Voiceprint recognition method, device, medium and equipment

Abstract

The present application discloses a voiceprint recognition method, device, medium and equipment, and relates to the field of speech recognition technology. The method includes: obtaining speech information to be recognized; converting the speech information to be recognized to obtain candidate speech information; inputting the speech information to be recognized and the candidate speech information into a voiceprint model respectively, performing voiceprint verification processing, and obtaining a first voiceprint verification result corresponding to the speech information to be recognized and a second voiceprint verification result corresponding to the candidate speech information; determining the voiceprint recognition result for the speech information to be recognized based on the first voiceprint verification result and the second voiceprint verification result. The technical solution provided by the present application can improve the defense performance against adversarial attacks and enhance the reliability and security of voiceprint recognition.

Description

Technical Field

[0001] The present application relates to the field of speech recognition technology, and specifically to voiceprint recognition methods, devices, media and equipment. Background Art

[0002] Artificial Intelligence (AI) is a comprehensive technology in computer science. By studying the design principles and implementation methods of various intelligent machines, it enables machines to have the functions of perception, reasoning and decision-making. Artificial Intelligence technology is a comprehensive discipline that covers a wide range of fields, such as natural language processing, machine learning, deep learning and other major directions. With the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0003] At present, voiceprint recognition based on deep neural networks has achieved very good recognition results and has been widely used in financial payment, remote biometric authentication, security, public security investigation and forensic identification. However, unprotected voiceprint recognition systems have great security risks and may be attacked by recording playback, speech synthesis, speech conversion, adversarial sample attacks, etc. The defense and detection technology for adversarial sample attacks is still in its infancy. The existing solutions mainly conduct adversarial training or introduce new network structures for active defense. These methods require more computing resources or increase the number of model parameters, and the effect of adversarial defense still has a lot of room for improvement. Summary of the invention

[0004] In order to improve the reliability and security of voiceprint recognition, this application provides a voiceprint recognition method, device, medium and equipment. The technical solution is as follows:

[0005] In a first aspect, the present application provides a voiceprint recognition method, the method comprising:

[0006] Acquire voice information to be recognized;

[0007] Converting the voice information to be recognized to obtain candidate voice information;

[0008] Inputting the voice information to be recognized and the candidate voice information into a voiceprint model respectively, performing voiceprint verification processing, and obtaining a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information;

[0009] A voiceprint recognition result for the voice information to be recognized is determined according to the first voiceprint verification result and the second voiceprint verification result.

[0010] In a second aspect, the present application provides a voiceprint recognition device, the device comprising:

[0011] An acquisition module, used for acquiring voice information to be recognized;

[0012] A conversion module, used for converting the voice information to be recognized to obtain candidate voice information;

[0013] A voiceprint verification module, used to input the voice information to be recognized and the candidate voice information into a voiceprint model respectively, perform voiceprint verification processing, and obtain a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information;

[0014] The recognition module is used to determine the voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result.

[0015] In a third aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by a processor to implement a voiceprint recognition method as described in the first aspect.

[0016] In a fourth aspect, the present application provides a computer device, comprising a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement a voiceprint recognition method as described in the first aspect.

[0017] In a fifth aspect, the present application provides a computer program product, wherein the computer program product includes computer instructions, and when the computer instructions are executed by a processor, a voiceprint recognition method as described in the first aspect is implemented.

[0018] The voiceprint recognition method, device, medium and equipment provided in this application have the following technical effects:

[0019] The solution provided by the present application converts the voice information to be recognized to obtain candidate voice information, and inputs the voice information to be recognized and the candidate voice information into the voiceprint model together for voiceprint verification processing, thereby obtaining the final voiceprint recognition result corresponding to the voice information to be recognized based on the first voiceprint verification result corresponding to the voice information to be recognized and the second voiceprint verification result corresponding to the candidate voice information. When applied to adversarial defense and detection, the technical solution provided by the present application is improved in the input preprocessing stage, and there is no need to conduct adversarial training on the original voiceprint model, nor is there any need to add a defense module for adversarial sample attacks. It only needs to add conversion processing of the voice to be recognized and judgment processing of multiple voiceprint verification results in the original voiceprint verification stage. The overall solution is simple and easy to implement, and has high versatility; and the technical solution provided by the present application can implement effective defense and detection against white-box attacks or black-box attacks based on adversarial samples without affecting the performance of the original voiceprint model, thereby improving the security and reliability of voiceprint recognition.

