A loan contract voice signing method, device and storage medium
Through voice information conversion and multi-model verification, combined with blockchain technology, the automated signing of loan contracts is achieved, solving the problems of cumbersome signing process and insufficient reliability of voiceprint extraction models, and improving the signing efficiency and accuracy.
Patent Information
- Application Number
- CN202410333235.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-22
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-03-22
AI Technical Summary
The signing process of traditional loan contracts is cumbersome and inefficient, which cannot meet the needs of fast loans, and the existing voiceprint extraction model is insufficient to achieve voice signing of loan contracts.
By receiving voice information to convert it into text, extracting key information, analyzing the purpose of the loan, matching the loan product, generating loan plans, and real-time approval is carried out after user confirmation, multiple voiceprint extraction models are used for verification, combining blockchain anti-counterfeiting technology and Pre-trained language model to realize the signing of a voice contract.
It improves the efficiency and accuracy of loan contract signing, realizes a de-artificial voice contract signing process, and improves the credibility of voiceprint verification. Customers can complete loan contract signing during the call.
Smart Images

Figure CN117995199B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a loan contract voice signing method, device and storage medium. Background Art
[0002] With the development of artificial intelligence and big data technology, the financial industry is accelerating its digital and intelligent transformation. In the field of consumer loans, traditional business processes are inefficient and cannot meet users' needs for quick loan issuance. Even customers who frequently apply for loans need to fill out forms and wait for manual loan approval when they apply for loans again. The cumbersome procedures and long waiting time result in a poor customer experience, and financial institutions also need to invest a lot of manpower costs to serve customers. Financial institutions store users' basic information, performance records, historical voice information, etc., and have the basic conditions for automatic services through artificial intelligence.
[0003] Existing technologies can already handle some business through intelligent voice interaction, and have been applied in post-loan services. However, it has not yet been applied to loan contract signing. This is because voice signing of loan contracts involves many steps and a long process. If voice signing of loan contracts is to be de-manualized, high requirements are placed on the accuracy and response time of each step. Moreover, if loan contracts are to be signed by voice, the reliability of the existing voiceprint extraction model needs to be further improved. Summary of the invention
[0004] In view of the shortcomings of the prior art, the present invention provides a loan contract voice signing method to solve the problem that the loan contract signing process is cumbersome and inefficient. The present invention also provides a device and a storage medium for executing the above-mentioned loan contract voice signing method.
[0005] In a first aspect, the present invention provides a method for signing a loan contract by voice, comprising the following steps:
[0006] S1: receiving voice information and converting the voice information into text information;
[0007] S2: extracting key information according to the text information, wherein the key information includes the purpose of the loan;
[0008] S3: Analyze the loan purpose to obtain the loan type and match the corresponding loan product; retrieve user information and generate a loan plan including loan amount, loan interest rate, and installment period based on the loan product and user information;
[0009] S4: converting the loan plan into voice information and feeding back the loan plan to the user;
[0010] S5: After the user confirms the loan plan, real-time approval is carried out based on the user information;
[0011] S6: After approval, a loan contract is generated and the user is guided to complete the contract signing. The contract signing is implemented by guiding the user to say a specified sentence, collecting the user's voice information, and using N voiceprint extraction models to extract and compare the voiceprint vectors of the user's historical voice information and the collected voice information. The N is ≥ 2, and the voice contract signing is completed when the verification is passed.
[0012] Preferably, step S6 includes the following sub-steps:
[0013] S61: extracting voiceprint vectors, using the trained N voiceprint extraction models to process the user's historical voice information and the collected voice information respectively, to obtain N voiceprint vectors respectively;
[0014] S62: concatenating voiceprint vectors, concatenating the N voiceprint vectors extracted in S61 to obtain a historical voiceprint vector and a current voiceprint vector;
[0015] S63: Compare the historical voiceprint vector with the current voiceprint vector to calculate the similarity;
[0016] S64: Compare the similarity calculated in S63 with the preset similarity threshold. When the similarity after comparison is greater than the preset similarity threshold, it is determined that they are the same user and the verification is successful; when the similarity after comparison is less than the preset similarity threshold, the verification fails.
[0017] Preferably, step S6 includes using blockchain anti-counterfeiting technology to record the user voice information.
[0018] Preferably, in step S64, cosine similarity is used as a similarity calculation method, and the preset similarity threshold a satisfies 0.7≤a≤0.9.
