Call quality optimization method and device based on voiceprint recognition and storage medium

Abstract

The invention discloses a call quality optimization method and device based on voiceprint recognition and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: responding to a service request, obtaining the initial voice data of a customer in a call, and extracting the voiceprint features of the customer; comparing the customer voiceprint characteristics with a customer voiceprint template in an identity information base, and determining target customer information corresponding to the customer voiceprint characteristics; based on the target customer information and the customer service call voice data, analyzing a business scene required by the target customer, and generating a corresponding exclusive service strategy; the method comprises the following steps: automatically labeling an emotion label and pushing a keyword and a service adjustment strategy according to an emotion analysis result of a target customer; and optimizing the exclusive service strategy according to the service adjustment strategy to obtain a target service strategy corresponding to the target customer. According to the method, the identity is confirmed through the voiceprint, and the scene emotion adapts to the optimization service, so that the problem of poor communication quality is solved, and the effect of improving the quality and efficiency of the whole call service process is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and particularly relates to a call quality optimization method, device and storage medium based on voiceprint recognition. BACKGROUND

[0002] In the telephone customer service scene of the securities industry, confirming the identity of a user is a key prerequisite for guaranteeing the safety of account funds, complying with industry regulations to handle transaction inquiries, business changes and other core operations, and is also a core link for preventing account impersonation risks and meeting industry regulatory requirements.

[0003] In related technologies, after a customer service personnel answers a user's call, the user's pre-set identity verification information is verified through oral inquiry, and after identity confirmation, subsequent business can be continued. This method relies on manual inquiry and subjective judgment to complete identity verification throughout the process, and in the process of calling up user information for checking, errors are inevitable and checking information takes too long, resulting in poor communication quality.

[0004] The above content is only used to assist in understanding the technical solutions of the present application and does not represent the acknowledgement of the above content as prior art. SUMMARY

[0005] The main purpose of the present application is to provide a call quality optimization method, device and storage medium based on voiceprint recognition, aiming to solve the technical problem of poor communication quality.

[0006] To achieve the above purpose, the present application provides a call quality optimization method based on voiceprint recognition, which comprises: In response to a service request of a user call, initial speech data of a customer in a call process is obtained, and customer voiceprint features in the initial speech data are extracted; An identity information library is called, the customer voiceprint features are compared and identified with customer voiceprint templates in the identity information library, and target customer information corresponding to the customer voiceprint features is determined; Based on the target customer information and customer service speech data, the business scene of target customer demand is analyzed, and an exclusive service strategy corresponding to the target customer is generated; Through the emotional analysis result of the target customer, an emotional label is automatically labeled and key words and service adjustment strategies are pushed; The exclusive service strategy is optimized according to the service adjustment strategy, and a target service strategy corresponding to the target customer is obtained.

[0007] In an embodiment, based on the service request of a customer call, the voice information of the initial stage of the customer call is captured through the audio acquisition component of the call terminal, and the raw speech data stream without processing is generated; The original voice data stream is denoised, dereverberated and format standardized, and environmental interference signals and invalid audio segments are filtered out to output pure initial voice data. The pure initial voice data is analyzed by using a preset voiceprint feature extraction algorithm, and voiceprint feature parameters unique in the pure initial voice data are extracted to form the customer voiceprint feature.

[0008] In an embodiment, a call instruction of the identity information library is triggered, and the customer voiceprint template corresponding to all registered customers is extracted from the identity information library. According to a preset voiceprint similarity comparison algorithm, the customer voiceprint feature is matched with each customer voiceprint template one by one to obtain a matching result list containing similarity values of each matching item. According to a preset similarity threshold, the optimal matching item in the matching result list is screened, and the customer identity data corresponding to the optimal matching item is associated and called from the identity information library to generate the target customer information.

[0009] In an embodiment, the target customer information and the customer service call voice data are structured and integrated, customer historical service records and key information representing the demand tendency in the call are extracted, and a customer demand related information set is output after integration. A preset business scenario recognition algorithm is used to classify and match the scenario features of the customer demand related information set, determine the specific business scenario type corresponding to the customer demand, and generate a corresponding business scenario identifier. Based on the business scenario identifier and the customer demand related information set, a preset service strategy rule library is combined to match the customer historical service preference and the business scenario adaptation requirement, and the exclusive service strategy is generated.

[0010] In an embodiment, the emotion intensity, emotion fluctuation amplitude and core emotion expression segment in the target customer corresponding emotion analysis result are extracted to obtain customer emotion feature data. According to a preset emotion label classification rule, the customer emotion feature data is matched and determined to automatically label the corresponding emotion label, and a customer emotion analysis result with an emotion label is obtained. The call core keywords triggering the corresponding emotion in the customer emotion analysis result are extracted, a preset emotion and service adjustment strategy mapping library is combined, and a targeted suggestion adapted to the current emotion state is generated. According to the targeted suggestion, associated sentence data and communication methods are screened and combined to generate the service adjustment strategy.

[0011] In an embodiment, the core optimization requirement, priority and adaptation rule in the service adjustment strategy are extracted to output a strategy optimization parameter set. According to the preset strategy optimization priority rule, the strategy optimization parameter set is adapted to the exclusive service strategy module by module, priority judgment and adjustment are performed on the service items with conflicts, and an adapted preliminary optimization service strategy is obtained; The integrity and logic of the preliminary optimization service strategy are checked, missing service linkages are supplemented, and unreasonable service process settings are corrected, so as to generate the target service strategy.

[0012] In an embodiment, the service adjustment strategy and the exclusive service strategy are split according to preset modules of service process, response priority, and exclusive benefits, and the corresponding service items under each module are checked one by one for adaptability, so as to obtain a module comparison result; The conflict service items identified in the module comparison result are extracted, the priority level of the conflict service items is quantitatively judged and sorted in combination with the strategy optimization priority rule, and a conflict service item priority judgment result is output; Based on the conflict service item priority judgment result, the core configuration of the high-priority service item is retained, the parameters of the low-priority conflict service item are adjusted, and the service items of each module after adjustment and the service items of the non-conflict module are reorganized and integrated according to the original service framework, so as to obtain the adapted preliminary optimization service strategy.

[0013] In an embodiment, the customer feedback data and the quality inspection feedback data after execution of the target service strategy are classified, sorted and packaged, so as to obtain a feedback data set; A preset feedback analysis algorithm is used to deeply analyze the feedback data set, identify the core dimension and specific adjustment direction that need to be optimized in the next round of call, and integrate an optimization demand list; According to the optimization demand list, the voiceprint feature extraction parameters, the service strategy rule library, and the emotion and service adjustment mapping relationship are updated, and an optimization configuration parameter set dedicated for the next round of call is generated, so as to provide updated execution basis for quality optimization of subsequent customer calls.

[0014] In addition, to achieve the above purpose, the present application also proposes a call quality optimization device, which comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the call quality optimization method based on voiceprint recognition as described above.

[0015] In addition, to achieve the above purpose, the present application also proposes a storage medium, which is a computer readable storage medium, and the storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the call quality optimization method based on voiceprint recognition as described above.

[0016] The application provides a call quality optimization method based on voiceprint recognition, comprising the following steps: obtaining initial voice data of a customer in a call process and extracting customer voiceprint features from the initial voice data in response to a service request of the customer, calling an identity information library to identify and compare the voiceprint features with customer voiceprint templates in the library to determine target customer information, analyzing a business scenario of customer demand based on the target customer information and customer service call voice data and generating a dedicated service strategy, automatically labeling an emotional label and pushing a keyword and a service adjustment strategy through an emotional analysis result of the target customer, and finally optimizing the dedicated service strategy according to the service adjustment strategy to obtain a target service strategy, which solves technical problems such as low identity verification efficiency, inaccurate matching of customer demand and business scenarios, and insufficient service strategy targeting caused by untimely adaptation of emotional state in traditional call services, and improves the accuracy of customer identity recognition, the accuracy of business scenario matching, the personalized adaptation capability of service strategy, and the satisfaction of customer call service experience.

[0017] In summary, the application extracts customer voiceprint features in initial voice data in response to a customer call request, calls an identity information library to determine target customer information, analyzes a business scenario based on customer service call voice data to generate a dedicated service strategy, labels a label through customer emotional analysis and pushes a service adjustment strategy, and finally optimizes to obtain a target service strategy, which solves the technical problem of poor communication quality, improves the customer identity recognition speed, business scenario matching efficiency and service strategy personalized adaptation response speed, and realizes the effect of improving the whole process of call service. BRIEF DESCRIPTION OF DRAWINGS

[0018] The accompanying drawings, which are incorporated into and form a part of the specification, illustrate an embodiment consistent with the present application and, together with the description, serve to explain the principles of the application.

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the accompanying drawings needed in the embodiment or prior art description will be briefly introduced. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0020] Figure 1 The flowchart of the first embodiment of the call quality optimization method based on voiceprint recognition of the application; Figure 2 The overall flowchart of the application; Figure 3 The flowchart of the seventh embodiment of the call quality optimization method based on voiceprint recognition of the application; Figure 4 The flowchart of the eighth embodiment of the call quality optimization method based on voiceprint recognition of the application; Figure 5A structure schematic diagram of a call quality optimization device of the present application.