[0020] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 It is a schematic diagram of an implementation environment of a voiceprint recognition method provided in an embodiment of the present application;

[0023] Figure 2 It is a flow chart of a voiceprint recognition method provided in an embodiment of the present application;

[0024] Figure 3 This is a flow chart of converting speech information to be recognized provided by an embodiment of the present application;

[0025] Figure 4 It is a schematic diagram of voiceprint recognition based on a voiceprint model provided in an embodiment of the present application;

[0026] Figure 5 It is a schematic diagram of a process for determining a voiceprint verification result provided by an embodiment of the present application;

[0027] Figure 6 This is a schematic diagram of a process for making a decision based on a voiceprint verification result provided by an embodiment of the present application;

[0028] Figure 7 It is a schematic diagram of a process of performing adversarial detection based on voiceprint verification results provided by an embodiment of the present application;

[0029] Figure 8 It is a schematic diagram of another flow of voiceprint recognition provided by an embodiment of the present application;

[0030] Fig. 9 is a schematic diagram of a voiceprint recognition device provided in an embodiment of the present application;

[0031] Fig.10 It is a schematic diagram of the hardware structure of a device for implementing a voiceprint recognition method provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making. Artificial intelligence technology is an interdisciplinary subject that covers a wide range of fields, including both hardware-level technology and software-level technology. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics.

[0033] The solution provided in the embodiment of this application involves artificial intelligence speech technology. The key technologies of speech technology include automatic speech recognition technology (ASR), speech synthesis technology (TTS) and voiceprint recognition technology. Enabling computers to listen, see, speak and feel is the development direction of human-computer interaction in the future, among which speech has become one of the most promising human-computer interaction methods in the future.

[0034] The solution provided in the embodiment of the present application can be deployed in the cloud, which also involves cloud technology, etc.

[0035] Cloud technology: refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology based on the cloud computing business model application, which can form a resource pool and be used on demand, flexible and convenient. The back-end services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, each item may have its own identification mark in the future, and all need to be transmitted to the back-end system for logical processing. Data of different levels will be processed separately. All kinds of industry data need strong system backing support, so cloud technology needs to be supported by cloud computing. Cloud computing is a computing model that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services as needed. The network that provides resources is called "cloud". The resources in the "cloud" are infinitely expandable in the eyes of users, and can be obtained at any time, used on demand, expanded at any time, and paid for by use. As a cloud computing basic capability provider, a cloud computing resource pool platform will be established, referred to as a cloud platform, generally referred to as Infrastructure as a Service (IaaS), and various types of virtual resources will be deployed in the resource pool for external customers to choose to use. The cloud computing resource pool mainly includes: computing devices (which can be virtualized machines, including operating systems), storage devices, and network devices.

[0036] The workflow of a typical voiceprint recognition system mainly involves two steps: "voiceprint reservation (registration)" and "voiceprint verification (test)". Reservation is to convert the user's voice into a speaker representation vector and store it. Voiceprint verification determines whether an unknown test voice comes from a specified speaker. The system converts the test voice into a speaker representation vector and compares it with the reserved user voice. If it is greater than a pre-set threshold, it is determined to belong to the same speaker; otherwise, if the score is less than the threshold, it is determined that they do not belong to the same speaker.

[0037] Voiceprint recognition has been widely used in financial payment, remote biometric authentication, security, public security investigation and forensic identification. Unprotected voiceprint systems have great security risks. After the user completes the registration, during the voiceprint verification step, voiceprint recognition may have security-related issues such as recording playback attack, speech synthesis attack, speech conversion attack, and adversarial sample attack. Recording playback attack, speech synthesis attack, and speech conversion attack are secondary recordings or synthesized sounds. Due to the frequency response of the device or the insufficient performance of the synthesis model, some frequency domains of the voice data used for the attack are missing and distorted, which will have different characteristics from real people speaking. Through a large number of positive and negative sample learning, the computer can easily distinguish whether the voice is synthesized from a recording or spoken by a real person.

[0038] Adversarial sample attacks, especially those in white-box situations, are more difficult to detect and defend against than the above three attacks. So far, the reason why deep learning models are vulnerable to adversarial sample attacks is still an open research topic and lacks a complete theoretical system. This problem also restricts the further development of deep learning. Most of the current research on adversarial samples is concentrated in the field of computer vision. Since speech signals are non-stationary signals, research on adversarial sample attacks and defenses for voiceprint recognition is still in its infancy. Especially in the defense against adversarial sample attacks, there are not many complete voiceprint-related research and solutions. At present, the defense methods against adversarial attacks are mainly divided into the following three solutions:

[0039] 1. Adversarial training: During each model training process, the model is retrained by injecting adversarial samples into the training set;

[0040] 2. Preprocess the input data: transform the input to make it difficult for attackers to calculate the gradient of the model, thereby achieving the purpose of defending against adversarial attacks;

[0041] 3. Distill the model: Use knowledge distillation to reduce the size of the network gradient and improve the ability to detect small perturbation adversarial samples.

