is completed, a download function for the video mediation video is provided, reducing the probability of secondary disputes and disagreements between the disputing parties and further ensuring the compliance and effectiveness of the mediation result. This method collects participant information through a video mediation participant information maintenance module and mediation compensation information through a video mediation meeting appointment module. Mediators with video service capabilities establish a communication link with respondents using a mobile app, while mediators with video service capabilities use a web client. A mediation information collection module records compensation information in real time during the mediation process. Mediation documents are pushed to the respondents immediately via transmission technology. A mediation document confirmation module ensures the authenticity and validity of the mediation documents. This method can effectively improve mediation efficiency, shorten the cycle, and fill the gap in existing technology for unit user scenarios. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the modules of the remote video mediation system for unit users in this application; Figure 2 This is a structural diagram of the voice information acquisition module. Figure 3 This is a schematic diagram of the process for generating a mediation agreement. Detailed Implementation

[0009] like Figure 1 As shown, this application includes a remote video mediation system for unit users. The system is based on a microservice distributed architecture and is engineered. The architecture is divided into a three-level layered topology: application layer, middleware layer, and persistent storage layer.

[0010] The application layer carries all core functional units for system business interactions, comprising six modular components: a video mediation participant information maintenance module, a video mediation meeting information reservation module, a mediation information collection module, a mediation document confirmation module, and a client. The client includes modules for mediators and respondents to participate in remote video mediation meetings. In this embodiment, the client includes a web page client for mediators and a mobile app for respondents. As the entry point for system human-computer interaction and business process scheduling, the application layer undertakes the entire business logic of remote video mediation, including process driving, information collection, audio-visual interaction, document generation, and electronic signing.

[0011] The middle service layer is the core supporting component for the system's distributed scheduling. It provides the underlying foundational capabilities for standardized data interaction, cross-service scheduling, and secure access control, serving as a crucial support layer for enabling collaborative operation of multi-terminal heterogeneous clients and multiple business modules. The middle service layer encapsulates a standardized underlying general-purpose interface service cluster, encompassing a unified front-end and back-end interaction interface management unit, a cross-microservice call interface encapsulation unit, and a multi-dimensional business data retrieval and query interface unit. It leverages standardized interfaces to achieve bidirectional data flow across modules, terminals, and services. Simultaneously, it integrates an asynchronous message queue data transmission unit, a real-time data channel unit, and a case business process logic control unit, adaptable to differentiated data access, read / write, and interactive operation requirements in various business scenarios. From a security perspective, the middle service layer embeds a unified identity authentication and multi-factor authentication logic unit, completing access subject identity verification, operation permission isolation, and continuous maintenance of the session lifecycle, ensuring system stability and traceability of access behavior under high-concurrency scenarios.

[0012] The persistent storage layer is the system's distributed data persistence support component. It adopts a hybrid multi-level storage architecture, integrating three storage media—Oracle database, HBase database, and Redis distributed cache—to collaboratively achieve layered data persistence and high-speed read / write. Specifically, the Oracle database handles the persistent storage of structured business data, relying on transaction isolation, primary key constraints, and integrity constraints to ensure the consistency, integrity, and verifiability of structured data such as participant information, meeting appointment records, mediation transcripts, and official documents. The HBase database provides large-capacity persistent storage support for unstructured and semi-structured multimedia data, capable of storing multimedia binary resources such as on-site evidence images, full-process mediation audio and video recordings, and electronic signatures generated during remote video mediation, adapting to the storage and retrieval needs of multi-dimensional, large-volume, complex, and heterogeneous data. The Redis distributed cache component provides global high-speed memory read / write caching capabilities, caching hot data such as meeting online status, temporary access credentials, permission verification information, and frequently queried business data, significantly reducing the underlying database IO pressure and optimizing the system's end-to-end interface response latency and concurrency performance.

[0013] The video mediation participant information maintenance module receives participant information maintenance requests submitted by mediators. The backend of this module is independently developed based on Spring Boot microservices. Participant information includes: the participant's relationship to the parties involved, participant type, identity information, contact information, and organization information. The module's database has a built-in customizable and extensible participant role dictionary. This dictionary supports custom participant type fields, allowing mediators to configure participant roles. Participant roles include: parties involved, insurance company specialists, corporate legal counsel, organization representatives, lawyers, guardians, and witnesses.

[0014] The relationships between participants and parties supported in this application include family members such as parents, spouses, and siblings, as well as individuals who can act as agents in mediation, such as lawyers and friends. The types of participants include insurance company specialists, legal representatives, and institutional agents. Mediators can achieve a customized control design for organizational users by customizing the participant type field, thereby adapting to the needs of organizational users to handle mediation on behalf of others.

[0015] The video mediation meeting information reservation module receives meeting reservation requests submitted by mediators. These requests include meeting information such as start time, end time, video type, participants, and meeting title. Upon successful reservation, a unique meeting ID and temporary access verification code are automatically generated and sent to all participants via encrypted SMS. Two or more mediators can join the meeting simultaneously. The module uses Redis to cache real-time meeting status and participant online information, supporting concurrent entry by multiple participants. Participant information is validated using specified storage logic after being saved.

[0016] Before mediators can join a video mediation conference via a web client with video service capabilities, they must also complete the following verification: The meeting information is validated based on the first validation condition, which includes: the meeting start time is later than the current time, the meeting end time is later than the meeting start time, and the meeting name and participants are not empty. If the first verification condition is met, the mediator enters the video mediation conference through a web client with video service capabilities; otherwise, the first warning message is output.

[0017] Before the respondent joins a video mediation meeting via a mobile app, the following verifications are also required: Verify that the microphone and camera permissions of the mobile device are enabled, and ensure that the voice and video access permissions of the connected device are enabled. In response to receiving confirmation from the respondent regarding the notification of rights and obligations and confirming that the microphone and camera permissions on the mobile device are enabled, the respondent can enter the video mediation meeting via the mobile app; otherwise, a second warning message will be output.