[0021] The object realization, functional features and advantages of the present application will be further explained in conjunction with the embodiments and by referring to the accompanying drawings. DETAILED DESCRIPTION

[0022] It should be understood that the specific embodiments described herein are merely intended to explain the technical solutions of the present application, and are not intended to limit the present application.

[0023] In the related art, after a customer service personnel answers a user's call, the user's preset identity verification information is verified through oral inquiry, and after identity confirmation, subsequent business can be continued. This method relies on manual inquiry and subjective judgment to complete identity verification throughout the process, and in the process of calling user information for checking, errors are inevitable and checking information takes too long, resulting in poor communication quality.

[0024] The present application provides a solution: first, in response to a service request from a user, initial voice data of a customer during a call is obtained, and customer voiceprint features in the initial voice data are extracted, then an identity information library is called, the customer voiceprint features are compared and identified with customer voiceprint templates in the identity information library, target customer information corresponding to the customer voiceprint features is determined, then based on the target customer information and customer service call voice data, the business scenario of the target customer's demand is analyzed, the exclusive service strategy corresponding to the target customer is generated, then through the emotional analysis result of the target customer, an emotional label is automatically labeled and keywords and service adjustment strategies are pushed, and finally the exclusive service strategy is optimized according to the service adjustment strategy to obtain the target service strategy corresponding to the target customer.

[0025] It should be noted that the execution subject of the present embodiment can be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a call quality optimization device, etc. The present embodiment and the following embodiments will be described below taking the call quality optimization device as an example.

[0026] In order to better understand the technical solutions of the present application, the following will be described in detail in conjunction with the drawings and specific embodiments of the present application.

[0027] The present embodiment provides a call quality optimization method based on voiceprint recognition, which is described in detail with reference to Figure 1 , Figure 1 A flowchart of the first embodiment of the call quality optimization method based on voiceprint recognition of the present application.

[0028] In the present embodiment, the call quality optimization method based on voiceprint recognition comprises steps S10-S50: Step S10, in response to the service request called by the user, obtaining the initial voice data of the customer in the call process, and extracting the customer voiceprint features in the initial voice data.

[0029] In this embodiment, the service request called by the customer is a request for service initiated by the customer through the telephone. The initial voice data is the unprocessed voice information issued by the customer at the beginning of the call. The customer voiceprint features are the voice feature parameters in the customer's voice that are unique and can be used for identity recognition.

[0030] As an optional implementation, after responding to the service request, the voice information at the initial stage of the call is continuously collected, and the collected voice information is simultaneously subjected to noise reduction and de-echo processing to filter out environmental interference signals and invalid audio segments. Then, the processed voice information is analyzed frame by frame using a preset feature extraction algorithm to extract feature parameters representing the identity uniqueness, and the analysis window size is dynamically adjusted to adapt to different voice rhythms during the extraction process. At the same time, the extracted feature parameters are real-time filtered to retain stable feature information, and the customer voiceprint features are output. This method has fast response speed and can quickly output effective voiceprint features at the initial stage of the call, providing basic data for subsequent identity comparison.

[0031] As another optional implementation, after responding to the service request, the voice information within a preset time duration at the initial stage of the call is first cached, and after the caching is completed, the overall voice data is subjected to multiple rounds of noise reduction processing to sequentially filter out steady-state noise and transient noise. Then, the voice segment is screened to remove silent and invalid repeated segments. Subsequently, a multi-dimensional feature extraction method is used to extract feature parameters related to frequency spectrum, tone, and speech rate from the screened voice data. After the extraction is completed, all feature parameters are subjected to effectiveness verification to remove abnormal features deviating from the normal range, and the customer voiceprint features are obtained. This method has high feature extraction accuracy, effectively reduces the influence of interference factors, and improves the reliability and accuracy of identity recognition results.

[0032] Step S20, calling the identity information library, comparing the customer voiceprint features with the customer voiceprint templates in the identity information library, and determining the target customer information corresponding to the customer voiceprint features.

[0033] In this embodiment, the identity information library is a database storing registered customer voiceprint templates and corresponding identity data. The customer voiceprint template is a set of standard customer voiceprint features pre-stored in the identity information library for identity comparison. The target customer information is the customer identity-related data corresponding to the customer voiceprint features determined by comparison.

[0034] As an optional implementation, after calling the identity information library, the core dimensional features in the customer voiceprint features are extracted first, all customer voiceprint templates in the library are preliminarily screened based on the core features, and the obviously unmatched templates are removed to reduce the comparison range. Then, the pre-set similarity calculation method is used to accurately compare each feature item of the screened voiceprint template and the customer voiceprint features, and the similarity values of each template are recorded. Finally, the template with the highest similarity is selected according to the pre-set threshold, the corresponding customer identity data of the template is associated and called, and the target customer information is determined. This method has high comparison efficiency and can quickly reduce the calculation amount.

[0035] As another optional implementation, after calling the identity information library, the full-dimensional information of the customer voiceprint features is extracted, and the full-dimensional information is directly compared with each customer voiceprint template in the identity information library in multiple dimensions one by one, and the matching degree of each feature item is quantitatively evaluated. Then, the overall similarity is calculated by comprehensively evaluating all feature items to generate a similarity ranking list. Finally, the voiceprint template with the highest matching degree is selected from the list according to the pre-set optimal matching rule, the corresponding customer identity data is associated and obtained, and the target customer information is determined. This method has strong comprehensive matching and can fully exploit feature detail differences.

[0036] Step S30, based on the target customer information and the customer service call voice data, analyzing the business scene of the target customer demand to generate the exclusive service strategy corresponding to the target customer.

[0037] In this embodiment, the customer service call voice data is voice information generated by the interaction between the customer service and the customer in the call process. The business scene of the target customer demand is the specific business type and related scene attributes corresponding to the service request initiated by the customer. The exclusive service strategy is a service execution scheme customized for the individual characteristics and demand scene of the customer.

[0038] As an optional implementation, the target customer information and the customer service call voice data are integrated, and key information such as customer historical business records, demand keywords, and interaction frequency is extracted and classified and labeled. A pre-set business scene classification rule library is called, and the labeled information is matched with the scene features in the rule library one by one to determine the business scene type corresponding to the customer demand. The basic template corresponding to the scene is called from the service strategy template library, and the template parameters are supplemented and adjusted in combination with the customer historical service preferences and exclusive benefit information to determine the service process, response priority, and core service content, and an exclusive service strategy is generated. This method has high adaptation efficiency, can quickly reuse mature templates to reduce generation cost, and ensures the timeliness of service response.

[0039] As another optional implementation, the target customer information is deeply analyzed to mine potential information such as customer historical service preferences, business handling habits, and unmet needs, and the customer service call voice data is analyzed for semantics to extract deep demand intentions and associated business appeals, and a complete customer demand portrait is constructed based on the two types of analysis results. Based on the customer demand portrait, the core attributes of the business scene are matched and the scene classification is refined, the service process is dynamically designed according to the scene attributes and the demand portrait, the response priority, exclusive benefits, and service connection mechanism are configured to generate an exclusive service strategy. This method has high individualization degree and can accurately match the special needs of customers, significantly improving the recognition and satisfaction of customers to the service.

[0040] In step S40, an emotional label is automatically labeled and keywords and service adjustment strategies are pushed based on the emotional analysis result of the target customer.

[0041] In this embodiment, the emotional analysis result of the target customer is related data obtained by analyzing the emotional state of the target customer during the call. The emotional label is a feature label used to standardize the identification of the customer's emotional type. The keywords are core voice contents extracted from the call voice that trigger the customer's emotions. The service adjustment strategy is a specific scheme for optimizing the service execution mode according to the customer's emotional state.

[0042] As an optional implementation, the emotional analysis result of the target customer is received, and emotional intensity, fluctuation characteristics, and core emotional expression related data are extracted from the emotional analysis result. These data are compared with preset emotional label classification rules one by one to automatically label the corresponding emotional label. Meanwhile, keywords are extracted from the call voice segment associated with the emotional analysis result to screen out core words directly related to emotional expression. Then, a preset emotional and service adjustment strategy mapping library is called to match the standardized service adjustment scheme corresponding to the current emotional label and keywords, and the label labeling and service adjustment strategy pushing are completed. This method has high execution efficiency and can quickly output standardized results, which is suitable for large-scale service scenarios.

[0043] As another optional implementation, the emotional analysis result of the target customer is multi-dimensionally disassembled to mine the deep causes of the emotions and the logical association of the emotional changes. Based on the disassembly result, the emotional label is accurately labeled in combination with a refined emotional classification system, and the call voice is analyzed for semantics to extract core keywords and associated semantic words, forming a complete set of emotional associated words. Then, a personalized service adjustment strategy is dynamically constructed based on the emotional label type, intensity, cause, and keyword set to adapt to the current emotional state, and the service dialogue optimization direction, response rhythm adjustment points, and exclusive benefit adaptation scheme are determined to complete the label labeling and service adjustment strategy pushing. This method has high individualization degree of strategy and can accurately match the deep needs of customer emotions, significantly improving the pertinence and effectiveness of the service strategy.

[0044] Step S50, the exclusive service strategy is optimized according to the service adjustment strategy to obtain a target service strategy corresponding to the target customer.

