Intelligent telemarketing quality control method and system based on voiceprint recognition

Abstract

The invention provides an intelligent telemarketing quality control method and system based on voiceprint recognition, and relates to the technical field of telemarketing quality control. Comprising the following steps: extracting a first target voiceprint feature of target seat voice data by adopting an MFCC algorithm, and extracting a second target voiceprint feature of the target seat voice data by adopting a deep learning method; fusing the first target voiceprint feature and the second target voiceprint feature to obtain a target voiceprint fusion feature; matching the target voiceprint fusion feature with a real voiceprint feature of a target seat in a pre-established voiceprint library to obtain a matching result; and according to a matching result, verifying the voice data of the telemarketing target seat to realize telemarketing quality control. According to the method, high-reliability identity biological characteristic verification is seamlessly fused into the telemarketing / co-marketing quality inspection process, non-sensitive, automatic and full-process quality supervision is realized, the blank of the prior art is filled, and increasingly strict risk control and compliance requirements are met.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of consumer finance electric sales quality control, and particularly relates to an intelligent electric sales quality control method and system based on voiceprint recognition. BACKGROUND

[0002] In the financial, telecommunications, insurance and other industries that highly rely on telephone sales (electric sales) or collaborative sales (collaborative sales), it is crucial to ensure the quality, compliance and risk control of the sales process. The traditional electric sales / collaborative sales quality control (quality inspection) mainly relies on the following methods: 1. Manual sampling: quality inspectors randomly select a small number of call recordings for manual listening and scoring.

[0003] This approach has low coverage (usually less than 1-5%), low efficiency, high cost and strong subjectivity, making it difficult to fully reflect the true sales quality and unable to timely identify risks.

[0004] 2. Automatic quality inspection based on speech-to-text (ASR) and natural language processing (NLP).

[0005] This type of system analyzes call text content, identifies keywords, illegal phrases, service specifications, etc. Although it improves coverage and efficiency, it has fundamental limitations: A. Unable to verify the authenticity of the agent's identity: The system can only analyze "what was said," but cannot confirm "who said it." This leaves a huge loophole for serious violations of "person-card separation" (such as non-registered agents impersonating, agents taking turns to punch cards "flying cards," and agents leaving work being replaced by others). Impersonators can easily bypass content-based quality inspection by following the compliance phrases.

[0006] B. Invasive / interfering supervision: Some systems require agents to perform specific operations (such as entering a job number or pressing a key to confirm) at the beginning or end of a call, which interferes with the normal sales process, reduces agent experience and work efficiency, and may even cause resistance.

[0007] C. Data authenticity in question: Agents are aware of being monitored (especially when they need to actively cooperate), which may lead to distorted behavior and "performance-based service," making it impossible to reflect the true level of customer interaction.

[0008] 3. Monitoring based on video or other biometric features.

[0009] While this approach can verify identity (such as facial recognition), it has high implementation costs (requiring the deployment of specialized equipment), strong privacy invasion, low feasibility in agent work scenarios (especially remote or mobile scenarios), and difficulty in large-scale application. SUMMARY

[0010] In order to overcome the above-mentioned deficiencies of the prior art, the present application provides an intelligent electric sales quality control method and system based on voiceprint recognition, which seamlessly integrates high-reliability identity biometric verification into electric sales / collaborative sales quality inspection processes, realizes quality supervision without feeling, automation and full-process, so as to fill the gap in the prior art and meet the increasingly stringent risk control and compliance requirements.

[0011] To achieve the above object, one or more embodiments of the present application provide the following technical solutions: The first aspect of the present application provides an intelligent electric sales quality control method based on voiceprint recognition.

[0012] The intelligent electric sales quality control method based on voiceprint recognition comprises the following steps: Obtaining telephone sales target agent voice data to be recognized; Extracting first target voiceprint features of the target agent voice data using an MFCC algorithm and extracting second target voiceprint features of the target agent voice data using a deep learning method; Fusing the first target voiceprint features and the second target voiceprint features to obtain target voiceprint fusion features; Matching the target voiceprint fusion features with target agent real voiceprint features in a pre-established voiceprint library to obtain a matching result; Verifying the telephone sales target agent voice data according to the matching result to realize telephone sales quality control.

