A high-fidelity AI voice authentication method based on biometric recognition

Abstract

The present invention relates to the field of voice authentication technology, and in particular to a high-fidelity AI voice authentication method based on biometric identification, comprising the steps of: S1, generating an excitation schedule, a keyword data set, a security status type value set, multiple operation voice information sets and multiple security schemes; S2, segmenting multiple data blocks from multiple detection data, and respectively judging whether each valid data meets the conditions for triggering a collection device group; S3, outputting operation guidance voice information, triggering a collection device group to obtain voice data and auxiliary biometric data; S4, identifying face image data through a trained AI face recognition model, and outputting a state feature value; S5, identifying voice data through a trained AI voice recognition model, and outputting text format data; S6, performing voiceprint matching, and outputting a voiceprint matching value; S7, performing keyword matching, and executing a corresponding security scheme.

Description

Technical Field

[0001] The present invention relates to the field of voice authentication technology, and in particular to a high-fidelity AI voice authentication method based on biometric recognition. Background Art

[0002] Existing security authentication methods include password authentication, biometric authentication, etc. The so-called biometric authentication is to collect biometrics such as fingerprints, irises and voices, and compare them with pre-stored biometric information to complete security authentication; single authentication methods are easy to crack, so fusion authentication methods have emerged, especially in some application fields with high security requirements. Fusion authentication methods include fusing multiple biometrics and improving the reliability of authentication through multiple verifications. In reality, voice authentication methods include text-based and independent text. Text-based methods recognize keywords by reading a fixed text and perform voice authentication; independent text methods do not use text and only authenticate through sound quality.

[0003] When performing voice authentication, the authenticator may be alone or there may be other people around him. How to judge whether other people are safe? How to judge whether the authenticator is performing voice authentication without coercion? How to make an alarm while ensuring the safety of the authenticator? For this purpose, I propose a voice authentication method. Summary of the invention

[0004] The present invention aims to provide a technical solution to solve the problems in the background technology.

[0005] The technical solution of the present invention is: a high-fidelity AI voice authentication method based on biometric recognition, the method comprising the following steps:

[0006] S1, generating an excitation schedule, a keyword data set, a security status type value set, multiple operation voice information sets and multiple security solutions, sending the excitation schedule, keyword data set, security status type value set, operation voice information set and security solutions to cloud storage, stimulating multiple detection devices according to the excitation schedule, and obtaining multiple detection data fed back by the multiple detection devices;

[0007] S2, dividing a plurality of data blocks from the plurality of detection data, extracting a plurality of valid data from the plurality of data blocks, and respectively judging whether each valid data satisfies the condition of triggering the acquisition device group, if so, proceeding to the next step; otherwise, proceeding to step S1;

[0008] S3, outputting operation guidance voice information, triggering the collection device group to obtain voice data and auxiliary biometric data, storing the voice data and auxiliary biometric data locally and generating data attachments to send to cloud storage; the auxiliary biometric data includes one or more of face image data, iris data and fingerprint data;

[0009] S4, recognize the facial image data through the trained AI face recognition model and output the state feature value;

[0010] S5. Recognize the voice data through the trained AI voice recognition model and output the data in text format;

[0011] S6, perform voiceprint matching and output the voiceprint matching value. If the voiceprint matching value is within the set threshold range, proceed to step S7; otherwise, generate a prompt message and proceed to step S3;

[0012] S7, perform keyword matching and execute corresponding security solutions;

[0013] The keyword data set includes multiple types of data subsets; the security status type set includes multiple status feature values.

[0014] Preferably, the method further includes the following steps: matching the plurality of security schemes with the plurality of types of data subsets; matching the state feature values ​​with the corresponding data subsets, and matching the plurality of operation voice information sets with the security state type value sets.

[0015] Preferably, the security status type value set is A={I, J, K}, where I, J, K are status characteristic values, all are positive integers, and are arranged in descending order; the acquisition device group includes multiple high-fidelity audio acquisition devices and multiple high-definition cameras.