[0045] In this embodiment, the target service strategy is a final service execution scheme adapted to the customer demand and emotional state after the service adjustment strategy optimization.

[0046] As an optional implementation, the exclusive service strategy is first split into modules according to service flow, response priority, right configuration and other dimensions, and the core optimization points and adaptation rules of the service adjustment strategy are also split. The execution logic and parameter configuration of the corresponding modules of the two types of strategies are compared one by one to identify the modules with conflicts or needing optimization. The optimization modules in the service adjustment strategy are used to replace the corresponding modules with conflicts or needing optimization in the exclusive service strategy, and the original configuration of the non-conflicting modules is retained. Then, the replaced modules are interfaced and verified to ensure smooth data transmission and process transition between the modules, and finally the target service strategy is integrated. This method has a simple optimization process, high execution efficiency, and can quickly complete strategy iteration.

[0047] As another optional implementation, the core optimization logic, emotional adaptation requirements and priority sorting of the service adjustment strategy are first deeply analyzed. Then, the whole process framework, parameter setting and execution conditions of the exclusive service strategy are fully disassembled, and the core elements that can be complementary in the two types of strategies are extracted. The optimization requirements of the service adjustment strategy are penetrated into each execution link of the exclusive service strategy, the service flow node order, response timeliness parameter and right adaptation rule are dynamically adjusted, and the logical conflicts and execution contradictions between the strategies are eliminated. Subsequently, the integrated strategy is subjected to whole-process logical consistency verification, and the missing connection mechanisms and abnormal processing processes are supplemented, and finally the target service strategy is formed. In this method, the strategy integration degree is deep, the logical coherence is strong, and the customer demand and emotional state can be accurately adapted, and the customer's recognition and experience of the service are improved.

[0048] Exemplarily, in the context of a securities industry telephone customer service scenario, identity authentication and authorization: when a customer calls a securities customer service to handle business, there is no need for cumbersome password input. Only by speaking, the customer service system can confirm the customer's identity through voiceprint recognition technology, ensure the authenticity of the customer's identity, improve the convenience of service, and effectively reduce the security risks caused by information leakage. Transaction monitoring and risk control: the system can monitor customer transaction behavior in real time and analyze user behavior patterns using intelligent algorithms. After confirming the customer's identity through voiceprint recognition, combined with transaction data, it can determine whether the transaction is abnormal, such as whether there is someone else using the identity to transact, etc., effectively preventing abnormal transactions and protecting the fairness and stability of the market. Customer service optimization: voiceprint recognition technology can realize the intelligentization of customer service. According to the recognized customer identity, it can quickly call out the customer's relevant information and historical service records, improve the customer service response speed and accuracy, and provide more personalized services for customers. For example, provide exclusive services for high-end customers, provide appropriate investment suggestions according to the customer's trading habits and preferences, etc. Intelligent voice quality inspection: using voiceprint recognition and speaker segmentation and clustering technology, it can structure and analyze the details of the telephone recording, monitor the service quality of customer service personnel. By identifying the voiceprints of different speakers, it can determine the respective speaking content of the customer service personnel and the customer, analyze whether the customer service personnel are serving according to the script specifications, whether they are professional and have a good attitude, and also understand the customer's needs and satisfaction, providing a basis for improving service quality. Emotion analysis and early warning: the customer service call is connected to the voiceprint recognition system in real time, which can analyze the emotions in the voice. When the user's speech speed is too fast and the volume is too high, the system automatically marks "emotion warning" and pushes relevant "anxiety keywords" such as "loss" and "scammer", helping customer service personnel to timely detect changes in customer emotions and adjust service strategies to improve customer satisfaction.

[0049] Further, with reference to Figure 2 , Figure 2The present application is a whole flowchart. After the customer calls the securities customer service, the system initializes the voice collection to obtain the initial voice data of the first 10 seconds of the call, extracts 38-dimensional customer voiceprint features using the MFCC voiceprint feature extraction model. Start the voiceprint recognition module, call the identity information library to compare the features with 100,000 customer voiceprint templates in the library through the DTW comparison algorithm, when the similarity is ≥95%, the identity authentication is passed, and the corresponding target customer's historical position and past service record in the customer information library are synchronously called. Start the transaction monitoring, service optimization, and emotion analysis modules in parallel, analyze the customer service call voice data through the LSTM emotion classification model, label the "stable" emotion label, push the "position query" keyword and the service adjustment strategy of "priority response business consultation", generate a dedicated service strategy based on the customer position business scenario, and adjust the response priority from "ordinary" to "high" to get the target service strategy. The call generates a service record and a quality inspection report containing 2 service standards, the voiceprint template is updated offline and stored encrypted, and the system waits for the next service request.

[0050] By extracting the customer voiceprint features in the initial voice data in response to the customer's call request, calling the identity information library to determine the target customer information, combining the customer service call voice data to analyze the business scenario to generate a dedicated service strategy, labeling the label through customer emotion analysis and pushing the service adjustment strategy, and finally optimizing the target service strategy, the technical problem of poor communication quality is solved, the customer identity recognition speed, business scenario matching efficiency, and service strategy personalized adaptation response speed are improved, and the effect of improving the whole process of call service is realized.

[0051] Based on any of the above embodiments, in embodiment two of the present application, the step S10 includes steps A11-A13: Step A11, based on the service request of the customer call, the audio acquisition component of the call terminal captures the voice information of the initial stage of the customer call, and generates raw voice data stream without processing.

[0052] In this embodiment, the call terminal is a device for realizing voice interaction between the customer and the server. The audio acquisition component is a component integrated in the call terminal to pick up voice signals. The voice information of the initial stage of the customer call is the voice content sent by the customer in the early period after the call starts.

[0053] As an optional implementation, based on the service request of the customer call, the audio acquisition component of the call terminal is triggered to enter the working state, continuously capturing the voice information in the initial stage of the customer call, sampling the voice signal at fixed time intervals during the capturing process, converting the sampled signal into a continuous digital signal sequence, and directly packaging it into an original voice data stream without processing. The timestamp information at the sampling time is preserved during the packaging process to ensure the time sequence integrity of the data stream. This method has a simple capturing process and can quickly generate a data stream, providing timely data source support for subsequent voiceprint feature extraction.

[0054] Step A12, performing noise reduction, dereverberation, and format standardization processing on the original voice data stream, filtering out environmental interference signals and invalid audio segments, and outputting pure initial voice data.

[0055] In this embodiment, noise reduction is a processing operation to remove environmental noise from the voice, dereverberation is a processing operation to eliminate echo signals in the voice, and format standardization processing is an operation to convert the voice data into a unified format. Environmental interference signals are irrelevant signals in the voice from the environment. Invalid audio segments are segments of the voice that have no actual information. Pure initial voice data is voice data that has been processed to remove interference and invalid content.

[0056] As an optional implementation, the original voice data stream is first received, and noise reduction processing is performed in sequence to identify and weaken the environmental interference signals therein. Then, dereverberation processing is performed to eliminate repeated echo segments in the voice, and format standardization processing is performed to convert the processed voice data into a unified format. The processed voice data is then subjected to segment screening to filter out invalid audio segments such as silence and noise, and the screened voice data is finally spliced and integrated to output pure initial voice data. This method processes in a fixed sequence, and the operation process is stable and easy to execute, meeting the needs of conventional subsequent processing.

[0057] Step A13, using a preset voiceprint feature extraction algorithm to analyze the pure initial voice data, extracting voiceprint feature parameters that are unique in the pure initial voice data, and composing the customer voiceprint features.

[0058] In this embodiment, the preset voiceprint feature extraction algorithm is a pre-set calculation rule for extracting voiceprint features from voice data. The voiceprint feature parameter is a feature index that reflects the uniqueness of an individual in the voice.

[0059] As an optional implementation, pure initial voice data is received, the data is parsed in sections according to a preset voiceprint feature extraction algorithm, feature parameters capable of reflecting the uniqueness of each section are extracted, the extracted parameters are preliminarily screened, and parameters meeting the uniqueness determination standard are retained. Then the screened parameters are sequentially integrated according to a preset classification manner, and finally the customer voiceprint features are composed. This method has a simple extraction process and fast overall processing speed, and is suitable for scenes with high requirements for processing timeliness.