[0013] The second aspect of the present application provides an intelligent electric sales quality control system based on voiceprint recognition.

[0014] The intelligent electric sales quality control system based on voiceprint recognition comprises: A data acquisition module configured to acquire telephone sales target agent voice data to be recognized; A feature extraction module configured to extract first target voiceprint features of the target agent voice data using an MFCC algorithm and extract second target voiceprint features of the target agent voice data using a deep learning method; A feature fusion module configured to fuse the first target voiceprint features and the second target voiceprint features to obtain target voiceprint fusion features; A matching module configured to match the target voiceprint fusion features with target agent real voiceprint features in a pre-established voiceprint library to obtain a matching result; A verification module configured to verify the telephone sales target agent voice data according to the matching result to realize telephone sales quality control. The third aspect of the present application provides a computer readable storage medium, which stores a program, and the program is executed by a processor to implement the steps in the voiceprint recognition based intelligent electric sales quality control method according to the first aspect of the present application.

[0015] The fourth aspect of the present application provides an electronic device, which comprises a memory, a processor and a program stored in the memory and executable on the processor, and the processor implements the steps in the voiceprint recognition based intelligent electric sales quality control method according to the first aspect of the present application when executing the program.

[0016] The above one or more technical solutions have the following beneficial effects: The present application provides a voiceprint recognition based intelligent electric sales quality control method and system, which seamlessly integrates high-reliability identity biological feature verification into electric sales / collaborative sales quality inspection process, realizes non-susceptible, automatic and full-process quality supervision, fills the gap in the prior art, and meets increasingly stringent risk control and compliance requirements.

[0017] The present application takes the voiceprint recognition of the target agent as the identity biological feature for non-susceptible identification of telephone voice, and if the target voiceprint fusion feature and the real voiceprint feature of the target agent in the pre-established voiceprint library can be successfully matched, the dynamic comprehensive score of the target agent is further calculated, and in the calculation of the dynamic comprehensive score, the voiceprint recognition score, the service specification score and the compliance score of the speech technique are comprehensively considered. If the target voiceprint fusion feature and the real voiceprint feature of the target agent in the pre-established voiceprint library cannot be successfully matched, it is directly determined that the target agent verification fails, and the highest level risk mark is triggered.

[0018] In the present application, the MFCC algorithm is used to extract the first target voiceprint feature of the target agent voice data, the deep learning method is used to extract the second target voiceprint feature of the target agent voice data, the first target voiceprint feature and the second target voiceprint feature are fused to obtain the target voiceprint fusion feature, and the voiceprint fusion feature is matched to obtain a more accurate matching result and improve the accuracy of the matching.

[0019] The advantages of the additional aspects of the present application will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0020] The drawings accompanying the specification of the present application form a part thereof, serve to provide further understanding of the present application, and together with the description of the exemplary embodiments of the present application and the explanation thereof serve to explain the present application, and do not constitute improper limitations on the present application.

[0021] Figure 1 The method flowchart of the embodiment one.

[0022] Figure 2 Data flow chart for example one.

[0023] Figure 3 Second target voiceprint feature extraction flow chart for example one. DETAILED DESCRIPTION

[0024] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0025] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application.

[0026] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0027] Example one As mentioned earlier, the core pain points in the current field of electric sales / collaborative sales quality control are: 1. How to efficiently, accurately and large-scale verify the authenticity of the agent's identity in the call, and completely eliminate identity fraud risks such as impersonation? 2. How to minimize the interference with the normal work of the agent while achieving effective supervision, and improve the experience and acceptance? 3. How to improve the quality inspection coverage and efficiency while ensuring the authenticity and objectivity of the collected data? The existing technical solutions (manual sampling inspection, pure content quality inspection, video monitoring) cannot effectively solve the above pain points at the same time, especially in the non-intrusive identity verification aspect. This leads to serious compliance risks and business risks for enterprises, such as commission losses caused by fly orders, customer disputes and management efficiency bottlenecks.

[0028] In this context, an innovative technical solution is urgently needed to seamlessly integrate high-reliability identity biometric verification into the electric sales / collaborative sales quality inspection process, achieve non-intrusive, automated and full-process quality supervision, and fill the gap in existing technologies to meet increasingly stringent risk control and compliance requirements.