[0016] Preferably, step S4 comprises the following steps:

[0017] Use the VGGFace face detection model to identify face image data. If the face image includes multiple face information, the corresponding state feature value I is output;

[0018] Compare the multiple facial information after recognition with the pre-stored facial information respectively;

[0019] If at least one face information matches the pre-stored face information, the state feature value is changed to J and then output; if multiple face information do not match, the state feature value is changed to K and then output.

[0020] Preferably, step S7 includes:

[0021] Obtain state characteristic values; retrieve corresponding data subsets according to the state characteristic values;

[0022] Keywords in the text information are extracted, matched with the keywords in the data subset, and the matching degree is output; if the matching degree is within a preset language matching threshold range, the security solution corresponding to the data subset is executed.

[0023] Preferably, the step of stimulating a plurality of detection devices according to the stimulation schedule and obtaining a plurality of detection data fed back by the plurality of detection devices comprises the following steps:

[0024] Acquire the detection device information in the excitation schedule and the excitation time point corresponding to the detection device information;

[0025] When the excitation time point arrives, the corresponding detection device is excited;

[0026] Sending out a variety of detection signals through a plurality of detection devices, and receiving feedback detection signals;

[0027] The multiple detection signals include one or more of ultrasonic signals and laser signals.

[0028] Preferably, the acquiring of a plurality of detection data fed back by a plurality of detection devices and dividing the plurality of detection data into a plurality of data blocks comprises:

[0029] The detection data is divided into: distance type data block and thermal radiation type data block according to type;

[0030] The method of extracting a plurality of valid data from a plurality of data blocks is as follows: smp Sampling is performed from the distance type data block and the thermal radiation type data block. If sampling data cannot be obtained, the data block is an invalid data block; otherwise, it is a valid data block;

[0031] The step of respectively judging whether each valid data satisfies the condition of triggering the acquisition device group comprises:

[0032] Obtain the duration T1 of continuously receiving the returned ultrasonic signal; obtain the duration T2 of continuously receiving the returned laser signal;

[0033] Setting a time condition, wherein the time condition is the time T during which the detection device continuously detects the feedback signal;

[0034] When T1>T, T2>T, the acquisition device group is triggered, otherwise it is not triggered.

[0035] Preferably, matching the state characteristic value with the corresponding data subset includes:

[0036] Set the keyword data set to Q = {Q1, Q2, Q3} where Q1, Q2, Q3 are data subsets;

[0037] Data subset Q1 matches the state eigenvalue I, data subset Q2 matches the state eigenvalue J, and data subset Q3 matches the state eigenvalue K;

[0038] Matching multiple action voice information sets to security status type value sets includes:

[0039] Set multiple operation voice information sets as: K1, K2, K3, each operation voice information set stores different voice data;

[0040] The operation voice information set K1 matches the state feature value I, the operation voice information set K2 matches the state feature value J, and the operation voice information set K3 matches the state feature value K;

[0041] The keyword data set is composed of multiple types of keywords, and the keywords are used to identify the authorization status and security status of personnel; the multiple types of keywords are stored in different data subsets according to different types;

[0042] The security solutions corresponding to data subsets Q1, Q2, and Q3 are security solution one, security solution two, and security solution three, respectively. The security solution one is the default solution and serves as a preparation for accessing solution two. In the security solution two, if a person who matches the pre-stored facial information is matched through voiceprint, if the voice information matches the keyword in Q2, an alarm operation is performed. The security solution three sends the data attachment backed up in the cloud to the user end and notifies the personnel to process it. The AI ​​speech recognition model includes an acoustic module, a language module, and a voiceprint recognition model. The acoustic module adopts a CNN-CTC model. The speech module includes a CBHG and / or transformer model, and the training data sets include stc and primewords.

[0043] The present invention also provides an electronic device, comprising a processor, a memory connected to the processor, and a communication module, and the electronic device executes the high-fidelity AI voice authentication method based on biometric recognition.

[0044] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the high-fidelity AI voice authentication method based on biometric recognition.

[0045] Beneficial Effects

[0046] 1. In the present invention, the voiceprint recognition model is used to perform voiceprint recognition on the answering voice. If the voiceprint matching value output by the voiceprint model is within the preset voiceprint matching threshold, it means that the voiceprint segment belongs to the authorized person. The voice recognition model is used to convert the content of the answer into text format information. If the text information contains keywords in the data subset Q2, it means that the authorized person is under duress at this time, then the security plan 2 is executed, and the police are notified to the networked public security organs, and the collected facial information is sent at the same time.