[0060] Exemplarily, in the securities industry telephone customer service scene, the voiceprint recognition technology module is adopted, the system adopts a hybrid recognition architecture that fuses acoustic features and semantic features through deep learning, and the core implementation includes three sub-modules of feature extraction, model training and recognition comparison, which fully meets the specification requirements of the national standard for sound data collection, storage and exchange. The feature extraction sub-module: a pre-trained front-end network (such as Wav2Vec 2.0) is used to synchronously extract multi-dimensional acoustic features and semantic features. The acoustic features include Mel Frequency Cepstral Coefficients (MFCC, dimension 13-40), fundamental frequency (F0, extraction accuracy ±5Hz), short-time energy (frame length 20ms, frame shift 10ms) and spectral entropy. The semantic features are converted into text vectors in real time through an automatic speech recognition (ASR) encoder, realizing the binding of "voiceprint + semantics" dual features, and the feature data is stored in XML mode, ensuring cross-system compatibility. The model training sub-module: a deep neural network hybrid model is constructed, including a 6-layer convolutional neural network (CNN) for feature dimension reduction, a 2-layer bidirectional long short-term memory network (Bi-LSTM) for capturing time sequence dependency, and a cosine similarity classifier for the output layer. The training process uses data enhancement techniques to optimize robustness, including spectral enhancement (Gamma correction coefficient 0.8-1.2), speech speed adjustment (95%-105% of the original speech speed), reverberation addition (room impulse response simulation) and noise superposition (0-15dB SNR financial customer service scene noise such as background human voice and key sound). The model is trained on a heterogeneous dataset (containing 100,000+ securities customer voice samples covering different ages, dialects and call environments), with an Equal Error Rate (EER) ≤0.005 and an identification response time ≤1.5 seconds. The recognition comparison sub-module: a dual-mode mechanism of "real-time verification and offline updating" is adopted. In real-time verification, the customer's current voice features and the template features in the identity information library are calculated for cosine similarity, the similarity threshold is set to 0.85 (which can be dynamically adjusted according to the business security level), and the similarity ≥ threshold is considered as authentication passed; the offline updating mechanism automatically extracts the customer's daily call voice segments (after desensitization) in the early morning every day to update the voiceprint template, solving the problem of voiceprint drift with age and health status changes.

[0061] Due to the accurate capture, noise reduction, de-echo and format standardization processing by the audio acquisition component, combined with feature extraction, the problems of original speech interference and impure voiceprint feature information are solved, and the uniqueness and integrity of the customer voiceprint feature are improved.

[0062] Based on any of the above embodiments, in the third embodiment of the present application, the step S20 comprises steps B11-B13: Step B11, trigger the calling instruction of the identity information library, extract all registered customer corresponding customer voiceprint template from the identity information library.

[0063] In this embodiment, the calling instruction is a signal triggering the identity information library to perform data extraction operation. The registered customer refers to the customer who has completed information registration in the system.

[0064] As an optional implementation, after receiving the triggered identity information library calling instruction, the data reading function of the identity information library is started, all registered customer associated data stored in the library is retrieved in order according to the customer registration sequence, the voiceprint template data corresponding to each customer is separated from it, the integrity of the extracted voiceprint template data is checked to ensure that there is no missing feature parameter for each template. Then all the customer voiceprint templates that pass the check are arranged and summarized in a unified format to complete the extraction operation. This method has a direct extraction process, does not require additional processing steps, has high execution efficiency, and is suitable for scenarios with high extraction time efficiency requirements and small customer scale.

[0065] Step B12, according to the preset voiceprint similarity comparison algorithm, the customer voiceprint feature is matched with each customer voiceprint template one by one to obtain a matching result list containing the similarity values of each matching item.

[0066] In this embodiment, the preset voiceprint similarity comparison algorithm is a rule preset for calculating the matching degree of the customer voiceprint feature and the voiceprint template. The similarity value is a quantitative index representing the matching degree of the voiceprint feature and the template. The matching result list is an ordered data set recording the similarity values of all templates and voiceprint features.

[0067] As an optional implementation, after obtaining the customer voiceprint feature and all customer voiceprint templates, according to the preset voiceprint similarity comparison algorithm, each parameter of the customer voiceprint feature is matched with the corresponding parameter of each customer voiceprint template one by one to calculate the matching degree of each parameter pair. Then, according to the preset weight, the degrees of all parameter pairs are integrated to obtain the overall similarity value of each voiceprint template and the customer voiceprint feature. According to the retrieval order of the customer voiceprint template, the identification of each template and the corresponding similarity value are recorded in order to form a matching result list containing the similarity values of each matching item. This method has comprehensive matching logic and can fully consider the degrees of all feature parameters, which is suitable for scenarios with high matching accuracy requirements.

[0068] Step B13, screening the optimal matching item in the matching result list according to the preset similarity threshold, associating and calling the corresponding customer identity data of the optimal matching item from the identity information library, and integrating to generate the target customer information.

[0069] In this embodiment, the preset similarity threshold is a critical value pre-set for determining the effectiveness of the voiceprint matching. The optimal matching item is the voiceprint template matching record with the highest similarity in the matching result list and meeting the threshold requirement. The target customer information is the complete customer information set formed after integrating the customer identity data.

[0070] As an optional implementation, after obtaining the matching result list and the preset similarity threshold, all matching items in the list are traversed first, the similarity values of each matching item are extracted and compared with the threshold, and the effective matching items with values higher than the threshold are screened out. Then the effective matching items are sorted from high to low according to the similarity values, and the matching item at the top of the sorting is selected as the optimal matching item. Based on the template identifier of the optimal matching item, the corresponding customer identity data is directly called from the identity information library, the called data is verified for field integrity, the data passing the verification is integrated in a preset format, and the target customer information is generated. This method has high execution efficiency and can quickly lock the optimal matching item, which is suitable for regular scenarios with high time efficiency requirements for identity confirmation.

[0071] Illustratively, in the telephone customer service scenario of the securities industry, the identity information library module adopts a "distributed storage and encrypted isolation" architecture, which is divided into a basic information layer, a voiceprint template layer, and a security verification layer, meeting the third level requirements of the financial data security protection. The basic information layer stores the customer's core identity information, including name, ID number (SHA-256 encrypted storage), securities account number (mask display, only the last 4 digits are visible to customer service), contact number (AES-256 encryption), opening account information and business permission range, and the data copy is synchronized to the off-site disaster recovery center, with RTO≤4 hours and RPO≤15 minutes. The voiceprint template layer adopts a "main template and sub-template" structure, the main template is a feature vector generated from the standard voice collected during customer opening account (3 segments, each segment 5-8 seconds, containing random numbers and securities terminology); the sub-template is a feature vector generated from the effective voice segment extracted in daily conversation, with a quantity of ≥5. All voiceprint data is encoded in XML format according to GB / T 26237.13-2023 standard, and is associated with the basic information through a unique customer ID. The security verification layer deploys an access control list (ACL), which is only accessible by the voiceprint recognition server and authorized customer service terminals, and the access log is retained for ≥6 months. The voiceprint template is encrypted using the national encryption algorithm SM4, and the key is managed by a hardware encryption module (HSM), which supports regular key rotation (period ≤90 days) to prevent data leakage or tampering.

[0072] By calling the identity information library to extract the full-amount template, aligning and matching the algorithm one by one, and screening the optimal item by threshold, the problems of low efficiency of voiceprint comparison and inaccurate identity association are solved, and the accuracy and speed of target customer information generation are improved.

[0073] Based on any of the above embodiments, in the fourth embodiment of the present application, the step S30 comprises steps C11-C13: Step C11, the structured integration of the target customer information and the customer service call voice data extracts the customer historical business records and the key information representing the demand direction in the call, and outputs the customer demand related information set after integration.

[0074] In this embodiment, the customer historical business records are the relevant record data of the target customer's past business. The key information representing the demand direction in the call is the core voice content that can reflect the customer's demand direction in the call. The customer demand related information set is the complete demand data set formed after integrating the historical business records and the key information in the call.

[0075] As an optional implementation, according to the preset data classification standard, the target customer information is divided into identity basic information and historical business information modules, and the customer service call voice data is divided into multiple voice segments according to the time axis. The content of each segment is identified, and the key information representing the customer's demand direction is extracted. Then the extracted key information and the historical business records in the target customer information are matched according to the business type, the repeated information is removed, and the corresponding relationship between the historical records and the call demands is sorted and ordered, and the customer demand related information set is output. The method has clear integration process, simple operation and high efficiency, and can quickly output structured customer demand information to meet the demand analysis time requirement in the conventional business scenario.

[0076] Step C12, using a preset business scenario recognition algorithm, classifying and matching the scenario features of the customer demand related information set, determining the specific business scenario type corresponding to the customer demand, and generating the corresponding business scenario identifier.

[0077] In this embodiment, the preset business scenario recognition algorithm is a rule system preset for determining the business classification corresponding to the customer demand. The scenario feature is the unique attribute identifier of each type of business scenario. The specific business scenario type is the explicit business classification to which the customer demand belongs. The business scenario identifier is a symbol or code used to standardize the representation of the specific business scenario type.

[0078] As an optional implementation, after obtaining the set of customer demand related information, the explicit features directly related to the business scene are extracted according to the preset business scene recognition algorithm. The explicit features are compared with the standard features of various scenes in the preset business scene feature library one by one, the feature matching coincidence degree is calculated, and the scene type with the highest coincidence degree is selected as the specific business scene type corresponding to the customer demand. Then, the preset standardized identifier is called according to the scene type to generate the corresponding business scene identifier. This method has a simple recognition process and high efficiency depending on explicit feature matching.

[0079] As another optional implementation, a plurality of preset business scene recognition algorithms are called to independently analyze the scene features of the set of customer demand related information, and a plurality of preliminary scene determination results are obtained. The determination confidence and core matching features of each algorithm are extracted, the preliminary results are weighted and fused according to the preset weight system, the determination results with a confidence lower than a threshold are eliminated, and the final specific business scene type is determined through secondary verification based on the uniqueness and relevance of the scene features, and a business scene identifier containing the fusion basis and confidence identifier is generated. This method has high scene recognition accuracy and strong anti-interference ability.