[0042] However, the above solution has the following disadvantages:

[0043] 1. Adversarial training requires generating adversarial samples during model training, and then using the generated adversarial samples as input data to train the original network. Both processes require a lot of computing resources and computing time, and the training models obtained in the end can mostly only defend against specific adversarial sample algorithms. If the attacker modifies the adversarial attack algorithm, the model's defense capability will be greatly reduced.

[0044] 2. Most of the current defense methods for preprocessing input data use a generated neural network model to reconstruct the input data, such as denoising the adversarial samples based on variational auto-encoders (VAE) or generative adversarial networks (GAN), so that the data output by the denoising model is closer to the original noise-free data. These methods require the introduction of new neural networks, which increases the number of parameters in the voiceprint recognition system; and the voiceprint recognition system needs to spend more computing time on denoising the original audio during the inference process, and this method is difficult to resist white-box attacks.

[0045] 3. Distillation and regularization of the model may greatly damage the recognition performance and robustness of the voiceprint model, reducing the recognition performance of real samples that have not been attacked.

[0046] 4. The current defense strategies for adversarial samples in voiceprint recognition systems are mostly borrowed and transferred from computer vision research. Compared with image signals, speech signals are non-stationary signals, and the adversarial sample attack and defense of voiceprint recognition systems are still in the initial stage. These methods on images may not be applicable to speech data.

[0047] In order to improve the reliability and security of voiceprint recognition, the embodiments of the present application provide voiceprint recognition methods, devices, media and equipment. The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application. Examples of the embodiments are shown in the drawings, where the same or similar numbers throughout represent the same or similar elements or elements with the same or similar functions.

[0048] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0049] In order to facilitate understanding of the technical solutions described in the embodiments of the present application and the technical effects produced therefrom, the embodiments of the present application explain the relevant professional terms involved:

[0050] Voiceprint: Voiceprint is a behavioral characteristic. Since the tongue, oral cavity, nasal cavity, vocal cords, lungs and other organs used by each person when speaking are different in size and shape, and considering the differences in age, personality, language habits and other factors, it can be said that each speaker's voiceprint is unique. In terms of data, voiceprint can be represented by the sound wave spectrum that carries speech information displayed by electroacoustic instruments. Modern scientific research shows that voiceprints are not only specific, but also relatively stable.

[0051] Voiceprint Recognition: Voiceprint Recognition, VPR; Voiceprint recognition is also called speaker recognition (SRE), which is a type of biometric recognition technology. It is a technology that automatically identifies the identity of a speaker based on voice parameters ("voiceprint") that reflect the speaker's physiological and behavioral characteristics in the voice signal. According to the specific business scenario and recognition target, voiceprint recognition can be divided into voiceprint identification and voiceprint confirmation. Voiceprint identification refers to determining which person in the target speaker model set the voice to be recognized belongs to, which is a 1:N selection problem; while voiceprint confirmation is to determine whether the voice to be recognized comes from the target speaker it declares, which is a 1:1 judgment problem.

[0052] Adversarial sample attack: refers to artificially adding imperceptible perturbations to normal data samples to obtain newly generated adversarial samples, causing the model to make incorrect judgments about the newly generated samples.

[0053] White-box attack: The attacker knows the detailed information of the model, such as data preprocessing methods, model structure and model parameters. In some cases, the attacker can also master part or all of the training data information, such as loss function and gradient information. In extreme cases, the attacker even knows the defense strategy of the model.

[0054] Black box attack: The attacker does not know the key details of the model. The attacker can only access the input and output links and cannot substantially access any internal operations and data.

[0055] It is understandable that in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0056] See also Figure 1 , which is a schematic diagram of an implementation environment of a voiceprint recognition method provided in an embodiment of the present application, such as Figure 1 As shown, the implementation environment may include at least a client 01 and a server 02 .

[0057] Specifically, the client 01 may include devices such as smart phones, desktop computers, tablet computers, laptop computers, vehicle terminals, digital assistants, smart wearable devices, monitoring devices, and voice interaction devices, and may also include software running in the device, such as web pages provided by some service providers to users, or applications provided by these service providers to users. Specifically, the client 01 may be used to obtain voice information to be recognized.

