An intelligent call processing method and system fusing digital human interaction

Abstract

The application discloses a kind of intelligent call processing methods and systems of fusing digital human interaction, it is related to digital human technical field, method includes that intelligent call hub receives the call content initiated by user terminal, collects user terminal identification information and current service type information, generates unique session state code;Intelligent call hub is based on session state code associated user historical service data, generates and sends to digital human interaction service engine including session state code and service adaptation parameter digital human service request;Intelligent call hub receives the customized digital human model associated information returned by digital human interaction service engine, and digital human audio and video stream data is encapsulated as standardized data packet;Intelligent call hub is compressed to standardized data packet, and the lightweight transmission of digital human interaction data is completed.

Description

Technical Field

[0001] This invention relates to the field of digital human technology, and in particular to an intelligent call processing method and system that integrates digital human interaction. Background Technology

[0002] With the development of artificial intelligence and audio / video processing technologies, digital human interaction has become the core carrier of intelligent call services. Achieving human-like communication through digital humans can significantly enhance the user experience, and is therefore widely used in various service-oriented call scenarios. The core requirement of this type of intelligent call system is to achieve efficient connection throughout the entire process of user calls, data processing, digital human responses, and data transmission. This requires ensuring the real-time flow of interactive data while controlling server resource consumption to avoid service lag due to redundant data processing.

[0003] However, existing intelligent call processing solutions that integrate digital human interaction generally suffer from technical limitations due to the fragmentation of contextual data collaboration. This leads to low efficiency in the entire data flow and waste of system resources. Specifically, after a user initiates a call, the system lacks a unified standard for collecting and encoding basic information such as terminal identification and service type. The generated identification information is either highly redundant, consuming too many storage resources, or lacks core features, resulting in low matching efficiency when subsequently associating with the user's historical business data. This necessitates repeatedly retrieving all data, consuming additional computing power. When sending requests to the digital human interaction service engine, there is a lack of a context association mechanism based on unified encoding. The integration logic of business adaptation parameters and session information is loose, easily leading to problems where parameters do not match the current business scenario, indirectly increasing the redundancy of cross-module data interaction. Furthermore, the encapsulation and transmission of digital human audio and video stream data lack systematic design. Inconsistent data packet formats result in poor compression processing, either causing insufficient compression and excessive transmission bandwidth consumption, or excessive compression affecting audio and video quality, making it impossible to achieve a balance between lightweight transmission and interaction quality. Summary of the Invention

[0004] To address the technical problems existing in the prior art, this invention provides an intelligent call processing method that integrates digital human interaction.

[0005] The technical solution adopted in this invention is:

[0006] The first aspect of this application provides an intelligent call processing method that integrates digital human interaction, including the following:

[0007] The intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and the current service type information, and generates a unique session status code;

[0008] The intelligent call center associates user historical business data with session state coding, generates a digital human service request containing session state coding and business adaptation parameters, and sends it to the digital human interaction service engine.

[0009] The intelligent call center receives customized digital human model association information returned by the digital human interaction service engine and encapsulates the digital human audio and video stream data into standardized data packets;

[0010] The intelligent call center compresses standardized data packets to achieve lightweight transmission of digital human interaction data.

[0011] Preferably, the intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and current service type information, and generates a unique session status code including the following:

[0012] The intelligent call center extracts the core feature parameters of user terminal identification information and current service type information, and constructs a feature parameter matrix.

[0013] The feature parameter matrix is ​​reduced in dimension by a feature fusion algorithm to generate a low-dimensional core feature vector.

[0014] Core information condensation encoding is generated based on low-dimensional core feature vectors, and a reverse verification mapping table for encoding is constructed simultaneously.

[0015] Based on the encoding reverse verification mapping table, redundant fields are stripped, feature fields with invalid mappings are removed, and a unique and reversibly verifiable session state code is output.

[0016] Preferably, the intelligent call center associates user historical service data based on session state coding, including the following:

[0017] The intelligent call center retrieves the user's historical service dataset through session state coding and constructs a historical data time-series queue.

[0018] A data integrity verification algorithm is used to verify the historical data in the time-series queue frame by frame, and the valid data segments that pass the verification are marked.

[0019] By performing cosine similarity matching between business scenario label vectors and effective data segments, target data whose suitability meets the threshold is selected.

[0020] Business adaptation parameters are generated based on the target data, and outdated fields are written to the data update log to drive incremental updates of the historical business dataset.

[0021] Preferably, the step of reducing the dimensionality of the feature parameter matrix using a feature fusion algorithm to generate a low-dimensional core feature vector includes the following:

[0022] Perform singular value decomposition on the feature parameter matrix and extract the principal singular values ​​of the matrix to form the initial eigenvectors;

[0023] The initial eigenvectors are smoothed using the Laplacian operator to eliminate feature noise;

[0024] Huffman coding is used to compress the smoothed feature vectors to generate low-dimensional core feature vectors, with the vector dimensions mapped one-to-one with the business type.

[0025] Preferably, the intelligent call center receives customized digital human model association information returned by the digital human interaction service engine, and encapsulates the digital human audio and video stream data into standardized data packets including the following:

[0026] Analyze the associated information of the customized digital human model and extract basic parameters such as bitrate, frame rate, and resolution of the audio and video streams;

[0027] Build a dedicated context transport protocol, define four basic protocol fields: session state encoding, business execution status, core business data, and timestamp, and set field validation bits;

[0028] A fault-tolerant redundancy field is embedded in the protocol layer. This field stores the MD5 checksum of the core business data, forming a dual check mechanism with the core business data field.