[0047] 2. In the present invention, the keyword data set includes various preset keywords. In this embodiment, when the conditions for triggering the acquisition device are met, the devices in the acquisition device group perform facial image acquisition. If the acquired facial image includes multiple facial information, the multiple facial information are separately identified and compared with the pre-stored facial information. If there is an authorized face among the multiple faces, the other facial information is not recorded, that is, the match with the pre-stored face fails. At this time, the state feature value J is output, the corresponding data subset Q2 and the operation voice information set K2 are obtained, and the voice data in the operation voice information set K2 is output, and the authenticated person makes a corresponding answer based on the voice data.

[0048] 3. In the present invention, an excitation schedule, a keyword data set, a security status type value set and multiple security schemes are first generated. The excitation schedule consists of a detection device identification code and an excitation time. When the excitation time arrives, the corresponding detection device is excited. The detection devices include laser distance sensors, infrared detectors and ultrasonic radars, etc. The excitation time interval and excitation time length of each device can be set as needed. For example, the infrared detector is a passive detection device and can be set to start for a long time. During the working time of the infrared detector, other detection devices are started at intervals to improve the accuracy of detecting approaching objects.

[0049] 4. In the present invention, if multiple facial information do not match, the state feature value is changed to K and then output, indicating that the collected facial information is not the facial information of the authorized person. At this time, the collected facial image information is sent to the cloud, and the authorized person can obtain the corresponding facial information to determine whether it is a dangerous person. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 Shown is a flow chart of a high-fidelity AI voice authentication method based on biometric recognition according to the present invention. DETAILED DESCRIPTION

[0051] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be applied to other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0052] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0053] refer to Figure 1 The present invention provides a technical solution: a high-fidelity AI voice authentication method based on biometric recognition, the method comprising the following steps:

[0054] S1, generating an excitation schedule, a keyword data set, a security status type value set, multiple operation voice information sets and multiple security solutions, sending the excitation schedule, keyword data set, security status type value set, operation voice information set and security solutions to cloud storage, stimulating multiple detection devices according to the excitation schedule, and obtaining multiple detection data fed back by the multiple detection devices;

[0055] S2, dividing a plurality of data blocks from the plurality of detection data, extracting a plurality of valid data from the plurality of data blocks, and respectively judging whether each valid data satisfies the condition of triggering the acquisition device group, if so, proceeding to the next step; otherwise, proceeding to step S1;

[0056] S3, outputting operation guidance voice information, triggering the collection device group to obtain voice data and auxiliary biometric data, storing the voice data and auxiliary biometric data locally and generating data attachments to send to cloud storage; the auxiliary biometric data includes one or more of face image data, iris data and fingerprint data;

[0057] S4. Recognize the voice data through the trained AI voice recognition model and output the data in text format;

[0058] S5. Recognize the facial image data through the trained AI face recognition model and output the state feature value;

[0059] S6, perform voiceprint matching and output the voiceprint matching value. If the voiceprint matching value is within the set threshold range, proceed to step S7; otherwise, generate a prompt message and proceed to step S3;

[0060] S7, perform keyword matching and execute corresponding security solutions;

[0061] The keyword data set includes multiple types of data subsets; the security status type set includes multiple status feature values.

[0062] In some preferred embodiments, the following steps are also included: matching multiple security schemes with multiple types of data subsets; matching the state feature values ​​with corresponding data subsets, and matching multiple operation voice information sets with security state type value sets.

[0063] The security status type value set is A={I, J, K}, where I, J, and K are status characteristic values, all of which are positive integers and are arranged in descending order; the acquisition device group includes multiple high-fidelity audio acquisition devices and multiple high-definition cameras to obtain high-fidelity audio data and high-definition facial images, which is conducive to voice authentication and face authentication.

[0064] Step S4 includes the following steps: using the VGGFace face detection model to identify face image data, if the face image includes multiple face information, outputting the corresponding state feature value I; comparing the multiple face information after identification with the pre-stored face information respectively; if at least one face information matches the pre-stored face information, changing the state feature value to J before output; if multiple face information do not match, changing the state feature value to K before output.