[0018] Based on the basic information from the meeting, the person being mediated is able to join the video mediation meeting via a mobile app and cooperate with the mediator in conducting the mediation work, including: Access restrictions are implemented using identity verification technology; the person being mediated must undergo identity verification, such as via SMS verification code or identity check, to restrict access. For organizational users, a designated agent participates in video mediation through the participant information maintenance module; audio and video recordings are simultaneously stored during the mediation process to ensure the fairness and legality of the video mediation work.

[0019] For special parties, including minors and deaf-mute individuals, authorized relatives, witnesses, or agents are allowed to participate via video link, and their presence is recorded throughout the process to prove the legality of the procedure and reduce subsequent disputes.

[0020] This application relies on an identity verification mechanism to reduce the risks and hidden dangers caused by non-compliant operations; through SMS verification code verification and identity verification technology, it restricts unauthorized personnel from entering the video mediation meeting, so as to ensure that the entire mediation process is based on evidence and the steps are controllable; and the mediation video is encrypted and stored to achieve the isolation and protection of privacy data.

[0021] When the person being mediated needs to upload a mediation application, it is also necessary to verify whether the mobile app has the permission to use voice and video. If the permission is not enabled, a warning message will be displayed.

[0022] In this embodiment, the verification of whether the mediator and the mediator's clients have enabled voice and video access permissions is implemented based on the mediation information collection module, specifically including the following operations: Before the participants' clients officially enter the video mediation meeting, the system will forcibly intercept them to confirm whether each client has enabled camera and microphone permissions. If not, the client will be prompted to enable permissions via message. If permissions are enabled, the meeting will proceed to the next stage. The mediation information collection module allows mediators to use a web client to read the camera and microphone permission status via the navigator.permissions.query method; if the returned result is denied or prompt, it indicates that the permissions are not enabled. The mediation information collection module detects that if the respondent uses the ContextCompat.checkSelfPermission method on their mobile app, and the returned result is not PackageManager.PERMISSION_GRANTED, it indicates that the permission is not enabled.

[0023] The mediation information collection module supports mediators in collecting mediation and compensation information during multi-party video conferences, enabling real-time creation of mediation records. The mediation and compensation information includes: the basis for mediation, basic facts and losses, compensation items and amounts, compensation liability and proportion, and compensation methods and time limits.

[0024] By mandating verification of whether the client has enabled voice and video access permissions, problems such as black screens and inability to hear sound can be effectively avoided after participants enter a video mediation meeting, thus improving meeting efficiency.

[0025] The mediation information collection module includes: a voice information collection module, a screen information collection module, and a document creation module.

[0026] Specifically, the voice information acquisition module is based on speech recognition technology (ASR), including model noise reduction and sound source enhancement. Through relevant acoustic models (such as convolutional neural networks and temporal networks) and real-time decoding algorithms, combined with a traffic accident terminology knowledge base, it can translate the conversation between mediators and those being mediated in video mediation conferences in real time, effectively improving the accuracy of speech-to-text conversion; and it allows mediators to select conversation content to add to the mediation record.

[0027] The screen information collection module supports the mediator's screen sharing function, allowing the mediator to present evidence and related materials online, enabling information exchange with the mediator and replacing the offline direct display of original documents.

[0028] After information collection is completed, the document preparation module allows mediators to select the type of document to generate mediation documents based on the mediation situation. After generating an electronic link online, the document is pushed to the party being mediated, ensuring the authenticity and validity of the mediation documents. The types of mediation documents include: damage compensation mediation agreement and damage compensation mediation termination agreement.

[0029] During the mediation process, the mediation information collection module collects mediation compensation information. Based on this information, if a settlement is reached, a damage compensation mediation agreement is prepared and sent out; if no settlement is reached, a damage compensation mediation termination agreement is prepared and sent out, including: Mediators with video service capabilities establish communication with respondents with video service capabilities via a mobile app, and use voice recognition to translate the question-and-answer dialogue between the mediators and respondents into written records in real time. The real-time translated mediation content is screened and confirmed, and then filled into the mediation collection items to generate mediation documents; The mediation content is the core part of the mediation record. The mediator conducts mediation with the parties involved based on the nature and specific circumstances of the accident. The mediation content includes the relationship between the parties involved and the accident, the basic facts of the accident and the losses, the items and amounts of compensation, the liability and proportion of compensation, the method and time limit of compensation, and information on the relevant personnel involved.

[0030] The mediation document confirmation module is used to respond to the mediation party's operation of receiving mediation documents on the client side. It provides the mediation party with the function of previewing and signing the mediation documents. After confirmation, the signature information is pushed to the mediator for review. The mediation document confirmation module receives the mediation documents after the mediation party has signed and confirmed them, generates the mediation document content with the signature information and pushes it to the mediator. The mediation document confirmation module stores the mediation documents with the mediation party's signature as official documents.

[0031] Before a mediation document generated based on mediation compensation information is sent out, the mediation compensation information also needs to be verified according to the second verification condition: The second verification condition includes that the mediation start time is later than the current time and the mediation end time is later than the mediation start time. If the mediation result is an agreement, the mediation basis, basic facts and losses, compensation items and amounts, compensation liability and proportion, and compensation method and time limit are mandatory fields and cannot be empty. If the second verification condition is met, a mediation document is generated and pushed based on the mediation compensation information; otherwise, a third warning message is output.

[0032] Before generating a mediation document with the signatures of the parties involved, it is necessary to verify whether all parties involved have completed their signatures. If the verification passes, a mediation document with the signatures of the parties involved will be generated; otherwise, a message will be displayed indicating that the parties involved have not yet completed their signatures.

[0033] Before the scheduled start time of the meeting, the mediator enters the video mediation conference via a web client capable of running video services to conduct mediation work with the respondent. The respondent enters the video mediation conference via a mobile app capable of running video services to cooperate with the mediator in conducting the mediation work.