[0029] The audio and video stream data and basic parameters are encapsulated in the order of protocol fields to generate standardized data packets with double verification mechanism.

[0030] Preferably, the generation of the digital human service request, which includes session state encoding and service adaptation parameters, includes the following:

[0031] Based on the session state encoding, the business type identifier is parsed, and the corresponding parameters are called to generate a template library;

[0032] User interaction behavior data is captured by a real-time interactive behavior collector, and the time sequence of behavioral features is extracted.

[0033] The behavioral feature time series is input as a parameter to generate a template library, triggering dynamic interpolation adjustment of the template parameters;

[0034] The adjusted parameters are smoothed and denoised using the Kalman filter algorithm to output service adaptation parameters.

[0035] Preferably, the construction of the dedicated context transport protocol includes the following:

[0036] Define the fixed byte length and binary storage format of the protocol fields, and construct a field format dictionary;

[0037] Establish a linkage update rule between core business data fields and timestamp fields, so that field modifications trigger automatic timestamp rewriting;

[0038] Construct logical constraint equations between business execution status fields and core business data fields to limit the correlation of field values;

[0039] Integrate field format dictionaries, linked update rules, and constraint equations to form a dedicated context transmission protocol.

[0040] Preferably, after generating the standardized data packet with double verification mechanism, the method further includes the following:

[0041] The intelligent call center captures call interruption events through the call status monitoring module, generates breakpoint markers, synchronously calls the generated standardized data packets with double verification mechanism, and extracts the real-time synchronized business progress information throughout the entire process of generating the data packet, and constructs a breakpoint data snapshot containing standardized data packets and business progress information.

[0042] Construct an evaluation index system for the urgency of breakpoint business, using core business data in standardized data packages as the core evaluation dimension, and combine business progress information to complete the priority quantitative scoring of breakpoint data snapshots;

[0043] A dynamic hierarchical storage architecture is built based on the scoring results. A dedicated hot data area stores high-priority breakpoint data snapshots, and a cold data area stores low-priority breakpoint data snapshots. The snapshots fully retain the association mapping between standardized data packets and business progress information.

[0044] Configure a repeated call trigger threshold. When the frequency of calls from the same user terminal reaches the threshold, the migration of the cold data area interruption point data snapshot is triggered. During the migration process, the association mapping between standardized data packets and service progress information remains unchanged.

[0045] Preferably, a repeated call trigger threshold is configured. When the frequency of calls from the same user terminal reaches the threshold, the migration of the cold data area interruption point data snapshot is triggered. During the migration process, the association mapping between standardized data packets and service progress information remains unchanged, including the following:

[0046] Capture call requests initiated by user terminals and extract terminal identification features and call timestamps from the requests;

[0047] Based on the terminal identification features, query the breakpoint data index library and preload the breakpoint data snapshot corresponding to the terminal into the cache buffer in advance.

[0048] The system uses a two-dimensional judgment logic of terminal identifier matching and call time window verification to complete the identity verification for connecting calls. After the verification is successful, it directly calls the breakpoint data snapshot preloaded in the cache buffer, and drives the digital human interaction service engine to continue the business process based on the standardized data packets and business progress information therein.

[0049] The second aspect of this application provides an intelligent call processing method integrating digital human interaction. The method includes:

[0050] The user terminal is used to initiate a call to the intelligent call center, transmit its own terminal identification information and current service type information, and receive digital human interactive audio and video stream data fed back by the intelligent call center.

[0051] The intelligent call hub receives call content initiated by a user terminal, collects and parses terminal identification information and current service type information, and generates a unique session state code. Based on the session state code, it associates the user's historical service data, generates a digital human service request containing the session state code and service adaptation parameters, and sends it to the digital human interaction service engine. It receives customized digital human model association information returned by the digital human interaction service engine, encapsulates the digital human audio and video stream data into standardized data packets, compresses the standardized data packets, and transmits lightweight digital human interaction data to the user terminal.

[0052] The digital human interaction service engine is used to receive digital human service requests sent by the intelligent call center, match the corresponding digital human model resources based on the session state code and business adaptation parameters in the request, generate customized digital human model association information, and feed it back to the intelligent call center.

[0053] The beneficial effects of this invention are: by generating a unique session state code based on terminal identifier and service type information, a unified benchmark is provided for data interaction in each stage, which helps to avoid the problems of redundancy or feature loss in traditional coding, reduces the storage occupation of identifier information, improves the matching accuracy of subsequent associated user historical service data, and reduces the computing power consumption caused by repeatedly retrieving the full amount of data.

[0054] Using session state coding as the core link, historical business data is accurately matched with current business needs and business adaptation parameters are generated. This makes the content of digital human service requests more in line with scenario requirements, helps reduce the situation where parameters do not match business scenarios, and thus reduces the redundancy of secondary data interaction across modules.