[0080] Step C13, based on the business scene identifier and the set of customer demand related information, combining the preset service strategy rule library, matching the customer historical service preference with the business scene adaptation requirement, generating the exclusive service strategy.

[0081] In this embodiment, the preset service strategy rule library is a set of corresponding rules of various business scenes and service strategies stored in advance. The customer historical service preference is the tendency and selection habit shown by the customer in the past when accepting services. The business scene adaptation requirement is the adaptation standard of service process and response mode in a specific business scene.

[0082] As an optional implementation, the business scene identifier is associated with the set of customer demand related information, and the core demand points and scene key attributes in the associated information set are extracted. The preset service strategy rule library is called to retrieve the basic service strategy template corresponding to the scene. Then, the core tendency in the customer historical service preference is matched, the key parameters of the template are adjusted according to the adaptation requirements in the rule library, the configurations conflicting with the preference are eliminated, and finally the exclusive service strategy is generated. This method has a simple matching process, high execution efficiency, and can quickly output available strategies to meet the service adaptation requirements of most conventional business scenes and ensure the timeliness of service response.

[0083] For example, in a securities industry telephone customer service scenario, target customer information (including 3 historical fund redemption records and 2 account consultation records) is structured and integrated with 120 seconds of customer service call voice data. Five key pieces of information representing the customer's needs are extracted from the historical business records: transaction frequency, consultation type, and five key information points indicating the customer's request tendency, such as "account balance inquiry" and "holding details verification." After removing duplicate content, a set of customer demand-related information is output. A pre-set CNN business scenario recognition algorithm is used to classify and match the scenario features (such as "balance inquiry" and "holding details verification") in this information set, determining that the specific business scenario type corresponding to the customer's need is "account information inquiry scenario," and generating a corresponding business scenario identifier SC-001. Based on this business scenario identifier and the set of customer demand-related information, a pre-set service strategy rule library storing more than 200 scenario adaptation rules is invoked. This library matches the customer's historical service preferences (such as prioritizing APP push results and not requiring manual follow-up) with scenario adaptation requirements (such as real-time response and accurate data display), adjusts the service response priority to "Level 1," configures the detailed push channel to the customer's bound APP, and generates a dedicated service strategy including response process, push method, and information display dimensions.

[0084] By integrating structured data, combining scenario recognition with rule base for precise matching, the problems of ambiguous business scenario identification and low adaptability of service strategies to needs have been solved, thereby improving the accuracy of business scenario identification and the efficiency of customizing exclusive service strategies.

[0085] Based on any of the above embodiments, in Embodiment 5 of this application, step S40 includes steps D11 to D14: Step D11: Extract the emotional intensity, emotional fluctuation amplitude, and core emotional expression fragments from the emotional analysis results corresponding to the target customer, and integrate them to obtain customer emotional characteristic data.

[0086] In this embodiment, emotional intensity is a quantitative indicator of the strength of a customer's emotions. Emotional fluctuation amplitude is the range of change in a customer's emotions within a specific time period. The core emotional expression segment is the voice content segment that best reflects the customer's dominant emotion. Customer emotional feature data is a comprehensive data representation of the customer's emotional state formed by integrating emotional intensity, fluctuation amplitude, and core emotional expression segments.

[0087] As an optional implementation, the method obtains the sentiment analysis results corresponding to the target customer, directly extracts the labeled sentiment intensity data and sentiment fluctuation amplitude values ​​according to preset dimensions, and simultaneously retrieves voice segments directly related to sentiment expression from the analysis results, selecting the segments with the highest frequency and clearest sentiment indication as core sentiment expression segments. The extracted sentiment intensity and fluctuation amplitude data are formatted and then integrated with the core sentiment expression segments according to the logical order of intensity, fluctuation, and segment to obtain customer sentiment feature data. This method has a simple extraction process, high execution efficiency, and can quickly output results, meeting the rapid service adaptation needs of common scenarios.

[0088] As an alternative implementation, the sentiment analysis results of the target customers are tracked over time, and the sentiment data is segmented into time slices. The quantified value of the sentiment intensity and fluctuation nodes of each slice are extracted simultaneously. Semantic dependency analysis is then performed on sentiment-related speech segments to uncover the relationship between core sentiment expressions and context. Potential sentiment features are dynamically supplemented based on sentiment evolution trends. The extracted multi-dimensional information is verified and corrected in real time, and integrated according to temporal logic and semantic association to obtain customer sentiment feature data containing dynamically changing characteristics. This method can capture dynamic changes in sentiment and provides more comprehensive feature extraction.

[0089] Step D12: According to the preset emotion tag classification rules, match and determine the customer emotion feature data, automatically label the corresponding emotion tags, and obtain the customer emotion analysis results with emotion tags.

[0090] In this embodiment, the preset emotion tagging classification rules are pre-defined standards and corresponding logic used to determine emotion types. Emotion tags are feature identifiers used to standardize and identify customer emotion types. The customer emotion analysis result with emotion tags is analysis data containing complete customer emotion information after the emotion tags have been labeled.

[0091] As an optional implementation, this method extracts the intensity and core emotional expression fragments from customer emotional feature data. It then directly matches these fragments against the intensity range and expression characteristics of each emotion in a pre-defined emotional tagging system to determine the most suitable emotion type and automatically label it. The labeled emotional tags are then linked to the original customer emotional feature data, supplemented with annotation information, to obtain the customer emotional analysis results with emotional tags. This method features a direct judgment process, high execution efficiency, and rapid tagging, meeting the needs of rapid service response in typical scenarios.

[0092] Step D13: Extract the core keywords of the call that trigger the corresponding emotions from the customer emotion analysis results, and combine them with the preset emotion and service adjustment strategy mapping library to generate targeted suggestions that are adapted to the current emotional state.

[0093] In this embodiment, the core keywords that trigger the corresponding emotion in a call are key voice words that directly evoke a specific emotion in the customer during the call. The preset emotion-service adjustment strategy mapping library is a pre-stored set of associations between emotion types and corresponding service optimization schemes. The current emotional state is the specific emotional type and intensity level the customer is currently experiencing, and the targeted suggestions are specific guidance schemes for service adjustments adapted to the current emotional state.

[0094] As an optional implementation, after obtaining customer sentiment analysis results, call audio segments directly related to emotional expression are retrieved, and frequently occurring words with clear semantic meanings are extracted as core keywords triggering the corresponding emotions. These core keywords are then deduplicated and semantically categorized to determine a core keyword set. A pre-defined sentiment-service adjustment strategy mapping library is then invoked. Based on the sentiment type corresponding to the customer's current emotional state, the service adjustment strategies associated with that current emotional state are matched in the mapping library. Combining this with the semantic tendency of the core keywords, strategy content suitable for those core keywords is selected and integrated to form targeted suggestions. This method features a simple extraction and matching process, high execution efficiency, and rapid suggestion output.

[0095] Step D14: Based on the targeted suggestions, filter out the associated statement data and communication methods, and combine them to generate the service adjustment strategy.

[0096] In this embodiment, the associated statement data is data related to service communication scripts adapted to targeted suggestions. The communication method refers to the format and rhythm of interaction with customers during the service process.

[0097] As an optional implementation, after obtaining targeted suggestions, the core service optimization directions and adaptation requirements are extracted. A pre-defined statement database is accessed to filter statement data that semantically aligns with the optimization directions. The statement sequence is organized according to the communication logic, and the corresponding basic communication methods are identified based on the suggestions. The organized statement data and the determined communication methods are directly combined according to the correspondence between the wording and execution form to supplement basic execution specifications and generate a service adjustment strategy. This method features a simple combination process, high execution efficiency, and can quickly generate strategies, meeting the rapid service adaptation needs in high-concurrency scenarios.

[0098] As an alternative implementation method, this approach involves in-depth analysis of the emotional adaptation points and core needs in targeted suggestions, constructing a dynamic combination model of statement data and communication methods. The tone and frequency of statements are dynamically adjusted based on the intensity of the customer's emotions. Combined with historical service preferences, suitable communication channels and response rhythms are selected. Real-time semantic optimization of the statement data ensures alignment with the customer's expression habits. Simultaneously, combination adjustment trigger conditions are set, and the combined service adjustment strategy is validated in real time, supplementing dynamic adjustment rules and generating service adjustment strategies that include personalized combination logic and trigger mechanisms. This method offers a high degree of strategy personalization and strong adaptability, improving the smoothness of service communication and customer acceptance.