[0065] The step S7 includes: obtaining a state feature value; retrieving a corresponding data subset according to the state feature value; extracting keywords in the text information, matching them with the keywords in the data subset, and outputting a matching degree; and executing a security solution corresponding to the data subset if the matching degree is within a preset language matching threshold.

[0066] The method of stimulating multiple detection devices according to the stimulation schedule and obtaining multiple detection data fed back by the multiple detection devices includes the following steps: obtaining detection device information in the stimulation schedule and multiple stimulation time points corresponding to the detection device information; stimulating corresponding detection devices when the stimulation time points arrive; sending multiple detection signals through the multiple detection devices and receiving feedback detection signals; obtaining multiple detection data fed back by the multiple detection devices and dividing the detection data into distance type data blocks and thermal radiation type data blocks according to the types; and using a preset time interval T smp = 0.1s. A fixed number of samples are taken from the distance type data block and the thermal radiation type data block. If there are multiple cases where the sampled data cannot be obtained, the data block is an invalid data block; otherwise, it is a valid data block.

[0067] In some preferred embodiments, the present invention also provides a method for triggering a collection device group, including setting a distance parameter value L. In the distance type data (laser distance sensor data), if the measured distance is greater than L, the collection is not triggered, and if the measured distance is less than L, the collection is triggered, thereby triggering the collection device group to collect data in advance, such as obtaining facial image information in advance, and performing facial recognition in advance, which greatly reduces the reaction time for starting authentication.

[0068] Obtain the duration T1 of continuously receiving the returned ultrasonic signal and the duration T2 of continuously receiving the returned laser signal in the distance type data block; set the time T for the detection device to continuously detect the feedback signal; trigger the acquisition device group when T1>T, T2>T, otherwise not trigger.

[0069] Matching the state feature value with the corresponding data subset includes: setting the keyword data set to Q={Q1, Q2, Q3}, where Q1, Q2, Q3 are data subsets; matching the data subset Q1 with the state feature value I, matching the data subset Q2 with the state feature value J, and matching the data subset Q3 with the state feature value K.

[0070] Matching multiple operation voice information sets with security status type value sets includes: setting multiple operation voice information sets as: K1, K2, K3, each operation voice information set stores different voice data; operation voice information set K1 matches the status characteristic value I, operation voice information set K2 matches the status characteristic value J, and operation voice information set K3 matches the status characteristic value K; the keyword data set is composed of multiple types of keywords, and the keywords are used to identify the authorization status and security status of personnel; and the multiple types of keywords are stored in different data subsets according to different types.

[0071] In some preferred embodiments, multiple security schemes corresponding to data subsets Q1, Q2, and Q3 are respectively set as security scheme one, security scheme two, and security scheme three. The security scheme one is the default scheme, which serves as a preparation for accessing scheme two. In the security scheme two, if a person who matches the pre-stored facial information passes the voiceprint match, if the voice information matches the keyword in Q2, an alarm operation is performed; and the operation voice information (voice problem) in K2 is output in a loop to increase the verification time. The security scheme three sends the data attachment backed up in the cloud to the user end and notifies the personnel to process it. The AI ​​speech recognition model includes an acoustic module, a language module, and a voiceprint recognition model, and the acoustic module adopts a CNN-CTC model. The speech module includes a CBHG and / or transformer model, and the training data set includes stc and primewords.

[0072] The present invention also provides an electronic device, which includes a processor and a memory and a communication module connected to the processor, and the electronic device executes the high-fidelity AI voice authentication method based on biometric recognition.

[0073] The present invention also provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the high-fidelity AI voice authentication method based on biometric recognition.