[0029] Based on this, the present embodiment discloses an intelligent electric sales quality control method based on voiceprint recognition, which has the following innovative technologies: 1. Deep coupling mechanism of voiceprint and quality inspection.

[0030] 1) The voiceprint verification result is used as a necessary pass for the quality inspection process for the first time, solving the identity fraud vulnerability of traditional solutions. 2). The voiceprint similarity score is directly included in the total quality score, realizing the quantitative management of biometric features.

[0031] 2. Complex scene adaptive voiceprint engine.

[0032] 1). Triple robustness design of noise suppression (SE-MelGAN) + device calibration + dynamic model update, solving the application problem of electric sales environment; 2). Combined with speaker separation technology, accurately positioning the target seat in multi-person conversation.

[0033] 3. Non-intrusive full-process monitoring 1). Use the original speech stream to realize non-intrusive collection and avoid seat behavior distortion; 2). Voiceprint verification and content quality inspection are processed asynchronously, and the call duration is not increased.

[0034] 4. Biometric security system 1). Voiceprint features are stored in irreversible encryption (such as Bcrypt hash); 2). Use federated learning technology, model update does not need to upload original speech, meet GDPR / Personal Information Protection Law.

[0035] As shown in Figure 1 and Figure 2 , the intelligent electric sales quality control method based on voiceprint recognition includes the following steps: Obtain the telephone sales target seat voice data to be identified; Use the MFCC algorithm to extract the first target voiceprint feature of the target seat voice data, and use the deep learning method to extract the second target voiceprint feature of the target seat voice data; Fuse the first target voiceprint feature and the second target voiceprint feature to obtain the target voiceprint fusion feature; Match the target voiceprint fusion feature with the target seat real voiceprint feature in the pre-established voiceprint library to obtain the matching result; According to the matching result, verify the telephone sales target seat voice data, and realize telephone sales quality control.

[0036] This embodiment creatively uses the voiceprint verification result as the necessary pass for the quality inspection process, solving the identity fraud vulnerability of traditional solutions. The voiceprint verification result has the highest priority. When the voiceprint verification result is passed, further calculate the dynamic comprehensive score of the target seat, and comprehensively evaluate the service of the target seat.

[0037] Further, the telephone sales target seat voice data to be identified is obtained, specifically including: In a multi-person conversation scenario, the speaker separation (Diarization) technology is combined to locate the telephone sales target agent voice channel; Through the located telephone sales target agent voice channel, the telephone sales target agent call voice stream is obtained in real time; The call voice stream is segmented through VAD to obtain a call voice segment; The SE-MelGAN speech enhancement technology is used to preprocess the call voice segment to obtain telephone sales target agent voice data to be recognized.

[0038] The embodiment adopts a non-intrusive method, can realize non-intrusive real-time verification, can maximize the restoration of the real service process of the target agent, realize non-sensitization verification, and reduce the psychological pressure and resistance of the target agent personnel.

[0039] In terms of noise reduction design, the embodiment: The SE-MelGAN (speech enhancement) preprocessing noise environment voice is adopted; The multi-device calibration module (headset / telephone frequency response compensation) is embedded to eliminate the influence of hardware differences.

[0040] For the collection of telephone sales target agent voice data to be recognized, a dual-mode collection method is adopted: Real-time stream processing: The call voice stream is segmented through VAD (voice activity detection) to obtain an agent voice segment, and the voiceprint feature is extracted in real time; Offline asynchronous verification: The complete recording is verified again to ensure the robustness of the result.

[0041] Further, the MFCC algorithm is used to extract the first target voiceprint feature of the target agent voice data. The MFCC (Mel-scale Frequency Cepstral Coefficients) feature is a cepstrum parameter extracted in the Mel scale frequency domain. The Mel scale describes the nonlinear characteristics of the human ear frequency. The MFCC algorithm is used to extract the voiceprint feature of the voice data to be recognized, and the obtained MFCC feature is the first target voiceprint feature.