[0055] By encapsulating digital human audio and video stream data into standardized data packets, a unified format foundation is provided for subsequent compression processing. This improves the targeting of compression processing, helps to achieve lightweight data transmission while ensuring basic interaction quality, reduces transmission bandwidth usage, and alleviates system transmission pressure. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the method flow of Embodiment 1 of the present invention;

[0057] Figure 2 This is a schematic diagram of the system interaction process in Embodiment 2 of the present invention. Detailed Implementation

[0058] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0059] In existing technologies, there is no unified standard for encoding terminal identification and service type when a user makes a call, which is particularly problematic in real-world scenarios. For example, if a user uses a mobile phone with IMEI code 861234567890123 to call a bank's customer service hotline to inquire about their credit card statement, the current system may directly store the complete IMEI code in plaintext, while only using the word "inquiry" to indicate the service type. This encoding method not only leads to redundancy in terminal identification information and occupies additional storage resources, but also, because the service type encoding lacks core characteristics such as credit card and statement information, the system cannot accurately locate the user's historical credit card transaction data when subsequently linking it. It needs to repeatedly retrieve the user's full transaction records for matching. The encoding and data association process for a single user call requires a significant amount of additional server computing power, and may even cause user dissatisfaction due to excessively long matching times.

[0060] Furthermore, in requests sent to the digital human engine, business parameters and session information are not strongly correlated, leading to frequent mismatches. Taking the example of a user querying their credit card statement, the existing system's generated digital human service request only contains the vague parameter that the user needs to query a business, without binding it to session information such as the user's terminal identifier and credit card business attributes. When the request is sent to the digital human interaction service engine, the engine, lacking clear correlation criteria, may mistakenly match a digital human model specializing in mortgage business to the user. This results in the digital human being unable to directly answer credit card questions after connection, requiring the system to regenerate a request containing credit card domain parameters and resend it. This leads to a high cross-module data retransmission rate, severely impacting response speed. Users often have to wait on the phone for the system to re-transfer, resulting in a very poor user experience.

[0061] In addition, the inconsistent data packet formats of digital human audio and video streams make it difficult to balance efficiency and quality in compression processing. For example, in the scenario mentioned above, when the system finally matches the digital human for credit card transactions, the voice stream output by the digital human is encapsulated in MP3 format, while the video stream is encapsulated in AVI format, resulting in a messy data packet format. If processed according to a uniform compression rule, the MP3 format voice stream has limited compression effect due to poor compression targeting, resulting in excessive bandwidth consumption; while the AVI format video stream may experience stuttering, frame loss, and the user seeing the digital human's lips out of sync with the voice, or even the image freezing, making it impossible to balance transmission efficiency and interactive experience.

[0062] Example 1

[0063] To address the aforementioned issues, this embodiment provides an intelligent call processing method integrating digital human interaction, comprising the following steps:

[0064] Step 1: The intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and the current service type information, and generates a unique session status code.

[0065] The intelligent call hub integrates functions such as call reception, data processing, protocol encapsulation, and data transmission. It adopts a distributed architecture and supports concurrent processing by multiple users. User terminal identification information refers to the characteristic information used to uniquely identify user terminals, including the IMEI code of mobile terminals, the MAC address of PCs, and the device serial number of smart devices. Current service type information refers to the core demand identifier of the user initiating the call, such as credit card bill inquiry or mortgage application consultation in financial customer service scenarios, which is stored in the form of preset codes (such as 001-credit card, 002-mortgage). Session state code refers to a unique string generated based on terminal identification and service type, which is used for context association throughout the entire call process.

[0066] In the specific implementation process, the intelligent call center receives call content initiated by user terminals through the WebSocket communication protocol. This protocol supports full-duplex communication to ensure the real-time transmission of call content. It collects user terminal identification information and current service type information through the terminal communication interface. The terminal identification information is obtained through active reporting by the terminal, and the service type information is obtained by parsing the user call content through voice recognition. The collected information is processed for format standardization, such as retaining valid fields of the terminal identification, converting the service type into a preset code, and generating a unique session state code based on a hash algorithm. The code length is fixed at 32 bits to balance uniqueness and storage efficiency.

[0067] Step 2: The intelligent call center associates the user's historical business data with the session state code, generates a digital human service request containing the session state code and business adaptation parameters, and sends it to the digital human interaction service engine.

[0068] Among them, the user's historical business dataset refers to the storage of the user's past business records in the intelligent call system, including business type, processing time, interaction content, business results, etc., which are stored in a distributed database in time-series form; business adaptation parameters are feature parameters used to match the digital human model, such as the digital human's professional field corresponding to the business type, the user's preferred digital human image, interaction speech speed, etc.; digital human service request refers to the resource call instruction initiated by the digital human interaction service engine, and the resource call instruction includes session state encoding, business adaptation parameters and timestamp.

[0069] In the specific implementation process, the intelligent call center uses the session state code as an index to query the index table of the distributed database, quickly retrieves the user's historical business dataset, and avoids full data scanning. The historical business data is filtered to retain valid data related to the current business type, and user business preferences are extracted based on the valid data. Business adaptation parameters are generated by combining the current business type and user preferences. The session state code, business adaptation parameters, and current timestamp are integrated to generate a digital human service request. The service request is sent to the digital human interaction service engine via the HTTP / 2 protocol, which supports multiplexing and improves request transmission efficiency.

[0070] Step 3: The intelligent call center receives the customized digital human model association information returned by the digital human interaction service engine and encapsulates the digital human audio and video stream data into standardized data packets.

[0071] Customized digital human model association information refers to the matching results returned by the digital human interaction service engine, including digital human model ID, audio and video stream bitrate, frame rate, resolution, interaction permissions, etc.; standardized data packets refer to audio and video stream data carriers encapsulated according to a unified protocol, including a protocol header, data body, and checksum tail to ensure format consistency; and dedicated context transmission protocol is used to standardize data packet field formats and association rules.