[0034] In this embodiment, the mobile app with video service capabilities is developed natively for both Android and iOS, has a built-in handwritten electronic signature component, and is compatible with multiple network environments including 4G, 5G, and WiFi.

[0035] Mobile devices include mobile phones, tablets, or dedicated mobile devices, running as an app or accessed via a webpage. The identity of the person being mediated is verified before signing to prevent unauthorized signatures and ensure the authenticity of the mediation.

[0036] After the mediator and the person being mediated enter the video mediation conference through the client, they establish communication and use voice recognition technology to translate the dialogue into a mediation record in real time. The mediator then filters and confirms the real-time translated mediation record and fills it into the information items of the mediation compensation information collection module to generate detailed mediation record content. During the video mediation, the mediator can share their web client with the person being mediated to display mediation evidence, agreements, and other materials, replacing the need to present the original documents offline.

[0037] During the mediation process, the mediation information collection module gathers mediation compensation information. After mediation, based on the compensation information, it selects to create a damage compensation mediation agreement or a mediation termination agreement and pushes it to the mediator. The respondent receives the mediation document on a mobile app, confirms the content is correct, signs it, and sends it to the mediator. A web client with video service capabilities receives the signed mediation document, confirms the content and signature, and generates a mediation document with the respondent's signature based on the mediation document confirmation module. The mediation document with the respondent's signature is a standard document that can be downloaded, printed, and viewed. This method is suitable for the business model of unit users handling dispute mediation, improving the convenience and efficiency of mediation work and saving manpower and time costs.

[0038] like Figure 2 As shown, the voice information acquisition module includes: an acoustic noise reduction module, a weighted search decoding module, and a traffic accident terminology knowledge base module. This application establishes a voice translation link composed of the acoustic noise reduction preprocessing module, the weighted search decoding module, and the traffic accident terminology knowledge base module. Relying on multi-level technical means, it systematically improves the accuracy of traffic accident terminology recognition, automatically outputs standardized and compliant mediation texts tailored to traffic management operations, and achieves automated conversion of voice dialogue into standardized mediation records.

[0039] The acoustic noise reduction module is located at the very beginning of the entire link and is used to eliminate and mediate ambient noise, audio and video echoes and other voice interference. Weighted search decoding module: Receives the spectral feature stream output by the acoustic noise reduction module to generate a sequence, and calls the traffic accident terminology knowledge base in real time to perform path weighting and scoring, thereby increasing the priority of candidate paths for traffic accident terminology and systematically improving the accuracy of traffic accident terminology recognition. Traffic accident terminology knowledge base module: Provides services such as hierarchical professional terminology, weighted score constraints, colloquial synonym mapping, and business field binding data for the weighted search decoding modular output module.

[0040] The original mixed speech stream input from the participants' terminals is used as the system's speech input; it is then fed into the acoustic noise reduction module, which performs multi-channel splitting, echo cancellation, and layered frequency domain noise reduction, outputting standard spectral features; the spectral features output by the acoustic noise reduction module are then input into the weighted search decoding module, which initiates a real-time call request during the decoding process to read the terminology sequence, matching weights, and synonym mapping rules stored in the traffic accident terminology knowledge base module; the weighted search decoding module scores and filters the candidate paths of words based on the knowledge base weights, and then outputs standardized traffic accident mediation text for archiving of mediation records.

[0041] The acoustic noise reduction module includes a multi-channel audio splitting subunit, an echo cancellation subunit, and a layered frequency domain noise reduction subunit. The multi-channel audio splitting subunit segments the multi-channel mixed audio based on the unique identifier of the conference terminal device or the participant ID of the participant, decouples the original audio, isolates the voices of the mediator and the parties being mediated, and avoids the interference of human voices on terminology recognition.

[0042] The echo cancellation subunit uses the Normalized Least Mean Square (NLMS) adaptive filtering algorithm to synchronously pull the audio played from the far end of the device's speaker as a reference signal, dynamically update the filtering weights to cancel the playback echo collected by the microphone, and perform weight updates to minimize the error. w n+1 =w n +(1 / ||x n+1 || 2 )·e(n) ·x n ; In the formula, w n Let x be the filter coefficient at time n, e(n) be the error signal, and x be the filter coefficient at time n. n For audio input; ||·|| 2 To find the square of the second norm.

[0043] In the layered frequency domain noise reduction subunit, the following operations are performed: a1: Receives the speech slices output from the echo cancellation subunit, performs short-time Fourier transform (STFT), splits the amplitude spectrum from the original phase, and performs frequency domain segmentation; the segmented frequency domain includes: a low-frequency noise layer of 0-2kHz, a mid-voice layer of 2-4kHz, and a high-frequency technical terminology layer of 4-8kHz. And complete the three-layer frequency domain segmentation according to 0-2kHz, 2-4kHz, and 4-8kHz; a2: Each frequency band independently updates the noise floor in real time using the improved minimum control recursive average IMCRA algorithm, and calculates the a priori and post-affirmative signal-to-noise ratios and the probability of speech presence p(k,l) in layers, where k is the frequency index and l is the time index; a3: Minimum gain G based on the differences in each frequency band minSolve for the optimal frequency domain gain of OM-LSA (Optimally-Modified Log-Spectral Amplitude), correct the amplitude spectrum with gain and preserve the original phase; OM-LSA single-layer gain formula: G(k,l)=[G H_1 (k,l)] p(k,l) ·G min ((1-p(k,l)) ; In the formula, [G H_1 [(k,l)] represents the optimal gain when speech is present, G min This is the minimum gain limit for this frequency band; Different G values ​​are set for the low-frequency noise layer, the mid-frequency human voice layer, and the high-frequency technical terminology layer. min Value; specifically, for the low-frequency noise layer, set G. min The value is 0.05, which strengthens the suppression of background noise; for the middle vocal layer, G is set to... min The value is 0.15 for equalization and noise reduction; for the high-frequency technical terminology layer, G is set to... min The value is 0.35 to ensure that vocal terminology segments are not excessively attenuated; a4: Outputs the corrected amplitude spectrum and the original phase; The acoustic noise reduction module ultimately outputs the meeting IDs of the mediator and the mediator, as well as the noise-reduced spectral feature stream.