[0019] Preferably, step S1 includes the following sub-steps:
[0020] S11: dividing the speech information into time windows, and converting each window into a spectrogram or a Mel-spectrogram;
[0021] S12: Use convolutional neural network to extract speech features;
[0022] S13: adding position coding to the extracted speech features;
[0023] S14: Input the feature sequence with position encoding into the Transformer encoder;
[0024] S15: Use the CTC module to decode and obtain text information.
[0025] Preferably, step S2 includes inputting the text information into a BERT pre-trained model, and the BERT pre-trained model outputs the key information.
[0026] Preferably, step S2 includes inputting the text information into a GPT pre-training model, and the GPT pre-training model outputs the key information.
[0027] Preferably, step S5 includes the following sub-steps:
[0028] S51: Arrange and format the loan plan according to the definition of the rule engine, and input it into the rule engine;
[0029] S52: The rule engine inputs the result of rule judgment to the multi-layer perceptron model;
[0030] S53: The multi-layer perceptron model outputs the probability b that the loan plan is approved;
[0031] S54: When b is greater than or equal to a preset threshold, the approval is passed.
[0032] In a second aspect, the present invention provides a loan contract voice signing device, comprising:
[0033] A speech recognition module, used for receiving speech information and converting the speech information into text information;
[0034] A natural language understanding module, used to extract key information from text information, wherein the key information includes the purpose of the loan;
[0035] The personalized loan suggestion module is used to obtain the loan type through loan purpose analysis and match the corresponding loan product; generate a loan plan including the loan amount, loan interest rate, and installment period based on the loan product and user information; convert the loan plan into voice information and feedback the loan plan to the user;
[0036] Real-time loan approval module, used to conduct real-time approval based on user information after the user confirms the loan plan;
[0037] The voice contract confirmation module is used to generate a loan contract after approval and guide the user to complete the contract signing.
[0038] In a third aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the loan contract voice signing method as described above is implemented.
[0039] The loan contract voice signing device and the readable storage medium are used to execute the loan contract voice signing method as described above, and thus have all the beneficial technical effects brought about by the loan contract voice signing method, which will not be described one by one here.
[0040] The beneficial effects of the present invention are as follows: in the process of verifying the identity of a customer through voiceprint, the features of different voiceprint extraction models are integrated, thereby improving the overall characterization capability, enhancing the credibility of voiceprint verification, meeting the requirements for signing a loan contract, and realizing the de-manualization of voice signing of a loan contract. When a customer makes a phone call to apply for a loan, the voice signing of the loan contract can be completed during the call, thus solving the problem of cumbersome and inefficient loan contract signing process. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The above and other purposes, features and advantages of the present invention will become more clear by more specific description of the preferred embodiments of the present invention shown in the accompanying drawings. The same reference numerals indicate the same parts in all the accompanying drawings, and the accompanying drawings are not deliberately scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.
[0042] Figure 1 A flowchart of a voice signing method for a loan contract provided by an embodiment of the present invention;
[0043] Figure 2 A schematic block diagram of a loan contract voice signing device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0044] To facilitate understanding of the present invention, a more comprehensive description of the present invention will be given below with reference to the relevant drawings.
[0045] It should be noted that when an element is considered to be "connected" to another element, it may be directly connected to another element and integrated therewith, or there may be an intermediate element at the same time. The terms "installed", "one end", "the other end" and similar expressions used herein are for illustrative purposes only.
[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those generally understood by those skilled in the art. The terms used in the specification herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.
[0047] refer to Figure 1 The embodiment of the present invention provides a loan contract voice signing method, comprising the following steps:
[0048] S1: receiving voice information and converting the voice information into text information;
[0049] S2: extracting key information according to the text information, wherein the key information includes the purpose of the loan;
[0050] S3: Analyze the loan purpose to obtain the loan type and match the corresponding loan product; retrieve user information and generate a loan plan including loan amount, loan interest rate, and installment period based on the loan product and user information;
[0051] S4: converting the loan plan into voice information and feeding back the loan plan to the user;
[0052] S5: After the user confirms the loan plan, real-time approval is carried out based on the user information;
[0053] S6: After approval, a loan contract is generated and the user is guided to complete the contract signing. The contract signing is implemented by guiding the user to say a specified sentence, collecting the user's voice information, and using N voiceprint extraction models to extract and compare the voiceprint vectors of the user's historical voice information and the collected voice information. The N is ≥ 2, and the voice contract signing is completed after the verification is passed;
[0054] In a preferred embodiment, step S6 includes the following sub-steps:
[0055] S61: extracting voiceprint vectors, using the trained N voiceprint extraction models to process the user's historical voice information and the collected voice information respectively, to obtain N voiceprint vectors respectively;
[0056] S62: concatenating voiceprint vectors, concatenating the N voiceprint vectors extracted in S61 to obtain a historical voiceprint vector and a current voiceprint vector;
[0057] S63: Compare the historical voiceprint vector with the current voiceprint vector to calculate the similarity;
[0058] S64: Compare the similarity calculated in S63 with the preset similarity threshold. When the similarity after comparison is greater than the preset similarity threshold, it is determined that they are the same user and the verification is successful; when the similarity after comparison is less than the preset similarity threshold, the verification fails.