[0072] In the specific implementation process, the intelligent call center receives customized digital human model association information returned by the digital human interaction service engine, extracts basic parameters of audio and video streams such as bitrate, frame rate, and resolution through the JSON format parsing module, defines four basic fields—session state encoding, business execution status, core business data, and timestamp—in a dedicated context transmission protocol, and sets field verification bits, and encapsulates the digital human audio and video stream data in the order of protocol header-core business data-verification tail, where the verification tail contains data integrity verification information, and generates standardized data packets.

[0073] Step 4: The intelligent call center compresses the standardized data packets to complete the lightweight transmission of digital human interaction data.

[0074] Lightweight transmission refers to a transmission method that reduces data packet size and transmission bandwidth usage through efficient compression and link optimization while ensuring the quality of audio and video interaction. Lightweight compression algorithm refers to the combination of H.265 video compression algorithm and OPUS audio compression algorithm used in this embodiment, which balances compression ratio and quality.

[0075] In the specific implementation process, the intelligent call center calls the H.265 video compression algorithm to compress the video stream in the standardized data packets, and removes video redundancy through three steps: intra-frame prediction, transform coding, and entropy coding; it calls the OPUS audio compression algorithm to compress the audio stream. This algorithm supports adaptive bitrate and can dynamically adjust the compression intensity according to the current network bandwidth; the compressed audio and video streams are integrated and transmitted to the user terminal through QoS (Quality of Service) optimized links. The links adopt traffic priority scheduling, with digital human interaction data having higher priority than ordinary data.

[0076] Considering the technical problems in Step 1 of Embodiment 1, such as redundant encoding information due to the lack of core feature extraction, lack of reverse tracing capability due to the lack of correlation between encoding and original information, and poor cross-module adaptability due to the lack of unified encoding rules, in an optional implementation, the intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and current service type information, and generates a unique session state code including the following:

[0077] The intelligent call center extracts the core feature parameters of user terminal identification information and current service type information to construct a feature parameter matrix; it reduces the dimension of the feature parameter matrix through a feature fusion algorithm to generate a low-dimensional core feature vector; it generates a core information condensation code based on the low-dimensional core feature vector and simultaneously constructs a code reverse verification mapping table; it performs redundant field stripping based on the code reverse verification mapping table, removes feature fields with invalid mappings, and outputs a unique and reverse-verifiable session state code.

[0078] Among them, the core feature parameters refer to the key information extracted from the terminal identifier and service type information, such as the last 12 digits of the IMEI code and the preset code of the service type, etc., after removing irrelevant and redundant fields; the feature parameter matrix refers to the two-dimensional data structure formed by arranging the core feature parameters in rows or columns, which is used for the structured processing of features; the encoding reverse verification mapping table stores the correspondence between the session state code and the original core feature parameters, with the key being the code and the value being the original parameter, which is used for the reverse tracing of the code.

[0079] In the specific implementation process, the user terminal IMEI code is 861234567890123, and the business type is credit card bill inquiry (preset code 001, core keywords are bill and inquiry): Extract the core feature parameters: the last 12 digits of IMEI 234567890123, business code 001, and keyword mapping values ​​10 (bill) and 20 (inquiry); construct the feature parameter matrix A=[[234567890123,001,10,20]]; process it with the feature fusion algorithm to generate a low-dimensional core feature vector [234567890123,001,30] (30 is the result of keyword mapping value fusion); generate the condensed code 5D8F2A7C9E1B3D5F7H9J1K3L5M7N9P1Q, and verify the original core feature parameters corresponding to this code in the reverse verification mapping table; strip the first 6 digits of the unrelated IMEI and output the final session status code.

[0080] Considering that in step 2 of embodiment 1, only the encoding is used as an index to retrieve the user's full historical data without verifying the data's integrity and authenticity, in actual applications, some historical data may be missing fields due to storage errors or tampered with due to system vulnerabilities (such as incorrect bill amounts). This invalid data will directly lead to errors in subsequent parameter generation, such as generating installment recommendation parameters based on incorrect bill history, triggering user complaints. In an optional implementation, the intelligent call center associates user historical business data based on session state encoding, including the following:

[0081] The intelligent call center retrieves the user's historical business dataset through session state coding and constructs a historical data time-series queue. It uses a data integrity verification algorithm to verify the historical data in the time-series queue frame by frame and marks the valid data segments that pass the verification. It uses cosine similarity matching between the business scenario label vector and the valid data segments to filter target data whose adaptability meets the threshold. It generates business adaptation parameters based on the target data and writes outdated fields into the data update log to drive incremental updates of the historical business dataset.

[0082] Among them, the historical data time-series queue is a queue formed by arranging the retrieved user historical business data in ascending order according to the processing timestamp, which facilitates the filtering of valid data by time dimension; the data integrity verification algorithm is an algorithm used to verify whether the historical data is complete and has not been tampered with, and this embodiment adopts the CRC32 algorithm; the cosine similarity matching is an algorithm used to calculate the similarity between the current business and the historical data. The larger the similarity value, the higher the adaptability. The similarity formula is: cosθ=(a·b) / (|a|·|b|), where a is the current business label vector and b is the historical data label vector.