[0044] The following operations are performed in the weighted search decoding module: b1: The weighted search decoding module contains a pre-trained acoustic model based on a convolutional neural network and a pre-trained general language model. The weighted search decoding module receives the spectral feature stream output by the acoustic noise reduction module, outputs the temporal phoneme sequence through the acoustic model and sends it into the general speech model. The general language model generates multiple candidate word paths based on phonemes, and the basic score of the general vocabulary is uniformly set to 1.0. b2: Traverse the phoneme segments of each path, retrieve the traffic accident terminology knowledge base to match professional terms, and if a match is found, add a special weight of 3.5~5.0 to increase the path score; if no match is found, the score remains unchanged. b3: Compare the total scores after all path corrections, select the optimal word sequence, and output the final text.

[0045] The traffic accident terminology knowledge base module constructed in this application includes: a terminology collection and entry subunit, a hierarchical terminology storage unit, a dual-index hybrid retrieval unit, and a dynamic incremental update unit; The terminology collection and database sub-unit extracts standard professional terms from multiple terminology sources to form a raw terminology material pool; terminology sources include: official regulations, standard mediation documents, road coding specifications, and operation and management regulations. The hierarchical terminology storage unit includes a professional terminology database, comprising: an insurance compensation core sub-database, a unit user main terminology sub-database, a road traffic accident liability sub-database, and a spoken language mapping sub-database; The core sub-database for insurance claims is used to store core claims terms, such as "compulsory traffic accident liability insurance," "lost wages," and "compensation," and is used to prioritize the identification of key terms related to claims. The unit user subject terminology sub-database is used to store enterprise-specific terms for mediation, such as: "operating vehicles", "enterprise agents", "enterprise representatives", etc. The Road Traffic Accident Liability Sub-database is used for terms related to liability determination, such as "full liability," "equal liability," and "no liability." The colloquial mapping sub-database is used to store the one-to-one correspondence between colloquial terms and standard professional terms: for example, "medical expenses" corresponds to "medical fees", "car crash" corresponds to "vehicle collision", "road repair section" corresponds to "road construction section", etc. In practice, all data is collected based on a historical database of traffic accidents.

[0046] The hierarchical terminology storage unit stratifies terms according to their business importance and configures differentiated weights. The baseline weight for general vocabulary is fixed at 1.0, while the weight for specialized terminology is configured at 4.0.

[0047] Dual-index hybrid retrieval unit: To adapt to two types of tasks, namely real-time speech decoding and text standardization, two independent indexes are constructed. The first is the phoneme prefix index, which performs fuzzy prefix matching on real-time phoneme segments during the decoding process, matches professional terms, and adds weights. The second is the text hash index, which calls the content in the spoken language mapping sub-library to replace spoken expressions with official standard terms in batches, so as to achieve professionalization and standardization of output text.

[0048] Dynamic incremental update unit: updates and iterates the content in the hierarchical terminology storage unit of the knowledge base; extracts professional terms from documents and records through NLP algorithms, and outputs them to the hierarchical terminology storage unit according to terminology type; enables the knowledge base to have self-iteration and self-optimization capabilities, adapts to long-term business iteration, avoids the defects of knowledge base solidification and update lag, and solves the problems of business terminology updates, insurance type updates, and changes in spoken language habits.

[0049] This application is based on a voice information acquisition module that is specially customized for traffic accident mediation scenarios. When recognizing professional terms related to commercial vehicles, insurance claims, road codes, and accident liability, it can effectively solve the problems of homophony confusion and mistranslation of spoken language. At the same time, it optimizes the shortcomings of the output text being too colloquial and lacking professionalism, thereby improving the efficiency of automated remote mediation.

[0050] This application also includes a complete process for mediation agreements: such as Figure 3 As shown, after the mediator confirms that the signatures of all parties and the documents are correct, the data is securely transmitted to the intranet; after the mediation agreement is rendered in a standardized manner by the iReport template engine, it is encrypted using the national cryptographic algorithm, and then the encrypted document is transmitted to the external network, where the person being mediated can view the official mediation document with electronic signature on their mobile app.

[0051] Specifically, the mediation document confirmation module performs the mediation agreement generation operation, which includes the following steps.

[0052] c1: Build an iReport hierarchical template library based on the classification of unit users. The iReport hierarchical templates include various types such as templates for ordinary parties, corporate templates, and traffic accident templates. Unlike ordinary litigants, corporate users are involved in various scenarios during mediation, including corporate compensation in traffic accidents and government-enterprise disputes. Each document contains unique fields such as the company's unified social credit code and legal representative information, requiring strict adherence to standards for archiving. Therefore, this method establishes a hierarchical iReport template library based on corporate user classification to distinguish between different user categories.

[0053] c2: Preset user-specific field fill positions and fixed layer area for electronic signature information in the template; c3: After the mediator confirms that the signatures of all parties and the documents are correct, the data is transmitted to the intranet in an encrypted manner; The electronic signature layer and document content are encrypted using the SM4 algorithm in fragmentation. c4: Design an algorithm for converting data between internal and external networks, separating and storing electronic signature information in JSON format from external networks and other text information, and automatically translating it into report parameters that iReport can recognize.