[0099] For example, in the securities industry's telephone customer service scenario, the emotion recognition technology module works in parallel with the voiceprint recognition module. Based on the acoustic features of the speech signal and the text content, it achieves multi-dimensional emotion analysis, supporting both real-time alerts and post-event analysis. Its technical architecture references the deep learning framework of the Chime SDK's speech emotion analysis model. Emotion feature extraction: Acoustic dimension extraction includes speech rate (words / minute, threshold > 200 words / minute triggers alert), volume (effective value > 65dB triggers alert), tone fluctuation (fundamental frequency standard deviation > 15Hz), and pause frequency (pauses > 5 times per 10 seconds). Text dimension extraction uses real-time ASR transcription results to match a pre-defined securities industry emotion keyword library (anxiety: loss, trapped, scammer, complaint; calm: consultation, inquiry, understanding; joy: profit, gratitude). Emotion classification model: A two-stage training mode is adopted. The first stage trains the ASR model to simultaneously recognize text content and preliminary emotion labels (positive, neutral, negative). The second stage freezes the ASR encoder as the front end and connects to a 3-layer fully connected network to build an emotion classifier. For speech data with missing labels, Amazon Comprehend is used to generate pseudo-emotion labels; for ambiguous speech data, Amazon Transcribe is used to optimize the transcribed text to ensure the integrity of the model training data. A real-time alert mechanism is implemented: using 5-second speech segments as the analysis unit, the system outputs the emotion probability every 2.5 seconds and calculates the average emotion within a 30-second sliding window. When the negative emotion probability > 0.7 and any acoustic alert condition is triggered, the system automatically labels the emotion level (Level 1: Anxiety, Level 2: Anger) and pushes the alert information and related keywords to the customer service workbench, with an alert delay of ≤ 0.5 seconds.

[0100] By extracting multi-dimensional emotional features, accurately matching tags, mapping keywords to strategies, and combining statements with communication methods, the problem of inaccurate matching and lack of targeting between service adjustment strategies and customer emotional states has been solved, thereby improving the effectiveness of service adjustments and the efficiency of alleviating customer emotions.

[0101] Based on any of the above embodiments, in Embodiment Six of this application, step S50 includes steps E11 to E13: Step E11: Extract the core optimization requirements, priorities, and adaptation rules from the service adjustment strategy, and integrate and output the strategy optimization parameter set.

[0102] In this embodiment, the core optimization requirements are the most critical elements to be optimized in the service adjustment strategy. Priority is the criterion for ranking the importance of the core optimization requirements. Adaptation rules are the scenario conditions and matching logic applicable to the service adjustment strategy. The strategy optimization parameter set is a standardized set of parameter data formed by integrating the core optimization requirements, priorities, and adaptation rules.

[0103] As an optional implementation, after obtaining the service adjustment strategy, the core optimization requirements explicitly marked in the strategy are directly filtered according to preset parameter extraction dimensions. The priority ranking information specified in the strategy is extracted, and the applicable adaptation rules are broken down. The extracted core optimization requirements, priorities, and adaptation rules are deduplicated and standardized according to a unified data format. The standardized three items are then arranged in an orderly manner according to the correspondence between requirements, priorities, and rules, and the strategy optimization parameter set is integrated and output. This method has a direct and clear extraction process, high execution efficiency, and can quickly complete parameter integration, rapidly outputting a basic and complete set of strategy optimization parameters.

[0104] As an alternative implementation, after obtaining the service adjustment strategy, the explicit core optimization requirements, priorities, and adaptation rules are first extracted. Then, semantic analysis and logical deduction are used to uncover hidden optimization tendencies and potential adaptation conditions within the strategy. Implicit features are categorized, labeled, and their credibility verified. Explicit parameters and implicit features are integrated according to a core-auxiliary hierarchy, supplementing the derivation basis and application boundaries of implicit features. After being standardized, a set of strategy optimization parameters containing explicit parameters and implicit features is output. Hidden optimization tendencies refer to customer preference adaptations that are not explicitly stated but logically implied, while potential adaptation conditions refer to parameter adjustment thresholds under specific scenarios. This method provides more comprehensive parameter coverage, supports deep strategy optimization, and improves the accuracy of service optimization strategies.

[0105] Step E12: Based on the preset strategy optimization priority rules, the strategy optimization parameter set and the exclusive service strategy are compared module by module. The priority of conflicting service items is determined and adjusted to obtain the adapted preliminary optimization service strategy.

[0106] In this embodiment, the preset strategy optimization priority rule is a pre-defined standard used to determine the importance of strategy optimization items and the order of conflict handling. Conflicting service items are service execution contents that contradict each other in the parameter set and the dedicated service strategy. Priority determination and adjustment involves determining the importance of conflicting service items according to the rules and adjusting and retaining the higher-priority content. The initial optimized service strategy is the initial optimized service plan adapted to the parameter set after conflict adjustment.

[0107] As an optional implementation, the dedicated service strategy is divided into basic modules according to the service execution stage, and a preset set of strategy optimization priority rules and parameter sets is obtained. The optimization requirements and adaptation rules in the parameter sets are compared with the corresponding module content of the dedicated service strategy in module order to identify conflicting service items. The priority of the conflicting items is directly determined according to the strategy optimization priority rules, retaining the higher-priority service content and replacing the lower-priority items. After completing the comparison and adjustment of all modules, the optimized content of each module is integrated and connected to obtain the adapted preliminary optimized service strategy. This method features simple module division, efficient comparison and adjustment processes, and rapid result output.

[0108] Step E13: Verify the completeness and logic of the preliminary optimized service strategy, supplement missing service connection links, correct unreasonable service process settings, and generate the target service strategy.

[0109] In this embodiment, completeness refers to the attribute that a service strategy includes all necessary service steps. Logicality refers to the attribute that the service flow is logically connected and consistent. Service connection steps are the transitional and related steps between different service modules. An unreasonable service flow setting refers to a service execution order and specification that does not meet adaptation requirements.

[0110] As an optional implementation, after obtaining the initial optimized service strategy, the necessary service steps are checked one by one according to the preset integrity verification dimensions to ensure they are complete. Simultaneously, the process logic is analyzed according to the service execution order to identify nodes with broken connections and logical inconsistencies. Standardized transition steps are added to missing service connection points, and unreasonable process settings are corrected according to general adaptation specifications to ensure smooth connection of each step in logical order. Finally, the corrected strategy is comprehensively reviewed and integrated to generate the target service strategy. This method has a simple verification process, high execution efficiency, and can quickly fill in and correct core issues.

[0111] For example, in the securities industry telephone customer service scenario, extract the three core optimization requirements ("first feedback within 1 minute", "progress synchronization every 15 minutes", "proactive explanation of compensation plan"), priorities (the first two are level one, the last one is level two) and two adaptation rules ("only applicable to account business consultation scenarios" and "triggered when customer emotional intensity ≥ 0.8") from the service adjustment strategy (including 10 reassurance / progress description statements and the "priority response and phased feedback" communication method), and integrate and output a set of strategy optimization parameters containing 12 standardized parameters. Based on the preset strategy optimization priority rules (judged by two dimensions: "customer needs fit and service urgency"), the exclusive service strategy was divided into four modules: "response timeliness, feedback mechanism, compensation explanation, and script adaptation." Each module was adapted and compared against the parameter set. Two conflicting service items were identified: "feedback frequency (originally synchronized every 30 minutes)" and "compensation explanation (originally lacked clear details)." The higher priority content in the parameter set was prioritized according to the rules. The feedback frequency was adjusted to every 15 minutes, and three additional compensation explanation details were added, resulting in a preliminary optimized service strategy. The completeness and logic of this strategy were verified. Two service connection links were found to be missing: "compensation application guidance" and "anomaly handling connection." The unreasonable process setting of "synchronizing business progress first and then explaining compensation" was corrected (adjusted to "compensation explanation first and then synchronizing business progress"). After closed-loop verification confirmed no logical loopholes, the target service strategy was generated.

[0112] By accurately extracting parameters, adjusting module conflicts, and verifying and correcting the entire process, the problems of strategy optimization not meeting requirements and process breaks have been solved, thereby improving the adaptability of the target service strategy and the efficiency of its implementation.

[0113] Based on any of the above embodiments, in Embodiment Seven of this application, referring to Figure 3 , Figure 3 This is a flowchart illustrating the seventh embodiment of the call quality optimization method based on voiceprint recognition in this application. Step E12 includes steps F11-F13: Step F11: According to the preset modules of service process, response priority, and exclusive rights, the service adjustment strategy and the exclusive service strategy are split, and the adaptability of the corresponding service items under each module is verified one by one to obtain the module comparison results.

[0114] In this embodiment, the service flow refers to the sequential steps and operational procedures for service execution. Response priority is the criterion for ranking service items by urgency. Exclusive benefits are service discounts and guarantees customized for customers. Preset modules are pre-defined strategy breakdown and classification units. Corresponding service items are the service content corresponding to the functions under the same module in two strategies. Module comparison results are the conclusion data recording the verification status of each module.

[0115] As an optional implementation method, the service adjustment strategy and the exclusive service strategy are separated according to three modules: preset service process, response priority, and exclusive rights. All service items under each module are extracted. The corresponding service items of the two strategies are aligned one by one according to the module order, verifying the fit between core functions and execution requirements, and recording whether the service items are adapted and the core dimensions of any differences. For differences, only basic discrepancies are marked without in-depth analysis. The verification records of all modules are integrated to form a module comparison result that clearly lists the adaptation status. This method has a simple decomposition and verification process, high execution efficiency, and can quickly complete the comparison, meeting the needs of scenarios with high timeliness requirements and no need for in-depth analysis.

[0116] Step F12: Extract the identified conflicting service items from the module comparison results, quantify and sort the priority levels of the conflicting service items based on the strategy optimization priority rules, and output the priority determination results of the conflicting service items.