[0083] For example, the user historical data retrieved based on session state coding includes credit card bill inquiry in October 2025, credit card limit adjustment in September 2025, and mortgage consultation in August 2024. These are arranged in ascending order to form a time-series queue: Mortgage Consultation in August 2024 → Credit Card Limit Adjustment in September 2025 → Credit Card Bill Inquiry in October 2025. After CRC32 verification, all three data entries are valid. The current business label vector a = [001, Bill, Inquiry], and the label vector b1 = [001, Bill, Inquiry] for the historical data 2025-10 credit card bill inquiry, are calculated to have cosθ1 = 0.92 ≥ 0.8, thus identifying it as target data. Based on the target data, the following parameters are generated: Digital Human Domain: Credit Card; Image: Female; Speech Rate: 110 words / minute. The 2024-08 mortgage consultation is written to the update log and marked as data to be cleaned.

[0084] Considering that the matrix processing method using the feature fusion algorithm does not specify whether it uses matrix factorization, feature concatenation, or weighted summation, developers may adopt different methods due to differences in understanding during actual development. For example, simple weighted summation can lead to an imbalance in feature weights between terminal identifiers and business codes, resulting in a vector that fails to reflect the priority of core information and ultimately causing mismatches when encoding associated data. In an optional implementation, the step of reducing the dimensionality of the feature parameter matrix using the feature fusion algorithm to generate a low-dimensional core feature vector includes the following:

[0085] Singular value decomposition is performed on the feature parameter matrix to extract the principal singular values ​​of the matrix to form the initial feature vector; the initial feature vector is smoothed by the Laplacian operator to eliminate feature noise; Huffman coding is used to compress the smoothed feature vector to generate a low-dimensional core feature vector, with the vector dimension mapped one-to-one with the business type.

[0086] Singular Value Decomposition (SVD) is a matrix factorization algorithm used to extract the principal singular values ​​of a matrix, achieving dimensionality reduction; the Laplace operator is a second-order differential operator used to smooth eigenvectors and eliminate noise; Huffman coding is a lossless compression coding based on character frequency, used to further compress the length of eigenvectors.

[0087] In the specific implementation process, singular value decomposition is performed on the feature parameter matrix A constructed in Example 2. The decomposition formula is A=UΣV^T, where A is an m×n feature parameter matrix, U is an m×m left singular matrix, Σ is an m×n singular value diagonal matrix, and V^T is an n×n right singular matrix transpose. The first k largest singular values ​​in Σ (k is 5, satisfying variance contribution ≥90%) are extracted to form the initial feature vector.

[0088] The initial eigenvector is smoothed using the Laplacian operator, with the formula ∇²f(x,y)=[f(x+1,y)+f(x-1,y)+f(x,y+1)+f(x,y-1)-4f(x,y)] / h², where f(x,y) is the value of the eigenvector at position (x,y), and h is the smoothing step size (set to 1), eliminating abnormal fluctuations in the eigenvector.

[0089] After statistical smoothing, the frequency of each feature value in the feature vector is counted, and a Huffman tree is constructed. High-frequency feature values ​​are represented by short codes, and low-frequency feature values ​​are represented by long codes, generating a low-dimensional core feature vector. The vector dimension is mapped one-to-one with the business type (e.g., credit card business corresponds to dimension 5, and mortgage business corresponds to dimension 6).

[0090] Considering the technical problems in practical applications of step 3 of Embodiment 1, such as the lack of a unified standard for protocol field definitions leading to format chaos, non-standard encapsulation order and verification logic leading to poor adaptability, weak fault tolerance of a single verification mechanism leading to missed data errors, and lack of constraints on field associations leading to a disconnect between business status and data, in an optional implementation, the intelligent call center receives customized digital human model association information returned by the digital human interaction service engine, and encapsulates the digital human audio and video stream data into standardized data packets including the following:

[0091] The system analyzes the associated information of the customized digital human model and extracts basic parameters such as bitrate, frame rate, and resolution of the audio and video streams. It constructs a dedicated context transmission protocol, defines four basic protocol fields: session state encoding, business execution status, core business data, and timestamp, and sets field check bits. It embeds a fault-tolerant redundancy field at the protocol layer, which stores the MD5 checksum of the core business data, forming a dual check mechanism with the core business data field. It encapsulates the audio and video stream data and basic parameters in the order of protocol fields to generate standardized data packets with dual check mechanisms.

[0092] The MD5 checksum is a 128-bit hash value calculated from the core business data using the MD5 hash algorithm, used to verify data integrity. The dual verification mechanism is a verification mechanism composed of the check bits of the core business data field itself and the MD5 checksum, which improves the ability to identify data errors.

[0093] In the specific implementation process, the customized digital human model association information is parsed, and basic parameters such as bitrate, frame rate, and resolution of audio and video streams are extracted to determine the basic attributes of data packets. Four basic protocol fields are defined: session state encoding, business execution status, core business data, and timestamp, and fixed byte length and check bits are set for each field. The MD5 checksum of the core business data is calculated using the formula MD5(message) = 128-bit hash value. This field serves as a fault-tolerant redundancy field and forms a dual check mechanism with the core business data field. Audio and video stream data and basic parameters are encapsulated in the order of protocol header - core business data - fault-tolerant redundancy field - checksum tail to generate standardized data packets with a dual check mechanism.

[0094] Considering the technical problems in practical applications of step 2 in Embodiment 1, such as the lack of standardized parameter templates leading to chaotic generation logic, reliance on historical data without considering real-time behavior resulting in poor adaptability, lack of parameter denoising leading to large fluctuations, and lack of a dynamic update mechanism causing parameters to lag behind user habits, in an optional implementation, the generation of a digital human service request containing session state codes and business adaptation parameters includes the following:

[0095] The system includes: a parameter generation template library (a collection of parameter templates categorized by business type, such as credit card business templates and mortgage business templates, containing basic parameter configurations); a real-time interactive behavior collector (a module for capturing user interaction behavior data during calls); and a Kalman filter algorithm (an algorithm for smoothing and denoising dynamic parameters, improving parameter stability).