[0059] A multi-mode approach is used to extract and compare voiceprint vectors. The features of different voiceprint extraction models are integrated in the process of verifying customer identity through voiceprints, thereby improving the overall representation capability and the credibility of voiceprint verification, meeting the requirements for signing loan contracts. There is no need for manual service users. After dialing the service phone, the entire process of loan contract signing can be completed through one phone call, solving the problem of cumbersome and inefficient loan contract signing process.
[0060] In a preferred embodiment, step S6 includes using blockchain anti-counterfeiting technology to record the user voice information.
[0061] In a preferred embodiment, cosine similarity is used as a similarity calculation method in step S64, and the preset similarity threshold a satisfies 0.7≤a≤0.9.
[0062] In a preferred embodiment, step S1 includes the following sub-steps:
[0063] S11: dividing the speech information into time windows, and converting each window into a spectrogram or a Mel-spectrogram;
[0064] S12: Use convolutional neural network to extract speech features;
[0065] S13: adding position coding to the extracted speech features;
[0066] S14: Input the feature sequence with position encoding into the Transformer encoder;
[0067] S15: Use the CTC module to decode and obtain text information.
[0068] Since the process of voice signing of loan contracts is long and complex, higher requirements are placed on the accuracy and fault tolerance of speech recognition. By improving the process of converting voice information into text information, the transformer encoder encodes the speech feature sequence into a high-level semantic representation, and the attention mechanism helps to capture important information in the sequence. CTC decoding does not require alignment information and can process sequences of variable length, thereby improving fault tolerance.
[0069] In Example 1, step S2 includes inputting the text information into a BERT pre-trained model, and the BERT pre-trained model outputs the key information.
[0070] In Example 2, step S2 includes inputting the text information into a GPT pre-training model, and the GPT pre-training model outputs the key information. The process of extracting the key information adopts named entity recognition, and can also adopt keyword extraction.
[0071] In a preferred embodiment, step S5 includes the following sub-steps:
[0072] S51: Arrange and format the loan plan according to the definition of the rule engine, and input it into the rule engine;
[0073] S52: The rule engine inputs the result of rule judgment to the multi-layer perceptron model;
[0074] S53: The multi-layer perceptron model outputs the probability b that the loan plan is approved;
[0075] S54: When b is greater than or equal to a preset threshold, the approval is passed.
[0076] The rule engine can screen and preliminarily evaluate the applicant's information based on the set of rules set by the financial institution, and is used to handle some clear rules, such as credit score, income level and debt situation. These rules can be defined and formulated according to risk control strategies and business needs to ensure that approval decisions comply with internal regulations and policy requirements. The multi-layer perceptron algorithm can be used to handle more complex situations, such as the applicant's credit record, employment status and repayment ability. By training on a large amount of historical data, the multi-layer perceptron model can learn complex patterns and laws, thereby performing a more accurate risk assessment of the applicant. Combining the rule engine and the multi-layer perceptron model for approval can speed up the approval process, and the approval result can be returned within half a minute.
[0077] See also Figure 2 The present invention provides a loan contract voice signing device, comprising:
[0078] A speech recognition module, used for receiving speech information and converting the speech information into text information;
[0079] A natural language understanding module, used to extract key information from text information, wherein the key information includes the purpose of the loan;
[0080] The personalized loan suggestion module is used to obtain the loan type through loan purpose analysis and match the corresponding loan product; generate a loan plan including the loan amount, loan interest rate, and installment period based on the loan product and user information; convert the loan plan into voice information and feedback the loan plan to the user;
[0081] Real-time loan approval module, used to conduct real-time approval based on user information after the user confirms the loan plan;
[0082] The voice contract confirmation module is used to generate a loan contract after approval and guide the user to complete the contract signing.