[0074] Principle description: First, generate an excitation schedule, a keyword data set, a security status type value set, and multiple security schemes. The excitation schedule consists of a detection device identification code and an excitation time. When the excitation time arrives, the corresponding detection device is excited. The detection devices include laser distance sensors, infrared detectors, and ultrasonic radars. Each device is set with multiple excitation time points. The time interval and excitation time length of each excitation time point can be set as needed or by the processor clock. For example, the infrared detector is a passive detection device and can be set to start for a long time. During the working time of the infrared detector, other detection devices are started at a time interval of multiple clock cycles, which can improve the accuracy of detecting the approach of objects. When the acquisition trigger conditions are met, the acquisition device group can be started to collect in advance;

[0075] The keyword data set includes various preset keywords. In this embodiment, when the conditions for triggering the acquisition device are met, the devices in the acquisition device group acquire face images. If the acquired face image includes multiple face information, the multiple face information are identified separately and compared with the pre-stored face information; specifically:

[0076] The pre-trained VGGFace model is first used to detect and align faces in the input image with faces in the pre-stored images. The model then converts these facial images into high-dimensional feature vectors, and finally compares these vectors using Euclidean distance or other similarity metrics for recognition.

[0077] If there is one authorized face among multiple faces, the information of other faces is not recorded, that is, the match with the pre-stored face fails. At this time, the state feature value J is output, the corresponding data subset Q2 and the operation voice information set K2 are obtained, and the voice data in the operation voice information set K2 is output, and the authenticated person makes a corresponding answer based on the voice data.

[0078] The voiceprint recognition model is used to perform voiceprint recognition on the answering voice. The voiceprint model in this embodiment implements voiceprint recognition based on the MFCC feature template matching algorithm. The feature template is the pre-stored voice feature of the authorized person. The output is the voiceprint matching value. If the voiceprint matching value output by the voiceprint model is within the preset voiceprint matching threshold, it means that the voiceprint segment is that of the authorized person. The principle of the model is existing technology and will not be repeated here.

[0079] The answer content is converted into text format information through the speech recognition model, and then keyword extraction is performed. If the text information contains keywords in the data subset Q2, it means that the authorized person is under duress, and security plan 2 is executed.

[0080] In some embodiments, if multiple facial information do not match, the state feature value is changed to K and then output, indicating that the collected facial information is not the facial information of authorized personnel. At this time, the collected facial image information is sent to the cloud, and the authorized personnel can obtain the corresponding facial information to determine whether it is a dangerous person.

[0081] The embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. The embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part, and / or installed from a removable medium. When the computer program is executed by the central processing unit (CPU), the above-mentioned functions defined in the method of the present application are executed. It should be noted that the above-mentioned computer-readable medium of the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of an electrical, magnetic, optical, electromagnetic, infrared segment, or semiconductor, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection with one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a 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, apparatus, or device. In the present application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries a computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, electrical wire, optical cable, RF, etc., or any suitable combination of the foregoing.

[0082] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0083] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functional and structural principles of the present invention have been demonstrated and explained in the embodiments. Without departing from the principles, the implementation methods of the present invention may be deformed or modified in any way.