[0042] As shown in Figure 3 The deep learning method is used to extract the second target voiceprint feature of the target agent voice data, and the specific process is as follows: Step 1: Feature data collection: Collect a large amount of agent voice data and perform fine labeling (identity ID, emotion label, compliance label, etc.); Step 2: Preprocessing: Voice Activity Detection (VAD) processing is performed to retain only the segments containing speech; then, enhancements are applied to the audio, including: adding background noise, reverberation effects simulating room impulse responses (RIR), randomly changing the speed and volume of the sound, etc. Step 3: Feature extraction: ECAPA-TDNN is used as the backbone network, combined with multi-task learning (MTL), the main task is identity verification with AAM-Softmax loss, and the auxiliary task is emotion or stress classification, to extract a fixed-dimensional embedding vector, which is a "second target voiceprint feature" that integrates identity, emotion, style, and other multi-dimensional information. Step 4: Application and feedback: Based on the application results of identity verification, quality analysis, clustering analysis, etc., further deep learning is performed to continuously fine-tune and optimize the "second target voiceprint feature".

[0043] Further, a weighted fusion method is used to fuse the first target voiceprint feature and the second target voiceprint feature to obtain a target voiceprint fusion feature.

[0044] Further, the target voiceprint fusion feature is matched with the target agent real voiceprint feature in the pre-established voiceprint library to obtain a matching result, specifically: The cosine similarity algorithm is used to calculate the similarity between the target voiceprint fusion feature and the target agent real voiceprint feature in the pre-established voiceprint library to obtain a voiceprint similarity score; The voiceprint similarity score is compared with the threshold value of the voiceprint similarity score corresponding to the target agent, and if the voiceprint similarity score is greater than the threshold value of the voiceprint similarity score, the matching is successful; The pre-established voiceprint library stores the target agent real voiceprint feature.

[0045] The target agent real voiceprint feature stored in the pre-established voiceprint library refers to the voiceprint feature corresponding to the user identifier that is pre-recorded.

[0046] This embodiment judges whether the voiceprint is the target agent himself, and in specific implementation, it is done through the voiceprint similarity score. First, a voiceprint similarity is calculated through the cosine similarity, and then the voiceprint similarity is converted into a score form. In the conversion, a pre-set mapping table can be used, which stores a plurality of cosine similarities and corresponding score relationships. Finally, a voiceprint similarity score (0-100) is obtained, and the threshold value can be flexibly configured (such as ≥85 points to pass).

[0047] Further, based on the telephone sales target agent voice data to be identified, the service specification score and the speech technique compliance score of the target agent are calculated, and the specific process is: Pre-set phone sales violation word library, silence duration threshold and speech rate requirement; Based on AI semantic recognition technology, the semantic understanding of the phone sales target agent voice data to be recognized is performed to determine whether it contains violation words in the phone sales violation word library; The total number of violation words is accumulated to obtain a first cumulative number, and the ratio of the first cumulative number to the total number of words in the phone sales target agent voice data to be recognized is calculated to obtain a compliance score of the script; The cumulative number of times that the actual silence duration does not meet the silence duration threshold and the actual speech rate does not meet the speech rate requirement in the phone sales target agent voice data to be recognized is determined to obtain a second cumulative number; Based on the mapping relationship between the second cumulative number and the corresponding duration of the phone sales target agent voice data to be recognized, a service specification score is obtained.

[0048] The embodiment provides a double-layer monitoring means that fuses voiceprints and content. In the telephone sales quality control, the voiceprint recognition has the highest priority. When the voiceprint recognition is successfully matched, further identification of service specification and compliance of the script is performed. When the voiceprint verification fails, the highest risk level is automatically triggered without content analysis.

[0049] Specifically, in the identification of service specification and compliance of the script, a rule + AI dual-engine quality inspection is adopted: Basic rule library: pre-set violation word library (such as “guaranteed return” and “absolute safety”), silence duration threshold, speech rate requirement and other rules.

[0050] Based on the phone sales target agent voice data to be recognized, the identification of service specification and compliance of the script is performed in combination with the basic rule library, silence duration threshold, speech rate requirement and other rules.

[0051] In the specific implementation of the identification of service specification and compliance of the script, based on the AI semantic analysis technology, the text is translated through voice ASR, and then the translated text is identified for compliance of the script, customer emotional fluctuation and business key point coverage based on the NLP model (BERT).