[0117] In this embodiment, the identified conflicting service items are those services that are clearly marked as contradictory and cannot be executed simultaneously in the module comparison results. The priority determination result of conflicting service items is the final data set recording the level and order of all conflicting service items.

[0118] As an optional implementation, the filtering module compares all identified conflicting service items in the results and extracts the core functional attributes of each conflicting service item. It then calls a preset strategy optimization priority rule, quantifies the conflicting service items according to the single core dimension specified in the rule, and assigns them corresponding priority levels. Finally, all conflicting service items are arranged in descending order of priority level, and conflicting service items of the same level are sorted sequentially according to the extraction order. The final output is a conflicting service item priority determination result containing the conflicting service item name, priority level, and sorting result. This method has a simple and intuitive determination process, high execution efficiency, and can quickly complete priority division and sorting.

[0119] Step F13: Based on the priority determination result of the conflicting service items, retain the core configuration of the high-priority service items, adjust the parameters of the low-priority conflicting service items, and reorganize and integrate the adjusted service items of each module with the service items of the non-conflicting modules according to the original service framework to obtain the adapted preliminary optimized service strategy.

[0120] As an optional implementation, all high-priority service items in the conflict service item priority determination results are extracted and their core configurations are fixed. For low-priority conflict service items, parameters are directly adjusted according to the adaptation requirements of the high-priority core configurations to ensure that the parameters do not conflict with the high-priority content. Then, strictly following the module division and process sequence of the original service framework, the adjusted conflict service items in each module are sequentially recombined with the non-conflicting module service items to ensure that the module positions and process logic conform to the original framework specifications. After recombining, a basic connection check is performed to obtain a preliminary optimized service strategy. This method is simple and intuitive, highly efficient, and can quickly complete strategy recombining, meeting the rapid iteration needs of common scenarios.

[0121] For example, in a securities industry telephone customer service scenario, based on preset modules of service process, response priority, and exclusive benefits, the service adjustment strategy (5 items in the service process module, 3 levels in the response priority module, and 4 items in the exclusive benefits module) and the exclusive service strategy (3 items in the service process module, 2 levels in the response priority module, and 3 items in the exclusive benefits module) are broken down. Adaptability checks are performed on each service item under each module, identifying three conflicting service items: "service process initiation node," "response timeliness standard," and "exclusive benefits redemption method." This yields module comparison results with clearly defined conflict locations and differences. These three conflicting service items are extracted, and combined with strategy optimization priority rules based on the Analytic Hierarchy Process (AHP), priority levels are quantified from three dimensions: customer need fit, execution urgency, and scenario adaptability. "Response timeliness standard" is determined as a first-level priority, "service process initiation node" as a second-level priority, and "exclusive benefits redemption method" as a third-level priority. After sorting by priority from highest to lowest, the output shows the conflicting service item priority determination results, including the conflicting service item name, priority level, sorting result, and weight of the determination dimensions. Based on this judgment, the core configuration of the first-priority "response timeliness standard" is retained, the trigger parameters of the second-priority "service process initiation node" and the execution path parameters of the third-priority "exclusive rights redemption method" are adjusted, and the 3 conflicting service items and 10 non-conflicting module service items (4 service processes, 2 response priorities, and 4 exclusive rights) of each module are reorganized and integrated according to the original service framework module hierarchy and process logic to ensure smooth connection between modules and obtain an adapted preliminary optimized service strategy.

[0122] By using precise verification through module splitting, quantitative determination of conflict priorities, and reorganization and integration within the framework, the problems of chaotic policy conflict adaptation and unclear integration logic have been solved, thereby improving the accuracy of the initial optimization service policy adaptation and the efficiency of integration execution.

[0123] Based on any of the above embodiments, in Embodiment Eight of this application, referring to Figure 4 , Figure 4This is a flowchart illustrating the eighth embodiment of the call quality optimization method based on voiceprint recognition in this application. Following step S50, steps G11-G13 are also included: Step G11: Classify, organize, and encapsulate the customer feedback data and quality inspection feedback data after the execution of the target service strategy to obtain the feedback dataset.

[0124] In this embodiment, the quality inspection feedback data is the evaluation data generated after quality inspection of the target service strategy execution process and results. The feedback dataset is a standardized set of feedback data formed after being categorized, organized, and packaged.

[0125] As an optional implementation method, customer feedback data and quality inspection feedback data after the execution of the target service strategy are acquired. These two types of data are then categorized and sorted according to preset feedback types, eliminating invalid and duplicate data, and arranging similar data in chronological order. The sorted data undergoes basic format standardization processing, supplementing core identifier fields such as data source and collection time. Finally, it is integrated and encapsulated according to the logical structure of customer feedback and quality inspection feedback to obtain the feedback dataset. This method features clear classification logic, a simple operation process, and high execution efficiency, meeting the basic data statistics and viewing needs in typical scenarios.

[0126] Step G12 involves using a preset feedback analysis algorithm to deeply analyze the feedback dataset, identify the core dimensions and specific adjustment directions that need to be optimized in the next round of calls, and integrate them to obtain an optimization requirement list.

[0127] In this embodiment, the preset feedback analysis algorithm is a pre-defined analytical logic and method used to parse feedback data and identify optimization directions. The core dimensions to be optimized in the next round of calls are the key areas that need to be improved in subsequent call services. The specific adjustment directions are actionable improvement paths formulated for the core dimensions. The optimization requirement list is a standardized improvement requirement document formed by integrating the core dimensions and adjustment directions.

[0128] As an optional implementation, a pre-defined feedback analysis algorithm is invoked to extract frequently occurring problem keywords and negative evaluation tendencies from the feedback dataset. Problems are categorized according to service modules, and the modules with the highest frequency and widest impact are identified as core optimization dimensions. A pre-defined adjustment direction library for corresponding modules is directly matched to the algorithm, selecting the most suitable improvement paths. These paths are then sorted and integrated according to the correspondence between core dimensions and specific adjustment directions to obtain a list of optimization requirements. This method features a simple analysis process, high execution efficiency, and the ability to quickly identify explicit optimization points, meeting the rapid improvement needs in short-cycle iteration scenarios.

[0129] As an alternative implementation, the feedback dataset is hierarchically split, and a multi-dimensional analysis framework is constructed based on service modules, feedback types, and severity. Association analysis algorithms are used to uncover implicit correlations between problem points in customer feedback and evaluation results in quality inspection feedback. Historical optimization records and current feedback data are simultaneously integrated to identify common optimization needs across modules and deduce targeted adjustment directions. The data is then hierarchically sorted by correlation strength and optimization urgency, supplemented with correlation evidence and expected improvement effects, resulting in a list of optimization needs containing cross-dimensional correlation information. This method provides a more comprehensive identification of optimization needs, a more systematic approach to adjustments, and a more guiding basis for subsequent parameter updates.

[0130] Step G13: According to the optimization requirement list, update the voiceprint feature extraction parameters, service strategy rule base, and emotion and service adjustment mapping relationship to generate a set of optimization configuration parameters for the next round of calls, providing an updated execution basis for subsequent customer call quality optimization.

[0131] In this embodiment, the voiceprint feature extraction parameters are configuration data used to extract customer voiceprint features. The service policy rule base is a collection storing rules corresponding to various business scenarios and service policies. The emotion-service adjustment mapping relationship is the association data between emotion types and corresponding service optimization schemes. The optimization configuration parameter set for the next round of calls is an updated standardized parameter set adapted for subsequent call quality optimization.

[0132] As an optional implementation, the adjustment requirements corresponding to each optimization item in the optimization requirement list are extracted, and the corresponding configuration items in the voiceprint feature extraction parameters, service strategy rule base, and emotion-service adjustment mapping relationship are directly matched. The parameter values ​​are updated according to the adjustment requirements, rule content is supplemented or modified, and mapping relationship logic is corrected. The updated three items undergo basic format standardization and duplicate item removal. Based on the structure of voiceprint parameters, rule base, and mapping relationship, a set of optimized configuration parameters for the next round of calls is generated, providing an updated execution basis for subsequent customer call quality optimization. This method has a direct update process, high execution efficiency, and can quickly complete configuration iteration and output a usable set of optimized configuration parameters.

[0133] For example, in the securities industry's telephone customer service scenario, 120 customer feedback data points (including 80 satisfaction ratings, 30 problem feedbacks, and 10 improvement suggestions) and 80 quality inspection feedback data points (including three types of evaluation results: service process compliance, response timeliness, and script adaptation) after the implementation of the target service strategy were categorized, organized, and packaged. These were then sorted by "feedback type and related service modules," 15 invalid and duplicate data points were removed, and fields such as feedback source, collection time, and severity were added, resulting in a feedback dataset containing 185 valid data points. A pre-defined LDA topic model feedback analysis algorithm was used to deeply analyze this dataset, uncovering high-frequency problem themes and implicit related needs. Three core dimensions requiring optimization in the next round of calls were identified: "optimization of response timeliness," "improvement of script adaptation accuracy," and "simplification of rights redemption process." Six specific adjustment directions were derived, including "shortening the initial response time to within 30 seconds," "adding three types of scripts adapted to moderately negative emotions," and "simplifying the rights redemption review process." This resulted in a standardized list of optimization requirements. According to the optimization requirements list, the voiceprint feature extraction parameters were updated (the feature extraction dimension was expanded from 15 to 20), the service strategy rule base was updated (12 new emotion and speech matching rules were added), and the emotion and service adjustment mapping relationship was updated (5 sets of exclusive service adjustment schemes corresponding to moderate emotions were added). A set of dedicated optimization configuration parameters for the next round of calls was generated, which includes 45 configuration items in 3 categories of updated parameters, providing an updated execution basis for the quality optimization of subsequent customer calls.