[0096] In the specific implementation process, based on the session state encoding parsing of the business type identifier, the corresponding business template in the template library is generated by calling the parameters (such as the credit card business template, with basic parameters of speech rate of 100 words / minute and neutral image); user interaction behavior data (such as speech rate, response interval, etc.) is captured by the real-time interactive behavior collector, and the behavioral feature time series is extracted; the behavioral feature time series is input into the parameters to generate the template library, and the template parameters are dynamically adjusted by the linear interpolation algorithm to make the parameters fit the user's behavioral habits; the adjusted parameters are smoothed and denoised by the Kalman filter algorithm to output stable business adaptation parameters. The state equation of the Kalman filter is X(k)=AX(k-1)+BU(k)+W(k), and the observation equation is Z(k)=HX(k)+V(k), where A=1, B=0, H=1, and the process noise W(k) and observation noise V(k) are both taken as 0.01.

[0097] Considering the technical problems existing in the practical application of the dedicated context transport protocol in Embodiment 5, such as the lack of quantitative standardization of protocol field formats leading to parsing ambiguity, the lack of triggering rules for field linkage leading to timing disorder, the lack of logical constraints leading to state mismatch, and the lack of a standard process for protocol integration leading to implementation difficulties, in an optional implementation, the construction of the dedicated context transport protocol includes the following:

[0098] Define the fixed byte length and binary storage format of the protocol fields and construct a field format dictionary; establish linkage update rules between core business data fields and timestamp fields, with field modifications triggering automatic timestamp rewriting; construct logical constraint equations between business execution status fields and core business data fields to limit the correlation of field values; integrate the field format dictionary, linkage update rules, and constraint equations to form a dedicated context transmission protocol.

[0099] Among them, the field format dictionary records the fixed byte length, binary storage format, and data type of each field in the protocol, and is used to standardize field definitions; the linkage update rules define the association update rules between the core business data fields and the timestamp fields; and the logical constraint equations are mathematical equations that limit the correlation between the values ​​of the business execution status fields and the core business data fields.

[0100] In the specific implementation process, the fixed byte length of each field in the protocol is set (32 bytes for session state encoding, 1 byte for business execution status, variable length for core business data, and 8 bytes for timestamp) and the binary storage format (big-endian), and the data type of each field (string, integer, byte stream, etc.) is defined to form a field format dictionary. When the core business data field is modified, the timestamp field is automatically rewritten to the current time to ensure the correlation between data and time. The value of the business execution status field is set to be determined by the core business data field, and the constraint equation is business execution status = f(core business data). For example, if the core business data is a successful query, the business execution status is encoded as 1; if the core business data is in the process of querying, the encoding is 0. The field format dictionary, linkage update rules, and logical constraint equations are integrated to form a complete dedicated context transmission protocol for data packet encapsulation.

[0101] Considering the issue of the lack of breakpoint protection mechanism for data packets in Embodiment 5, in an optional implementation, after generating the standardized data packet with double verification mechanism, the following content is also included:

[0102] The intelligent call hub captures call interruption events through the call status monitoring module, generates breakpoint markers, synchronously calls the generated standardized data packets with double verification mechanisms, and extracts the real-time synchronized business progress information throughout the entire process of generating the data packet. It then constructs a breakpoint data snapshot containing both the standardized data packet and the business progress information. A breakpoint business urgency evaluation index system is built, using core business data in the standardized data packet as the core evaluation dimension, and combining it with business progress information to complete the priority quantification scoring of the breakpoint data snapshot. Based on the scoring results, a dynamic hierarchical storage architecture is constructed, with a dedicated hot data area storing high-priority breakpoint data snapshots and a cold data area storing low-priority breakpoint data snapshots. The snapshots fully retain the association mapping between the standardized data packet and the business progress information. A repeated call trigger threshold is configured; when the call frequency of the same user terminal reaches the threshold, the migration of the breakpoint data snapshot in the cold data area is triggered, maintaining the association mapping between the standardized data packet and the business progress information unchanged during the migration process.

[0103] Among them, the call status monitoring module is used to monitor the call connection status in real time, with a sampling frequency of 5 times / second, which can quickly capture interruption events; breakpoint data snapshot: a dataset containing standardized data packets and business progress information at the time of interruption, used for subsequent business continuation; dynamic hierarchical storage architecture: a storage architecture containing dedicated hot data area and cold data area, which allocates storage resources according to data priority.

[0104] In the specific implementation process, the intelligent call center monitors the call status in real time through the call status monitoring module. When a call connection interruption is detected, a breakpoint marker is generated, and the interruption time and terminal identifier are recorded. The generated standardized data packet with dual verification mechanism is called synchronously, and the business progress information synchronized in real time throughout the entire process of generating the data packet (such as the bill amount has been broadcast, but the repayment date has not been broadcast) is extracted to construct a breakpoint data snapshot containing the standardized data packet and business progress information. A breakpoint business urgency evaluation index system is constructed, with the importance of core business data (weight 0.6) and business progress (weight 0.4) as evaluation dimensions to complete the priority quantitative scoring of the breakpoint data snapshot (maximum score 10 points). Based on the scoring results, high-priority (score ≥ 8 points) breakpoint data snapshots are stored in a dedicated hot data area (high-speed storage), and low-priority (score < 8 points) are stored in a cold data area. A repeated call trigger threshold (such as 2 times) is configured. When the call frequency of the same user terminal reaches the threshold, the cold data area snapshot is triggered to migrate to the hot data area.