[0052] Further: If the target voiceprint fusion feature can be successfully matched with the real voiceprint feature of the target agent in the pre-established voiceprint library, the dynamic comprehensive score of the target agent is further calculated: The dynamic comprehensive score of the target agent = (voiceprint similarity score × α) + (compliance score of the script × β) + (service specification score × γ); Wherein, α, β and γ are weights; If the target voiceprint fusion feature cannot be matched successfully with the real voiceprint feature of the target agent in the pre-established voiceprint library, it is directly determined that the target agent verification fails, triggering the highest level risk mark; The threshold of the voiceprint similarity score of the target agent triggering the highest level risk mark is raised.

[0053] The weights are configurable, for example, in a high-risk scenario of identity fraud, set α=50%, β=30%, and γ=20%.

[0054] Complaint handling closed loop: One-key complaint: when the agent questions the quality inspection result, the system automatically associates the original recording, voiceprint comparison report, and violation segment mark.

[0055] Three-dimensional review: artificial reviewers can simultaneously view voiceprint verification results + violation content text + time axis positioning, improving processing efficiency by more than 50%.

[0056] The present embodiment can achieve: (1) Managed distribution function: Marking high-risk agents → automatically increasing voiceprint verification frequency to 100% + content quality inspection intensity; Project-level hosting: mandatory full-quantity voiceprint verification + keyword scanning during the new product online period.

[0057] (2) Full-link traceability: Each call recording generates a unique quality inspection number, which runs through the voiceprint report → violation mark → complaint record → final score; Supporting reverse checking of original data by number to meet audit compliance requirements.

[0058] The following describes the technical effects achieved by the present embodiment in a comprehensive manner: I. Building an insurmountable biological defense line to eradicate the identity fraud problem: Effects: 100% real-time verification of agent identity: mandatory voiceprint comparison for each distribution call to eliminate fraud such as "person-card separation" type fly single, shift replacement, and account subletting from the source.

[0059] Impersonation detection rate > 99.5%: higher than manual sampling inspection (<30% detection rate) and pure content quality inspection (0% detection rate), which reduces commission fraud losses by 75%+ (based on data from a certain insurance customer).

[0060] Advantages: Technical uniqueness: full-quantity biological feature verification that cannot be achieved by traditional solutions, making the present embodiment the ultimate means of compliance and risk control.

[0061] Risk prepositioning: voiceprint verification failure alarm in real time, without the need for post-facto accountability, avoiding loss expansion.

[0062] II. Revolutionary breakthrough in non-intrusive supervision, reshaping quality inspection paradigm: Effects: Zero interference for agents: No need to press keys / report ID numbers, no awareness of call process, agent satisfaction improved by 40% (survey of a certain bank).

[0063] 100% quality inspection coverage: Breakthrough in manual sampling inspection <5%, achieve full-service non-corner monitoring.

[0064] Data authenticity jumps: Capture real service status in non-interference environment, customer complaint analysis accuracy improved by 55%.

[0065] Advantages: Experience and regulation: Break the dead cycle of traditional solutions "strict regulation ⇋ agent resistance".

[0066] Implicit cost elimination: Save the time cost of agents cooperating with quality inspection, release 80+ working hours per person per year for revenue generation.

[0067] III. Automation-driven cost reduction and efficiency improvement, quality inspection efficiency exponentially increased: Effects: Quality inspection efficiency improved by 80%: Voiceprint verification (0.5 seconds / segment) + AI content analysis instead of 90% manual listening.

[0068] Complaint handling period shortened to 2 hours: Full-link data automatic association (traditional mode needs 1-3 days).

[0069] Human cost reduced by 60%: 2000-agent center quality inspection team reduced from 50 to 20 people.

[0070] Advantages: Resource precise placement: Focus on high-risk recording review, value density improved by 300%.

[0071] Millisecond-level risk response: Managed distribution function realizes real-time interception in high-risk scenarios (such as sensitive words + voiceprint anomaly double triggering).

[0072] IV. From fuzzy score to precise data governance, evaluation system upgraded: Effects: Quality score credibility improved by 90%: Voiceprint score (weight α) as a hard indicator, eliminating data distortion caused by identity fraud.