[0083] The present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the voice signing method for a loan contract as described above is implemented.
[0084] In a preferred embodiment, the training process of the voiceprint vector extraction model is: use a million-scale speaker speech data set to train the model. The data set should include rich and diverse speech information to ensure that the model has a wide range of speech feature generalization capabilities. By using different data compositions, such as one-third of the training data extracted by step, and one-half of the training data on average; model structures such as CAM++, ERes2Net, etc. and loss functions such as SphereFace2loss function, ArcMargin Softmax, etc., different voiceprint vector extraction models are trained.
[0085] In the present application, unless otherwise clearly specified and limited, a first feature being “above” or “below” a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being “above”, “above”, and “above” a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being “below”, “below”, and “below” a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.
[0086] In the description of this specification, the description with reference to the terms "preferred embodiment", "further embodiment", "other embodiments" or "specific example" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.
[0087] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.
Claims
1. A method for signing a loan contract by voice, comprising the following steps: S1: receiving voice information and converting the voice information into text information; S2: extracting key information according to the text information, wherein the key information includes the purpose of the loan; S3: Analyze the loan purpose to obtain the loan type and match the corresponding loan product; retrieve user information and generate a loan plan including loan amount, loan interest rate, and installment period based on the loan product and user information; S4: converting the loan plan into voice information and feeding back the loan plan to the user; S5: After the user confirms the loan plan, real-time approval is carried out based on the user information; S6: After approval, a loan contract is generated and the user is guided to complete the contract signing. The contract signing is implemented by guiding the user to say a specified sentence, collecting the user's voice information, and using N voiceprint extraction models to extract and compare the voiceprint vectors of the user's historical voice information and the collected voice information. The N is ≥ 2, and the voice contract signing is completed after the verification is passed; Step S1 includes the following sub-steps: S11: dividing the speech information into time windows, and converting each window into a spectrogram or a Mel-spectrogram; S12: Use convolutional neural network to extract speech features; S13: adding position coding to the extracted speech features; S14: Input the feature sequence with position encoding into the Transformer encoder; S15: Use CTC module to decode and obtain text information; Step S2 includes inputting the text information into a BERT pre-training model, and the BERT pre-training model outputs the key information, or inputting the text information into a GPT pre-training model, and the GPT pre-training model outputs the key information; Step S5 includes the following sub-steps: S51: Arrange and format the loan plan according to the definition of the rule engine, and input it into the rule engine; S52: The rule engine inputs the result of rule judgment to the multi-layer perceptron model; S53: The multi-layer perceptron model outputs the probability b that the loan plan is approved; S54: When b is greater than or equal to a preset threshold, the approval is passed.
2. A loan contract voice signing method according to claim 1, characterized in that: Step S6 includes the following sub-steps: S61: extracting voiceprint vectors, using the trained N voiceprint extraction models to process the user's historical voice information and the collected voice information respectively, to obtain N voiceprint vectors respectively; S62: concatenating voiceprint vectors, concatenating the N voiceprint vectors extracted in S61 to obtain a historical voiceprint vector and a current voiceprint vector; S63: Compare the historical voiceprint vector with the current voiceprint vector to calculate the similarity; S64: Compare the similarity calculated in S63 with the preset similarity threshold. When the similarity after comparison is greater than the preset similarity threshold, it is determined that they are the same user and the verification is successful; when the similarity after comparison is less than the preset similarity threshold, the verification fails.
3. A loan contract voice signing method according to claim 2, characterized in that: Step S6 includes using blockchain anti-counterfeiting technology to record the user voice information.
4. A loan contract voice signing method according to claim 3, characterized in that: In step S64, cosine similarity is used as a similarity calculation method, and the preset similarity threshold a satisfies 0.7≤a≤0.
9.
5. A loan contract voice signing device, characterized in that: include: A speech recognition module, used for receiving speech information and converting the speech information into text information; A natural language understanding module, used to extract key information from text information, wherein the key information includes the purpose of the loan; The personalized loan suggestion module is used to obtain the loan type through loan purpose analysis and match the corresponding loan product; generate a loan plan including the loan amount, loan interest rate, and installment period based on the loan product and user information; convert the loan plan into voice information and feedback the loan plan to the user; Real-time loan approval module, used to conduct real-time approval based on user information after the user confirms the loan plan; The voice contract confirmation module is used to generate a loan contract after approval and guide the user to complete the contract signing.
6. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the voice signing method for a loan contract as described in any one of claims 1 to 4 is implemented.