[0054] The internal / external network data conversion algorithm includes the following operations: d1: Receive complete JSON data transmitted over the internet based on a predefined entity bean; d2: Split the JSON data into two independent data streams; Independent data streams include: text-based business data streams and electronic signature data streams; The text-based business data stream includes structured data such as: the basis for mediation, basic facts and losses, compensation items and amounts, compensation liability and proportion, and compensation methods and time limits; The electronic signature data stream includes: the Base64 string of the original binary image of the electronic signature, which is uniformly serialized during transmission; d3: Decodes the Base64 string to restore the original byte array of the electronic signature; simultaneously parses the header image information of the byte array to obtain the pixel width and height of the electronic signature image, generates iReport layer parameters, and outputs them to the document dynamic generation algorithm for PDF rendering.

[0055] The iReport template has a built-in dynamic document generation algorithm that automatically adjusts the position of the electronic signature layer according to the length of the document content, avoiding document misalignment caused by signature information obscuring document content. The dynamic document generation algorithm includes the following steps.

[0056] c5: Based on the document type identifier carried by the internal and external network data conversion algorithm, retrieve the corresponding iReport template. The template has two preset independent layer areas: the main text rendering layer and the electronic signature layer reserved layer. c6: Pre-render the main text; Calculate the boundary of the area occupied by the complete text content in memory, and obtain the set of coordinates of the remaining blank area of ​​the page. The region boundaries include: page boundary coordinates, total occupied height, and page break position; c7: Read the layer parameters such as pixel width and height of the signature image calculated in the internal and external network data conversion algorithm, combine them with the data of the area boundary occupied by the main text content, traverse the blank area of ​​the page, and select the optimal coordinates for placing the signature image; c8: Dynamic layer rendering and document compositing; Using the optimal coordinate information, the independently encapsulated electronic signature layer is rendered to the target location, effectively avoiding the problem of the signature obscuring the document content during transmission; the main text rendering layer and the reserved electronic signature layer are merged to generate a complete PDF of the mediation agreement.

[0057] c9: Send the mediation document with signature information to the mediator.

[0058] c10: The mediation document confirmation module stores mediation documents with the signature of the mediated party as official documents.

[0059] During storage, SM4 encryption is used, and the national standard SM3 hash operation is called synchronously to generate fixed-length ciphertext. The original SM3 ciphertext is stored in the database to determine whether the document has been tampered with. For each user type of mediator, an SM2 private key is assigned to them to digitally sign the original SM3 ciphertext stored in the database.

[0060] This application employs a combined storage system of SM4 symmetric encryption, SM3 hash digest, and SM2 digital signature. The core storage component uses SM4 encryption as its core, supplemented by SM3 integrity verification and SM2 signature binding. First, the electronic signature layer and document content are encrypted in segments using the SM4 algorithm, significantly improving the speed of batch document encryption. During SM4 encryption, a fixed-length ciphertext is generated simultaneously using SM3 hash operations. The original SM3 ciphertext is stored in the database to determine whether the document has been tampered with. Finally, for different user types such as ordinary parties, corporate entities, and mediators, their respective SM2 private keys are assigned to digitally sign the aforementioned SM3 hash digest, ensuring the authenticity of the electronic signature.

[0061] The middle service layer is equipped with an asynchronous message queue transmission unit. The electronic signature information of the mediated party and the push messages for mediation documents are exchanged across the network via the queue. During peak business periods, a large backlog of messages in the queue may occur, potentially leading to severe data synchronization delays or message loss. This could result in the mediated party not receiving the membership SMS or document preview link in real time. To address data failures during peak business periods, this application designs transmission logic for large volumes of unstructured data such as mediation documents and electronic signature information. Specifically, the mediation document confirmation module also includes the following operations: e1: Set up a separate business queue for large-volume unstructured data; large-volume unstructured data includes: electronic signature information of the parties to the mediation and mediation documents; e2: An asynchronous message queue transmission unit is installed in the middle service layer, and the push of large-volume unstructured data messages is completed through the queue for cross-network exchange; e3: Configure an automatic retry policy for each cross-network exchange message: When queue data consumption fails, re-push according to the preset time interval and number of pushes (fre); if consumption still fails after fre retries, write it into the system error log, and manually resend the electronic signature information and mediation document message according to the log. In this embodiment, fre=3 times, and the corresponding sending intervals for each time are: 1 minute, 5 minutes, and 10 minutes, that is: when queue data consumption fails, re-push at intervals of 1 minute, 5 minutes, and 10 minutes; if consumption still fails after three retries, write it into the system error log, and manually resend the electronic signature information and mediation document message according to the log, without directly discarding valid business data.

[0062] The mediation method based on the aforementioned remote video system for unit users includes the following steps.

[0063] S1: Dictionary of personnel roles for mediators; The system receives a request from the mediator to maintain the participant information through the video mediation participant information maintenance module; The personnel role dictionary includes: a custom participant type field; this field allows mediators to customize the roles of participants. In this embodiment, the personnel roles include: mediator, parties involved, insurance company specialist, corporate legal representative, unit / organization representative, lawyer, guardian, and witness; This application is designed for corporate users to handle matters on behalf of others, addressing the limitation of individuals being unable to do so. By configuring multiple types of participants, such as insurance company specialists, legal representatives, and institutional agents, and classifying and controlling participation permissions based on personnel type, it is suitable for the business models of dispute mediation handled by enterprises, institutions, transportation companies, and insurance companies. This overcomes the limitation of existing remote mediation technologies that are only designed for individual parties and do not support corporate users to participate in the process on behalf of others.

[0064] S2: The mediator submits a meeting reservation request for a remote video mediation meeting and controls the meeting information of the video mediation meeting; The system completes the reservation of a video mediation meeting based on the meeting information submitted through the video mediation meeting reservation module.

[0065] S3: The system completes the reservation of video mediation meetings based on meeting information. After successful reservation, a unique meeting ID and temporary access verification code are automatically generated and pushed to all participants through an encrypted SMS channel. The system uses Redis to cache the real-time status of the meeting and the online information of the participants, and supports multiple people to join the meeting concurrently.

[0066] S4: Before the meeting starts, based on the meeting information, enable the mediators and mediators to enter the video mediation meeting.