[0105] Considering the low efficiency and inaccurate verification of breakpoint recovery in Implementation Example 8, in an optional implementation, a repeated call trigger threshold is configured. When the frequency of calls from the same user terminal reaches the threshold, the migration of the cold data area breakpoint data snapshot is triggered. During the migration process, the association mapping between standardized data packets and service progress information remains unchanged, including the following:

[0106] The system captures call requests initiated by user terminals and extracts terminal identification features and call timestamps from the requests. Based on the terminal identification features, it queries the breakpoint data index and preloads the breakpoint data snapshot corresponding to the terminal into the cache buffer. Through a two-dimensional judgment logic of terminal identification matching and call time window verification, it completes the identity verification for connecting calls. After the verification is successful, it directly calls the preloaded breakpoint data snapshot in the cache buffer and drives the digital human interaction service engine to continue the business process based on the standardized data packets and business progress information therein.

[0107] Among them, the fast matching judgment module is used to quickly capture call requests and match breakpoint data, with a response time of ≤10ms; the cache buffer is a high-speed cache area used to preload breakpoint data snapshots, with a read and write speed 10 times that of ordinary storage; and the call time window verification is a verification mechanism to verify whether the time interval between the second call and the interruption is within a reasonable range (such as within 30 seconds).

[0108] In the specific implementation process, the high-speed matching and judgment module captures the secondary call request initiated by the user terminal, extracts the terminal identification features and call timestamp from the request, queries the breakpoint data index based on the terminal identification features, and preloads the breakpoint data snapshot corresponding to the terminal into the cache buffer in advance. The identity verification is completed through a two-dimensional judgment logic of terminal identification matching + call time window verification - if the terminal identification matches the snapshot terminal identification and the time interval between the call time and the interruption is ≤30 seconds (time window threshold), the verification is passed. After the verification is passed, the preloaded breakpoint data snapshot in the cache buffer is directly called. Based on the standardized data packets and business progress information in it, the digital human interaction service engine is driven to continue the business process without having to re-execute the data query and parameter generation steps.

[0109] Example 2 provides an intelligent call processing method integrating digital human interaction, applying the intelligent call processing method integrating digital human interaction described in Example 1 above, such as... Figure 2 As shown, it includes:

[0110] The user terminal is used to initiate a call to the intelligent call center, transmit its own terminal identification information and current service type information, and receive digital human interactive audio and video stream data fed back by the intelligent call center.

[0111] The intelligent call hub receives call content initiated by a user terminal, collects and parses terminal identification information and current service type information, and generates a unique session state code. Based on the session state code, it associates the user's historical service data, generates a digital human service request containing the session state code and service adaptation parameters, and sends it to the digital human interaction service engine. It receives customized digital human model association information returned by the digital human interaction service engine, encapsulates the digital human audio and video stream data into standardized data packets, compresses the standardized data packets, and transmits lightweight digital human interaction data to the user terminal.

[0112] The digital human interaction service engine is used to receive digital human service requests sent by the intelligent call center, match the corresponding digital human model resources based on the session state code and business adaptation parameters in the request, generate customized digital human model association information, and feed it back to the intelligent call center.

[0113] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent call processing integrating digital human interaction, characterized in that, Includes the following: The intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and the current service type information, and generates a unique session status code; The intelligent call center associates user historical business data with session state coding, generates a digital human service request containing session state coding and business adaptation parameters, and sends it to the digital human interaction service engine. The intelligent call center receives customized digital human model association information returned by the digital human interaction service engine and encapsulates the digital human audio and video stream data into standardized data packets; The intelligent call center compresses standardized data packets to achieve lightweight transmission of digital human interaction data; The intelligent call center receives the call content initiated by the user terminal, collects the user terminal identification information and the current service type information, and generates a unique session status code including the following: The intelligent call center extracts the core feature parameters of user terminal identification information and current service type information. The core feature parameters are key information extracted from terminal identification and service type information. Irrelevant and redundant fields are removed, and the core feature parameters are arranged in rows or columns to form a feature parameter matrix. The feature parameter matrix is ​​reduced in dimension by a feature fusion algorithm to generate a low-dimensional core feature vector. Core information condensation encoding is generated based on low-dimensional core feature vectors, and a reverse verification mapping table for encoding is constructed simultaneously. Based on the encoding reverse verification mapping table, redundant fields are stripped, feature fields with invalid mappings are removed, and a unique and reversibly verifiable session state code is output. The intelligent call center compresses standardized data packets to achieve lightweight transmission of digital human interaction data, including the following: The intelligent call center compresses standardized data packets and transmits them to the user terminal through a QoS-optimized link. The link uses traffic priority scheduling, with digital human interaction data having higher priority than ordinary data, thus completing the lightweight transmission of digital human interaction data.

2. The intelligent call processing method integrating digital human interaction according to claim 1, characterized in that, The intelligent call center associates user historical service data with session state coding, including the following: The intelligent call center retrieves the user's historical business dataset through session state coding and constructs a historical data time-series queue. A data integrity verification algorithm is used to verify the historical data in the time-series queue frame by frame, and the valid data segments that pass the verification are marked. By performing cosine similarity matching between business scenario label vectors and effective data segments, target data whose suitability meets the threshold is selected. Business adaptation parameters are generated based on the target data, and outdated fields are written to the data update log to drive incremental updates of the historical business dataset.