[0073] Full-process traceability efficiency improved by 10 times: Locate problem nodes through quality inspection number in 5 seconds (traditional verification needs to cross-system call >5 minutes).

[0074] Dynamic weight regulation: Voiceprint weight α adjusted to 70% during new product online period, quickly identify weak points in training.

[0075] Advantages: Scientific decision-making: Managers can optimize training strategies based on multi-dimensional data heat maps (such as "high compliance score + low voiceprint score" agent groups).

[0076] Audit-free: Biometric verification records + full-link logs meet financial-level audit requirements.

[0077] Embodiment Two The embodiment discloses an intelligent electric sales quality control system based on voiceprint recognition.

[0078] The intelligent electric sales quality control system based on voiceprint recognition comprises: A data acquisition module configured to acquire telephone sales target agent voice data to be recognized; A feature extraction module configured to extract first target voiceprint features of the target agent voice data using an MFCC algorithm and extract second target voiceprint features of the target agent voice data using a deep learning method; A feature fusion module configured to fuse the first target voiceprint features and the second target voiceprint features to obtain target voiceprint fusion features; A matching module configured to match the target voiceprint fusion features with target agent real voiceprint features in a pre-established voiceprint library to obtain a matching result; A verification module configured to verify the telephone sales target agent voice data according to the matching result to realize telephone sales quality control. Embodiment Three The purpose of the embodiment is to provide a computer-readable storage medium.

[0079] The computer-readable storage medium stores a computer program, which is executed by a processor to implement the steps in the intelligent electric sales quality control method based on voiceprint recognition according to Embodiment 1 of the present disclosure.

[0080] Embodiment Four The purpose of the embodiment is to provide an electronic device.

[0081] The electronic device comprises a memory, a processor, and a program stored in the memory and executable on the processor, and the processor executes the program to implement the steps in the intelligent electric sales quality control method based on voiceprint recognition according to Embodiment 1 of the present disclosure.

[0082] The steps involved in the apparatuses of the above embodiments two, three and four correspond to the method of embodiment one, and the specific implementation can refer to the relevant description of embodiment one. The term "computer readable storage medium" should be understood as including a single medium or multiple media of one or more instruction sets; it should also be understood as including any medium capable of storing, encoding or carrying the instruction set for execution by the processor and causing the processor to perform any of the methods in the present application.

[0083] Those skilled in the art should understand that each module or step of the present application described above can be realized by a general computer device, alternatively, they can be realized by program codes executable by a computing device, so that they can be stored in a storage device for execution by a computing device, or they can be respectively made into each integrated circuit module, or a plurality of modules or steps among them can be made into a single integrated circuit module to realize. The present application is not limited to any specific combination of hardware and software.

[0084] Although the specific embodiments of the present application are described above in combination with the drawings, it is not a limitation on the scope of protection of the present application, and those skilled in the art should understand that various modifications or changes made by those skilled in the art on the basis of the technical solutions of the present application without creative labor are still within the scope of protection of the present application.

Claims

1. A method for intelligent electric selling quality control based on voiceprint recognition, characterized in that, The method comprises the following steps: Obtaining telephone sales target agent voice data to be identified; Extracting first target voiceprint features of the target agent voice data by using an MFCC algorithm and extracting second target voiceprint features of the target agent voice data by using a deep learning method; Fusing the first target voiceprint features and the second target voiceprint features to obtain target voiceprint fusion features; Matching the target voiceprint fusion features with target agent real voiceprint features in a pre-established voiceprint library to obtain a matching result; Verifying the telephone sales target agent voice data according to the matching result to realize telephone sales quality control.

2. The voiceprint recognition-based intelligent electric dialing quality management method of claim 1, wherein, Obtaining telephone sales target agent voice data to be identified, specifically comprising: In a multi-person conversation scene, combining a speaker separation technology to locate a telephone sales target agent sound channel; Obtaining a telephone sales target agent call voice stream in real time through the located telephone sales target agent sound channel; Segmenting the call voice stream through VAD to obtain a call voice segment; Preprocessing the call voice segment by using an SE-MelGAN voice enhancement technology to obtain the telephone sales target agent voice data to be identified. 3.The voiceprint recognition-based intelligent electric dialing quality management method of claim 1, wherein, Extracting the second target voiceprint features of the target agent voice data by using a deep learning method, specifically comprising: Using ECAPA-TDNN as a backbone network, combining multi-task learning, a main task being identity verification of AAM-Softmax loss, and an auxiliary task being emotion or stress classification, extracting an embedding vector fused with multi-dimensional information including identity, emotion and style to obtain the second target voiceprint features.