[0067] S5: During a video mediation conference, the system generates basic information items for the mediation information collection module based on the maintained video mediation conference reservation information; and supports mediators to conduct remote video mediation, collect mediation compensation information, and realize the real-time production process of mediation records.

[0068] S6: After completing the collection of mediation information, the mediator can create mediation records in real time based on the collected mediation compensation information; The mediator selects the type of document to be prepared based on the mediation situation, generates a mediation document from the written record, and sends an electronic link of the mediation document to the party being mediated online. If a settlement is reached through mediation, a damages settlement agreement shall be prepared. If no agreement can be reached, a settlement agreement for damages shall be issued.

[0069] S7: The person being mediated receives the mediation document on the mobile app, previews the document through the mediation document confirmation module, and signs the document on the mobile app after confirming that the content of the mediation document is correct. The document with the signature information is then generated and pushed to the mediator.

[0070] S8: The mediator receives the mediation document with signature information on the web platform, and after confirming that the content and signature information of the mediation document are correct, generates a mediation document with the signature information of the mediated party and saves it as an official document.

[0071] S9: The entire mediation process is recorded on video and audio simultaneously, and a complete record is kept. After the video mediation is completed, the function of downloading the video mediation video is provided.

[0072] This application utilizes an audio-visual visual interactive method to conduct business, fully and objectively reproducing the communication content, evidence presentation process, and confirmation by all parties at the mediation site. The entire mediation process is simultaneously recorded on video and audio, with complete record-keeping. After the video mediation is completed, a download function for the video mediation video is provided, reducing the probability of secondary disputes and disagreements between the disputing parties, and further ensuring the compliance and effectiveness of the mediation result. The remote video mediation system designed in this application can overcome the limitations of time and space, efficiently conduct mediation work, shorten the dispute resolution cycle, and achieve the standardized advancement and efficient operation of law enforcement procedures.

[0073] After using the technical solution of this invention, a communication link is established between the mediator (with video service capabilities) using a web client and the respondent (with video service capabilities) using a mobile app. The mediation information collection module records compensation information during the mediation process in real time, and the mediation documents are pushed to the respondent immediately via transmission technology. The mediation document confirmation module ensures the authenticity and validity of the mediation documents. This method provides a remote video mediation technology for organizational users. Compared with existing technologies, it adapts to the business models of enterprises, transportation companies, and insurance institutions handling dispute mediation by configuring multiple participation roles such as agency representatives, legal agents, and insurance specialists. It overcomes the limitation of existing traffic management remote mediation platforms, which are only designed for individual accident parties and do not support participation by organizational personnel. Through remote audio and video transmission technology, it breaks the dependence of traditional offline mediation on physical space, enabling remote connection between mediators and respondents in cross-regional scenarios. This not only significantly reduces explicit costs such as travel and time, but also lowers the participation threshold for respondents due to geographical barriers, effectively improving participation and cooperation in mediation work. This method enables remote screen sharing, breaking down barriers to collaboration among ordinary parties, institutional users, and case-handling units. Compared to traditional offline oral mediation and paper-based record-keeping, this method, relying on Automatic Speech Recognition (ASR) technology and a traffic accident terminology knowledge base, can completely, objectively, and continuously reconstruct the communication content, evidence presentation process, and confirmations made by all parties at the mediation site. The designed mediation document confirmation process effectively avoids problems such as omissions, subsequent alterations, and recording errors in manual record-keeping, significantly improving the authenticity, objectivity, and legal validity of the mediation process and its results.

Claims

1. A remote video mediation system for unit users, characterized in that, It includes: a video mediation participant information maintenance module, a video mediation meeting information reservation module, a mediation information collection module, a mediation document confirmation module, and a client application; The video mediation participant information maintenance module is used to receive participant information maintenance requests submitted by mediators; the video mediation participant information maintenance module sets up a participant role dictionary, which supports custom participant type fields; through the custom participant type fields, mediators can customize the roles of participants; The video mediation meeting information reservation module is used to receive meeting reservation requests submitted by mediators. The meeting reservation request includes meeting information, including: start time, end time, participants, and meeting title. After successful reservation, a unique meeting ID and a temporary access verification code are automatically generated and pushed to all participants via an encrypted SMS channel. The mediation information collection module supports mediators in collecting mediation compensation information during multi-party video conferences, enabling real-time creation of mediation records. The mediation compensation information includes: the basis for mediation, basic facts and losses, compensation items and amounts, compensation liability and proportion, and compensation method and time limit. The mediation information collection module includes: a voice information collection module, a screen information collection module, and a document production module; The voice information acquisition module uses voice recognition technology to translate the conversation between the mediator and the mediator in a video mediation conference in real time; it also allows the mediator to select conversation content to add to the mediation record. The screen information acquisition module supports the mediator's screen sharing function, allowing the mediator to present evidence and related materials online and exchange information with the mediator, replacing the offline direct display of original documents. After information collection is completed, the document creation module allows mediators to select the type of document to generate mediation documents based on the mediation situation, and then generate an electronic link online and push it to the party being mediated. The types of mediation documents include: damage compensation mediation agreement and damage compensation mediation termination agreement. The mediation document confirmation module is used to respond to the mediation party's operation of receiving the mediation document on the client. The mediation document confirmation module receives the mediation document after the mediation party signs and confirms it, generates the mediation document content with signature information and pushes it to the mediator; the mediation document confirmation module stores the mediation document with the mediation party's signature as an official document. The client includes: mediators and those being mediated participating in remote video mediation conferences based on the client.