3. The intelligent call processing method integrating digital human interaction according to claim 1, characterized in that, The step of reducing the dimensionality of the feature parameter matrix using a feature fusion algorithm to generate a low-dimensional core feature vector includes the following: Perform singular value decomposition on the feature parameter matrix and extract the principal singular values ​​of the matrix to form the initial eigenvectors; The initial eigenvectors are smoothed using the Laplacian operator to eliminate feature noise; Huffman coding is used to compress the smoothed feature vectors to generate low-dimensional core feature vectors, with the vector dimensions mapped one-to-one with the business type.

4. The intelligent call processing method integrating digital human interaction according to claim 1, characterized in that, The intelligent call center receives customized digital human model association information returned by the digital human interaction service engine, and encapsulates the digital human audio and video stream data into standardized data packets including the following: Analyze the associated information of the customized digital human model and extract the basic parameters of the audio and video streams, including bit rate, frame rate, and resolution; Build a dedicated context transport protocol, define four basic protocol fields: session state encoding, business execution status, core business data, and timestamp, and set field validation bits; A fault-tolerant redundancy field is embedded in the protocol layer. This field stores the MD5 checksum of the core business data, forming a dual check mechanism with the core business data field. The audio and video stream data and basic parameters are encapsulated in the order of protocol fields to generate standardized data packets with double verification mechanism.

5. The intelligent call processing method integrating digital human interaction according to claim 1, characterized in that, The process of generating a digital human service request that includes session state encoding and business adaptation parameters includes the following: Based on the session state encoding, the business type identifier is parsed, and the corresponding parameters are called to generate a template library; User interaction behavior data is captured by a real-time interactive behavior collector, and the time sequence of behavioral features is extracted. The behavioral feature time series is input as a parameter to generate a template library, triggering dynamic interpolation adjustment of the template parameters; The adjusted parameters are smoothed and denoised using the Kalman filter algorithm to output service adaptation parameters.

6. The intelligent call processing method integrating digital human interaction according to claim 4, characterized in that, The construction of the dedicated context transport protocol includes the following: Define the fixed byte length and binary storage format of the protocol fields, and construct a field format dictionary; Establish a linkage update rule between core business data fields and timestamp fields, so that field modifications trigger automatic timestamp rewriting; Construct logical constraint equations between business execution status fields and core business data fields to limit the correlation of field values; Integrate field format dictionaries, linked update rules, and constraint equations to form a dedicated context transmission protocol.

7. The intelligent call processing method integrating digital human interaction according to claim 4, characterized in that, After generating the standardized data packet with double verification mechanism, the following content is also included: The intelligent call center captures call interruption events through the call status monitoring module, generates breakpoint markers, synchronously calls the generated standardized data packets with double verification mechanism, and extracts the real-time synchronized business progress information throughout the entire process of generating the data packet, and constructs a breakpoint data snapshot containing standardized data packets and business progress information. Construct an evaluation index system for the urgency of breakpoint business, using core business data in standardized data packages as the core evaluation dimension, and combine business progress information to complete the priority quantitative scoring of breakpoint data snapshots; A dynamic hierarchical storage architecture is built based on the scoring results. A dedicated hot data area stores high-priority breakpoint data snapshots, and a cold data area stores low-priority breakpoint data snapshots. The snapshots fully retain the association mapping between standardized data packets and business progress information. Configure a repeated call trigger threshold. When the frequency of calls from the same user terminal reaches the threshold, the migration of the cold data area interruption point data snapshot is triggered. During the migration process, the association mapping between standardized data packets and service progress information remains unchanged.

8. The intelligent call processing method integrating digital human interaction according to claim 7, characterized in that, Configure a repeated call trigger threshold. When the frequency of calls from the same user terminal reaches the threshold, the migration of the cold data area interruption point data snapshot is triggered. During the migration process, the association mapping between standardized data packets and service progress information remains unchanged, including the following: Capture call requests initiated by user terminals and extract terminal identification features and call timestamps from the requests; Based on the terminal identification features, query the breakpoint data index library and preload the breakpoint data snapshot corresponding to the terminal into the cache buffer in advance. The system uses a two-dimensional judgment logic of terminal identifier matching and call time window verification to complete the identity verification for connecting calls. After the verification is successful, it directly calls the breakpoint data snapshot preloaded in the cache buffer, and drives the digital human interaction service engine to continue the business process based on the standardized data packets and business progress information therein.

9. An intelligent call processing system integrating digital human interaction, characterized in that, The intelligent call processing method integrating digital human interaction according to any one of claims 1-8 includes: The user terminal is used to initiate a call to the intelligent call center, transmit its own terminal identification information and current service type information, and receive digital human interactive audio and video stream data fed back by the intelligent call center. The intelligent call hub receives call content initiated by a user terminal, collects and parses terminal identification information and current service type information, and generates a unique session state code. Based on the session state code, it associates the user's historical service data, generates a digital human service request containing the session state code and service adaptation parameters, and sends it to the digital human interaction service engine. It receives customized digital human model association information returned by the digital human interaction service engine, encapsulates the digital human audio and video stream data into standardized data packets, compresses the standardized data packets, and transmits lightweight digital human interaction data to the user terminal. The digital human interaction service engine is used to receive digital human service requests sent by the intelligent call center, match the corresponding digital human model resources based on the session state code and business adaptation parameters in the request, generate customized digital human model association information, and feed it back to the intelligent call center.

Images