4. The voiceprint recognition-based intelligent electric dialing quality management method of claim 1, wherein, Fusing the first target voiceprint features and the second target voiceprint features by using a weighted fusion method to obtain the target voiceprint fusion features.

5. The voiceprint recognition-based intelligent electric dialing quality management and control method of claim 1, wherein, Matching the target voiceprint fusion features with target agent real voiceprint features in a pre-established voiceprint library to obtain a matching result, specifically comprising: Calculating a voiceprint similarity score by using a cosine similarity algorithm for similarity calculation of the target voiceprint fusion features and the target agent real voiceprint features in the pre-established voiceprint library; Comparing the voiceprint similarity score with a threshold value of the voiceprint similarity score corresponding to the target agent, and if the voiceprint similarity score is greater than the threshold value of the voiceprint similarity score, the matching is successful. The pre-established voiceprint library stores the target agent real voiceprint features.

6. The voiceprint recognition-based intelligent electric dialing quality management and control method of claim 5, wherein, Further comprising, based on the telephone sales target agent voice data to be identified, calculating a service specification score and a speech compliance score of the target agent, and the specific process is: Pre-setting a telephone sales violation word library, a silence duration threshold value and a speech speed requirement; Based on an AI semantic recognition technology, performing semantic understanding on the telephone sales target agent voice data to be identified to determine whether the telephone sales target agent voice data to be identified contains a violation word in the telephone sales violation word library; Accumulating a total number of the violation words to obtain a first accumulated number, calculating a ratio of the first accumulated number to a total number of words in the telephone sales target agent voice data to be identified to obtain the speech compliance score; Determining an accumulated number of times that an actual silence duration does not conform to the silence duration threshold value and an actual speech speed does not conform to the speech speed requirement in the telephone sales target agent voice data to be identified to obtain a second accumulated number; Based on a mapping relationship between the second cumulative number and a time length corresponding to the telephone sales target agent voice data to be identified, a service specification score is obtained.

7. The voiceprint recognition-based intelligent electric sales quality control method of claim 6, characterized in that: If the target voiceprint fusion feature and the real voiceprint feature of the target agent in the pre-established voiceprint library can be successfully matched, the dynamic comprehensive score of the target agent is further calculated: The comprehensive score = (voiceprint similarity score × a) + (script compliance score × b) + (service specification score × g); wherein a, b and g are weights; If the target voiceprint fusion feature and the real voiceprint feature of the target agent in the pre-established voiceprint library cannot be successfully matched, it is directly determined that the target agent verification fails, triggering a highest level risk mark; The threshold of the voiceprint similarity score of the target agent triggering the highest level risk mark is raised.

8. The intelligent electric selling quality control system based on voiceprint recognition, characterized in that, It comprises: A data acquisition module configured to acquire telephone sales target agent voice data to be identified; A feature extraction module configured to extract a first target voiceprint feature of the target agent voice data using an MFCC algorithm and extract a second target voiceprint feature of the target agent voice data using a deep learning method; A feature fusion module configured to fuse the first target voiceprint feature and the second target voiceprint feature to obtain a target voiceprint fusion feature; A matching module configured to match the target voiceprint fusion feature with a real voiceprint feature of the target agent in a pre-established voiceprint library to obtain a matching result; A verification module configured to verify the telephone sales target agent voice data according to the matching result to realize telephone sales quality control.

9. A computer-readable storage medium having stored thereon a program, characterized in that, The program is executed by the processor to realize the steps in the voiceprint recognition-based intelligent electric sales quality control method of any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored on the memory and executable on the processor, characterized in that, The processor executes the program to realize the steps in the voiceprint recognition-based intelligent electric sales quality control method of any one of claims 1-7.

Images