2. The remote video mediation system for unit users according to claim 1, characterized in that: The voice information acquisition module includes: an acoustic noise reduction module, a weighted search decoding module, and a traffic accident terminology knowledge base module; The acoustic noise reduction module is used to eliminate ambient noise, audio and video echoes, and other speech interference at the mediation site. The weighted search decoding module receives the spectral feature stream generated sequence from the acoustic noise reduction module and calls the traffic accident terminology knowledge base in real time to perform path weighting and scoring, thereby increasing the priority of candidate paths for traffic accident terminology and systematically improving the accuracy of traffic accident terminology recognition. The traffic accident terminology knowledge base module provides hierarchical professional terminology, weighted score constraints, colloquial synonym mapping, and business field binding data services to the weighted search decoding modular output module. The original mixed speech stream input from the participants' terminals is used as the system's speech input; it is then fed into the acoustic noise reduction module, which performs multi-channel splitting, echo cancellation, and layered frequency domain noise reduction, outputting standard spectral features. The spectral features output by the acoustic noise reduction module are then input into the weighted search decoding module. During the decoding process, a real-time call request is initiated to read the terminology sequences, matching weights, and synonym mapping rules stored in the traffic accident terminology knowledge base module. The weighted search decoding module scores and filters candidate word paths based on the knowledge base weights, and then outputs standardized traffic accident mediation text for archiving mediation records.

3. The remote video mediation system for unit users according to claim 2, characterized in that: The acoustic noise reduction module includes: a multi-channel audio splitting subunit, an echo cancellation subunit, and a layered frequency domain noise reduction subunit; the multi-channel audio splitting subunit segments the multi-channel mixed audio according to the unique identifier of the conference terminal device or the participant ID of the participant, decouples the original audio, and isolates the voice of each participant; The echo cancellation subunit uses a normalized least mean square adaptive filtering algorithm to synchronously pull the audio played at the far end of the device's speaker as a reference signal and dynamically update the filtering weights to cancel the playback echo collected by the microphone. w n+1 =w n +(1 / ||x n+1 || 2 )·e(n)x n ; In the formula, w n Let x be the filter coefficient at time n, e(n) be the error signal, and x be the filter coefficient at time n. n For audio input; ||·|| 2 To find the square of the second norm; The hierarchical frequency domain noise reduction subunit performs the following operations: a1: Receives the speech slices output by the echo cancellation subunit, performs short-time Fourier transform, splits the amplitude spectrum from the original phase, and performs frequency domain segmentation; the segmented frequency domain includes: a low-frequency noise layer of 0-2kHz, a mid-voice layer of 2-4kHz, and a high-frequency technical terminology layer of 4-8kHz. And complete the three-layer frequency domain segmentation according to 0-2kHz, 2-4kHz, and 4-8kHz; a2: Each frequency band independently updates the noise floor in real time using the improved minimum control recursive average IMCRA algorithm, and calculates the a priori and post-affirmative signal-to-noise ratios and the probability of speech presence p(k,l) in layers, where k is the frequency index and l is the time index; a3: Based on the minimum gain G differentiated by each frequency band min Solve for the optimal frequency domain gain of OM-LSA, correct the amplitude spectrum with gain and preserve the original phase; G(k,l)=[G H_1 (k,l)] p(k,l) ·G min ((1-p(k,l)) ; In the formula, [G H_1 [(k,l)] represents the optimal gain when speech is present, G min This is the minimum gain limit for this frequency band; Different G values ​​are set for the low-frequency noise layer, the mid-frequency human voice layer, and the high-frequency technical terminology layer. min value; a4: Outputs the corrected amplitude spectrum and the original phase; The acoustic noise reduction module ultimately outputs the meeting IDs of the mediator and the mediator, as well as the noise-reduced spectral feature stream.

4. The remote video mediation system for unit users according to claim 2, characterized in that: The weighted search decoding module performs the following operations: b1: The weighted search decoding module has a pre-trained acoustic model based on a convolutional neural network and a pre-trained general language model built in. The weighted search decoding module receives the spectral feature stream output by the acoustic noise reduction module, outputs a temporal phoneme sequence through the acoustic model and sends it into the general speech model. The general language model generates multiple candidate word paths based on phonemes, and the basic score of the general words is uniformly set to 1.

0. b2: Traverse the phoneme segments of each path, retrieve the professional terms from the traffic accident terminology knowledge base, and if a match is found, add a special weight of 3.5~5.0 to increase the path score; if no match is found, the score remains unchanged. b3: Compare the total scores after all path corrections, select the optimal word sequence, and output the final text.

5. The remote video mediation system for unit users according to claim 2, characterized in that: The traffic accident terminology knowledge base module includes: a terminology collection and entry sub-unit, a hierarchical terminology storage unit, a dual-index hybrid retrieval unit, and a dynamic incremental update unit; The terminology collection and database sub-unit extracts standard professional terms from multiple terminology sources to form a raw terminology material pool; the terminology sources include: official legal provisions, standard mediation documents, road coding specifications, and operation and management regulations. The hierarchical terminology storage unit is equipped with a professional terminology database, including: an insurance compensation core sub-database, a unit user main terminology sub-database, a road traffic accident liability sub-database, and a spoken language mapping sub-database; The core insurance claims sub-database is used to store core claims terms and is used to prioritize the identification of key terms related to claims. The unit user subject terminology sub-database is used to store enterprise-specific mediation terminology; The road traffic accident liability sub-database is used for liability determination terminology; The colloquial mapping sub-library is used to store the one-to-one correspondence between colloquial terms and standard professional terms; The hierarchical term storage unit hierarchizes terms according to their business importance and configures differentiated weights; The dual-index hybrid retrieval unit constructs two independent indexes to adapt to the two types of tasks: real-time speech decoding and text standardization. The first is a phoneme prefix index, which performs fuzzy prefix matching on real-time phoneme segments during the decoding process, matches professional terms, and adds weights. The second is a text hash index, which calls the content in the spoken language mapping sub-library to replace spoken expressions with official standard terms in batches. The dynamic incremental update unit updates and iterates the content in the hierarchical terminology storage unit of the knowledge base; it extracts professional terms from documents and records using NLP algorithms and outputs them to the hierarchical terminology storage unit according to terminology type.