Claims

1. A high-fidelity AI voice authentication method based on biometric recognition, characterized in that: The method comprises the following steps: S1, generating an excitation schedule, a keyword data set, a security status type value set, multiple operation voice information sets and multiple security solutions, sending the excitation schedule, keyword data set, security status type value set, operation voice information set and security solutions to cloud storage, stimulating multiple detection devices according to the excitation schedule, and obtaining multiple detection data fed back by the multiple detection devices; S2, dividing a plurality of data blocks from the plurality of detection data, extracting a plurality of valid data from the plurality of data blocks, and respectively judging whether each valid data satisfies the condition of triggering the acquisition device group, if so, proceeding to the next step, otherwise proceeding to step S1; S3, outputting operation guidance voice information, triggering the collection device group to obtain voice data and auxiliary biometric data, storing the voice data and auxiliary biometric data locally and generating data attachments to send to cloud storage; the auxiliary biometric data includes one or more of face image data, iris data and fingerprint data; S4, recognize the facial image data through the trained AI face recognition model and output the state feature value; S5. Recognize the voice data through the trained AI voice recognition model and output the data in text format; S6. Perform voiceprint matching and output the voiceprint matching value; If the voiceprint matching value is within the set threshold range, the process proceeds to step S7; otherwise, a prompt message is generated and the process proceeds to step S3; S7, perform keyword matching and execute corresponding security solutions; The keyword data set includes multiple types of data subsets; the security status type value set includes multiple status feature values ​​that are all positive integers and arranged in descending order; the acquisition device group includes multiple high-fidelity audio acquisition devices and multiple high-definition cameras; The step of dividing the plurality of detection data into a plurality of data blocks comprises: dividing the detection data into distance type data blocks and thermal radiation type data blocks according to the type; The extracting of multiple valid data from multiple data blocks includes: smp Perform multiple sampling from the distance type data block and the thermal radiation type data block. If there are multiple cases where sampling data cannot be obtained, the data block is an invalid data block; otherwise, it is a valid data block; The step of respectively judging whether each valid data satisfies the condition of triggering the acquisition device group includes: Obtain the duration T1 of continuously receiving the returned ultrasonic signal; obtain the duration T2 of continuously receiving the returned laser signal; Setting a time condition, wherein the time condition is the time T during which the detection device continuously detects the feedback signal; When T1>T, T2>T, the acquisition device group is triggered, otherwise it is not triggered; Matching the state feature value with the corresponding data subset includes: setting the keyword data set to ,in is a subset of data; Data subset and state eigenvalue Matching, Data Subset and state eigenvalue Matching, Data Subset and state eigenvalue match; Matching multiple action voice information sets to security status type value sets includes: Set up multiple voice message sets , , ,Each operation voice information set stores multiple different voice data; Operation voice information set and state eigenvalue Matching, operating voice information set and state eigenvalue Matching, operating voice information set and state eigenvalue match; The keyword data set is composed of multiple types of keywords, and the keywords are used to identify the authorization status and security status of personnel; the multiple types of keywords are respectively stored in different data subsets.

2. According to claim 1, a high-fidelity AI voice authentication method based on biometric recognition is characterized in that: The following steps are also included: Matching a plurality of the security schemes with a plurality of types of data subsets; and matching the data subsets The corresponding security solutions are security solution 1, security solution 2 and security solution 3, respectively. The security solution 1 is the default solution and serves as a preparation for access solution 2. In the second security solution, if the person who matches the pre-stored face information matches the voiceprint, if the voice information matches When keywords are found in the text, an alarm is triggered; the security solution three sends the data attachments backed up in the cloud to the user end and notifies the personnel to process; the AI ​​speech recognition model includes an acoustic module, a language module and a voiceprint recognition model, and the acoustic module adopts a CNN-CTC model; the language module includes a CBHG and / or transformer model, and the training data sets include stc and primewords.

3. According to claim 1, a high-fidelity AI voice authentication method based on biometric recognition is characterized in that: The security status type value set is ,in is the state characteristic value.

4. According to claim 1, a high-fidelity AI voice authentication method based on biometric recognition is characterized in that: The step S5 comprises the following steps: Use the VGGFace face detection model to identify face image data. If the face image data includes multiple face information, the corresponding state feature value is output. ; Compare the multiple facial information after recognition with the pre-stored facial information respectively; If at least one face matches the pre-stored face, the status feature value is changed to After output; If multiple face information does not match, change the status feature value After output.

5. According to claim 1, a high-fidelity AI voice authentication method based on biometric recognition is characterized in that: The step S7 comprises: Get state characteristic values; Retrieve the corresponding data subset according to the state characteristic value; Extracting keywords from the text information, matching them with keywords in the data subset, and outputting the degree of matching; If the matching degree is within the preset language matching threshold range, the security solution corresponding to the data subset is executed.

6. According to claim 1, a high-fidelity AI voice authentication method based on biometric recognition is characterized in that: The method of stimulating a plurality of detection devices according to the stimulation schedule and obtaining a plurality of detection data fed back by the plurality of detection devices comprises the following steps: Acquire detection device information in an excitation schedule and a plurality of excitation time points corresponding to the detection device information; When the excitation time point arrives, the corresponding detection device is excited; Multiple detection signals are sent out through multiple detection devices, and feedback detection signals are received; the multiple detection signals include one or more of ultrasonic signals and laser signals.

7. An electronic device, comprising a processor, a memory connected to the processor, and a communication module, characterized in that: The electronic device executes a high-fidelity AI voice authentication method based on biometric recognition as described in any one of claims 1-6.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a high-fidelity AI voice authentication method based on biometric recognition as described in any one of claims 1 to 6.

Images