[0134] By standardizing and integrating feedback data, deeply analyzing algorithms, and dynamically updating core parameters, the problems of lack of accurate data support and lagging configuration iteration for service optimization have been solved, thereby improving the accuracy of subsequent call service adaptation and the efficiency of optimization iteration.

[0135] This application provides a call quality optimization device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the call quality optimization method based on voiceprint recognition in the first embodiment described above.

[0136] The following is for reference. Figure 5This document illustrates a structural schematic diagram of a call quality optimization device suitable for implementing embodiments of this application. The call quality optimization device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, professional calling terminals, personal digital assistants (PDAs), tablet computers (PADs), portable multimedia players (PMPs), network infrastructure equipment, etc., as well as fixed terminals such as network call optimization devices, desktop computers, etc. Figure 5 The call quality optimization device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0137] like Figure 5 As shown, the call quality optimization device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage device 1003 into a random access memory (RAM) 1004. The random access memory 1004 also stores various programs and data required for the operation of the call quality optimization device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the call quality optimization device to communicate wirelessly or wiredly with other devices to exchange data. Although call quality optimization devices with various systems are shown in the figures, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0138] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0139] The call quality optimization device provided in this application employs the voiceprint recognition-based call quality optimization method described in the above embodiments, which can solve the technical problem of poor communication quality. Compared with the prior art, the beneficial effects of the call quality optimization device provided in this application are the same as those of the voiceprint recognition-based call quality optimization method provided in the above embodiments, and other technical features in this call quality optimization device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0140] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0141] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0142] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the voiceprint recognition-based call quality optimization method in the above embodiments.

[0143] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination thereof.

[0144] The aforementioned computer-readable storage medium may be included in the call quality optimization device; or it may exist independently and not assembled into the call quality optimization device.

[0145] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a call quality optimization device, the call quality optimization device causes the following actions: In response to a user's service request, it acquires the customer's initial voice data during the call and extracts the customer's voiceprint features from the initial voice data; it calls an identity information database and compares the customer's voiceprint features with customer voiceprint templates in the identity information database to determine the target customer information corresponding to the customer's voiceprint features; based on the target customer information and customer service call voice data, it analyzes the business scenario of the target customer's needs and generates a dedicated service strategy corresponding to the target customer; based on the emotion analysis results for the target customer, it automatically labels emotion tags and pushes keywords and service adjustment strategies; and it optimizes the dedicated service strategy according to the service adjustment strategy to obtain the target service strategy corresponding to the target customer.

[0146] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0148] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0149] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described voiceprint recognition-based call quality optimization method, thereby solving the technical problem of poor communication quality. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the voiceprint recognition-based call quality optimization method provided in the above embodiments, and will not be repeated here.

[0150] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for optimizing call quality based on voiceprint recognition, characterized in that, The method includes: In response to a service request made by a user, the system obtains the customer's initial voice data during the call and extracts the customer's voiceprint features from the initial voice data. The identity information database is invoked, and the customer voiceprint features are compared with the customer voiceprint templates in the identity information database to determine the target customer information corresponding to the customer voiceprint features. Based on the target customer information and customer service call voice data, analyze the business scenarios that meet the needs of the target customers, and generate exclusive service strategies for the target customers. Based on the sentiment analysis results of the target customers, the system automatically labels sentiments and pushes keywords and service adjustment strategies. The exclusive service strategy is optimized according to the service adjustment strategy to obtain the target service strategy corresponding to the target customer.

2. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, The steps of responding to a user's service request by calling, obtaining the customer's initial voice data during the call, and extracting the customer's voiceprint features from the initial voice data include: Based on the service request made by the customer, the voice information of the customer in the initial stage of the call is captured by the audio acquisition component of the call terminal to generate an unprocessed raw voice data stream; The original speech data stream is subjected to noise reduction, echo removal, and format standardization to filter out environmental interference signals and invalid audio segments, and output clean initial speech data. The clean initial speech data is analyzed using a preset voiceprint feature extraction algorithm to extract unique voiceprint feature parameters from the clean initial speech data, forming the customer's voiceprint features.

3. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, The step of calling the identity information database and comparing the customer's voiceprint features with the customer's voiceprint templates in the identity information database to determine the target customer information corresponding to the customer's voiceprint features includes: Trigger the call command to the identity information database and extract the customer voiceprint templates corresponding to all registered customers from the identity information database; According to the preset voiceprint similarity comparison algorithm, the customer voiceprint features are matched one by one with each of the customer voiceprint templates to obtain a list of matching results containing the similarity values ​​of each matching item. The optimal matching item in the matching result list is selected based on a preset similarity threshold, and the customer identity data corresponding to the optimal matching item is retrieved from the identity information database and integrated to generate the target customer information.

4. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, The step of analyzing the business scenarios of target customer needs based on the target customer information and customer service call voice data, and generating a dedicated service strategy for the target customer, includes: The target customer information and the customer service call voice data are structured and integrated to extract key information that represents the customer's historical business records and the tendency of their demands during the call. After integration, a set of customer demand-related information is output. Using a preset business scenario recognition algorithm, the scenario features of the customer demand-related information set are classified and matched to determine the specific business scenario type corresponding to the customer demand and generate the corresponding business scenario identifier. Based on the business scenario identifier and the set of customer demand-related information, and combined with the preset service strategy rule base, the exclusive service strategy is generated by matching the customer's historical service preferences with the business scenario adaptation requirements.

5. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, The steps of automatically labeling emotions and pushing keywords and service adjustment strategies based on the sentiment analysis results of the target customers include: Extract the emotional intensity, emotional fluctuation amplitude, and core emotional expression fragments from the emotional analysis results corresponding to the target customer, and integrate them to obtain customer emotional characteristic data; According to the preset emotion tag classification rules, the customer emotion feature data is matched and judged, and the corresponding emotion tags are automatically labeled to obtain the customer emotion analysis results with emotion tags. Extract the core keywords of the call that trigger the corresponding emotion from the customer sentiment analysis results, and combine them with a preset emotion and service adjustment strategy mapping library to generate targeted suggestions that are adapted to the current emotional state. Based on the targeted recommendations, relevant statement data and communication methods are filtered out and combined to generate the service adjustment strategy.

6. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, The step of optimizing the exclusive service strategy according to the service adjustment strategy to obtain the target service strategy for the target customer includes: Extract the core optimization requirements, priorities, and adaptation rules from the service adjustment strategy, and integrate them to output a set of strategy optimization parameters; According to the preset strategy optimization priority rules, the strategy optimization parameter set and the exclusive service strategy are compared module by module, and the priority of conflicting service items is determined and adjusted to obtain the adapted preliminary optimization service strategy. Verify the completeness and logic of the preliminary service optimization strategy, supplement missing service connection links, correct unreasonable service process settings, and generate the target service strategy.

7. The call quality optimization method based on voiceprint recognition as described in claim 6, characterized in that, The step of adapting and comparing the set of strategy optimization parameters with the dedicated service strategy module by module according to the preset strategy optimization priority rules, and determining and adjusting the priority of conflicting service items to obtain the adapted preliminary optimization service strategy includes: The service adjustment strategy and the exclusive service strategy are broken down into preset modules based on service process, response priority, and exclusive rights. Adaptability verification is performed on the corresponding service items under each module to obtain module comparison results. Extract the identified conflicting service items from the module comparison results, combine the strategy optimization priority rules to quantify and rank the priority levels of the conflicting service items, and output the priority determination results of the conflicting service items. Based on the priority determination results of the conflicting service items, the core configuration of the high-priority service items is retained, the parameters of the low-priority conflicting service items are adjusted, and the adjusted service items of each module are reorganized and integrated with the service items of the non-conflicting modules according to the original service framework to obtain the adapted preliminary optimized service strategy.

8. The call quality optimization method based on voiceprint recognition as described in claim 1, characterized in that, After the step of optimizing the exclusive service strategy according to the service adjustment strategy to obtain the target service strategy corresponding to the target customer, the call quality optimization method based on voiceprint recognition further includes: The customer feedback data and quality inspection feedback data after the execution of the target service strategy are classified, organized, and packaged to obtain a feedback dataset. The feedback dataset is analyzed in depth using a preset feedback analysis algorithm to identify the core dimensions and specific adjustment directions that need to be optimized in the next round of calls, and to integrate them to obtain a list of optimization requirements. Based on the optimization requirement list, update the voiceprint feature extraction parameters, service strategy rule base, and emotion-service adjustment mapping relationship to generate a set of optimization configuration parameters for the next round of calls, providing an updated execution basis for subsequent customer call quality optimization.

9. A call quality optimization device, characterized in that, The call quality optimization device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the call quality optimization method based on voiceprint recognition as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the call quality optimization method based on voiceprint recognition as described in any one of claims 1 to 8.

Images