6. The remote video mediation system for unit users according to claim 1, characterized in that: The clients include: a web page client for mediators and a mobile app for the parties being mediated; The mediation information collection module also includes the following operations: Before the participants' clients officially enter the video mediation meeting, the system will forcibly intercept them to confirm whether each client has enabled camera and microphone permissions. If not, the client will be prompted to enable permissions via message. If permissions are enabled, the meeting will proceed to the next stage. The mediation information collection module allows mediators to use a web client to read the camera and microphone permission status via the navigator.permissions.query method; if the returned result is denied or prompt, it indicates that the permissions are not enabled. The mediation information collection module will prompt that permission is not enabled if the ContextCompat.checkSelfPermission method returned by the mediator using a mobile app is not PackageManager.PERMISSION_GRANTED.

7. The remote video mediation system for unit users according to claim 1, characterized in that: The mediation document confirmation module specifically includes the following operations: c1: Build an iReport hierarchical template library based on the classification of unit users. The iReport hierarchical templates include various types such as ordinary party templates, corporate templates, and traffic accident templates. c2: Preset user-specific field fill positions and fixed layer area for electronic signature information in the template; c3: After the mediator confirms that the signatures of all parties and the documents are correct, the data is transmitted to the intranet in an encrypted manner; The electronic signature layer and document content are encrypted using the SM4 algorithm in fragmentation. c4: Design an algorithm for converting data between internal and external networks, separating and storing electronic signature information in JSON format from external networks and other text information, and automatically translating it into report parameters that iReport can recognize; c5: Based on the document type identifier carried by the internal and external network data conversion algorithm, retrieve the corresponding iReport template. The template has two preset independent layer areas: the main text rendering layer and the electronic signature layer reserved layer. c6: Pre-render the main text; Calculate the boundary of the area occupied by the complete text content in memory, and obtain the set of coordinates of the remaining blank area of ​​the page. The region boundary includes: page boundary coordinates, total occupied height, and page break position; c7: Read the layer parameters of the signature image calculated in the internal and external network data conversion algorithm, combine them with the data of the area boundary, traverse the blank area of ​​the page, and select the optimal coordinates for placing the signature image; The layer parameters include: pixel width and height; c8: Dynamic layer rendering and document compositing; Based on the coordinate information of the optimal coordinates, the independently encapsulated electronic signature layer is rendered to the target position, and the main text rendering layer and the reserved electronic signature layer are merged to generate a complete PDF of the mediation agreement; c9: Send the mediation document with signature information to the mediator; c10: The mediation document confirmation module stores mediation documents with the signature of the mediated party as official documents; During storage, SM4 encryption is used, and the national standard SM3 hash operation is called synchronously to generate fixed-length ciphertext. The original SM3 ciphertext is stored in the database to determine whether the document has been tampered with. For each user type of mediator, an SM2 private key is assigned to them to digitally sign the original SM3 ciphertext stored in the database.

8. The remote video mediation system for unit users according to claim 7, characterized in that: The internal / external network data conversion algorithm includes the following operations: d1: Receive complete JSON data transmitted over the internet based on a predefined entity bean; d2: Split the JSON data into two independent data streams; The independent data streams include: text business data stream and electronic signature data stream; The text business data stream includes structured data such as: mediation basis, basic facts and losses, compensation items and amounts, compensation liability and proportion, compensation methods and time limits; The electronic signature data stream includes: a Base64 string that is uniformly serialized during transmission of the original binary image of the electronic signature; d3: Decodes the Base64 string to restore the original byte array of the digital signature; simultaneously parses the header image information of the byte array to obtain the pixel width and height of the digital signature image, and generates iReport layer parameters for subsequent PDF rendering.

9. The remote video mediation system for unit users according to claim 1, characterized in that: The mediation document confirmation module also includes the following operations: e1: A separate business queue is set up for large-volume unstructured data; the large-volume unstructured data includes: the electronic signature information of the mediated party and mediation documents; e2: An asynchronous message queue transmission unit is installed in the middle service layer, and the push of large-volume unstructured data messages is completed through the queue for cross-network exchange; e3: Configure an automatic retry policy for each cross-network exchange message: When queue data consumption fails, re-push according to the preset time interval and number of pushes (fre); if consumption still fails after fre retries, write it to the system error log, and manually resend the electronic signature information and mediation document message according to the log.

10. A remote video mediation method for unit users, characterized in that, It includes the following steps: S1: Dictionary of personnel roles for mediators; The personnel role dictionary includes: a custom participant type field; the custom participant type field allows mediators to customize the roles of participants. S2: The mediator submits a meeting reservation request for a remote video mediation meeting and controls the meeting information of the video mediation meeting; S3: Based on the meeting information, complete the reservation for the video mediation meeting. After successful reservation, a unique meeting ID and temporary access verification code are automatically generated and pushed to all participants via an encrypted SMS channel. S4: Before the meeting start time arrives, based on the meeting information, the mediators and mediators enter the video mediation meeting; S5: During video mediation meetings, mediators collect information on mediation and compensation, enabling real-time creation of mediation records. S6: After completing the collection of mediation information, the mediator selects the type of document to be prepared according to the mediation situation, generates the mediation document from the transcript, and sends the electronic link of the mediation document to the mediator online. If a settlement is reached through mediation, a damages settlement agreement shall be prepared. If no agreement can be reached, a settlement agreement for damages shall be prepared. S7: The respondent receives the mediation document on the mobile app, and after confirming that the content of the mediation document is correct and signing it, generates the mediation document content with the signature information and pushes it to the mediator. S8: The mediator receives the mediation document content with signature information on the web terminal, and after confirming that the content of the mediation document and the signature information are correct, generates a mediation document with the signature information of the mediated party and stores it as an official document; S9: The entire mediation process is recorded on video and audio simultaneously, and a complete record is kept. After the video mediation is completed, the function of downloading the video mediation video is provided.

Images