[0058] Specifically, the server 02 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server 02 may include a network communication unit, a processor, and a memory, etc. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application does not limit this. Specifically, the server 02 can be used to convert the voice information to be recognized sent by the client 01 to obtain candidate voice information; and input the voice information to be recognized and the candidate voice information into the voiceprint model together, perform voiceprint verification processing, and obtain the corresponding first voiceprint verification result and second voiceprint verification result respectively, and finally determine the voiceprint recognition result corresponding to the voice information to be recognized based on multiple voiceprint verification results.

[0059] The embodiments of the present application can also be implemented in combination with cloud technology. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software and network in a wide area network or a local area network to realize the calculation, storage, processing and sharing of data. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology based on the cloud computing business model application. Cloud technology needs to be supported by cloud computing. Cloud computing is a computing model that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services as needed. The network that provides resources is called a "cloud". Specifically, the server 02 and the database are located in the cloud, and the server 02 can be a physical machine or a virtualized machine.

[0060] The following introduces a voiceprint recognition method provided by this application. Figure 2 This is a flowchart of a voiceprint recognition method provided in an embodiment of the present application. The present application provides method operation steps as described in the embodiment or flowchart, but may include more or fewer operation steps based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many orders, and does not represent the only order of execution. When the actual system or server product is executed, it can be executed in sequence or in parallel according to the method shown in the embodiment or the accompanying drawings (for example, in a parallel processor or multi-threaded processing environment). Please refer to Figure 2 A voiceprint recognition method provided in an embodiment of the present application may include the following steps:

[0061] S210: Acquire voice information to be recognized.

[0062] In an embodiment of the present application, the voice information may be audio data, which may be expressed as an electrical signal in the time domain or frequency domain, and voiceprint recognition may be performed using the audio data. According to the specific business scenario and recognition target, voiceprint recognition can be divided into Speaker Identification (SI) and Speaker Verification (SV). Voiceprint identification refers to determining to which person in the target speaker model set the voice information to be recognized belongs, which is a 1:N selection problem, in which the target speaker model is a set of speakers for which voiceprint features have been reserved; and voiceprint verification is to determine whether the voice to be recognized comes from the declared target speaker, which is a 1:1 decision problem.

[0063] S230: Convert the voice information to be recognized to obtain candidate voice information.

[0064] The method provided in the embodiment of the present application is mainly improved in the input preprocessing stage, by converting the original voice information to be recognized to obtain one or more candidate voice information, so that the voice information to be recognized and one or more candidate voice information are input into the voiceprint model for voiceprint verification processing, instead of just inputting the voice information to be recognized, or just inputting the converted candidate voice information.

[0065] In one embodiment of the present application, the speech information to be recognized is subjected to batch random preprocessing and audio conversion without changing the speaker characteristics in the speech information to be recognized as much as possible, wherein the so-called speaker characteristics can be traditional acoustic features including Mel-Frequency Cepstral Coefficients (MFCC), Perceptual linear predictive coefficients (PLP), Deep Features (Deep Feature) or Power-Normalized Cepstral Coefficients (PNCC), etc. The above parameters can be used as acoustic features that are optional and perform well in the feature extraction level of voiceprint recognition. Exemplarily, as shown in formula (1), in the voiceprint verification stage, a batch audio conversion function g is used to process the speech information x to be recognized t Perform batch conversion to generate n candidate voice information, and form a set X with the voice information to be recognized:

[0066]

[0067] in, is the original audio corresponding to the voice information to be recognized, which is equivalent to x t The batch audio conversion function g refers to batch random preprocessing and audio conversion of the original audio without changing the speaker characteristics in the speech information to be recognized as much as possible. Represents the audio data corresponding to n pieces of candidate voice information.

[0068] In a possible implementation, Figure 3 Specifically, the conversion process may include the following:

[0069] S231: Add noise information to the voice information to be recognized to obtain first candidate voice information.

[0070] Exemplarily, in the absence of reverberation, noise within a certain signal-to-noise ratio limit is randomly added to obtain the first candidate voice information. The added noise can be noise that does not change the speaker characteristics of the original audio, such as real environmental noise, Gaussian white noise, etc., that is, adding additive noise.

[0071] Exemplarily, a certain amount of reverberation is introduced into the speech information to be recognized to obtain the first candidate speech information. Alternatively, the first candidate speech information with reverberation can be obtained by adding multiplicative noise, that is, performing a convolution operation on the speech information to be recognized and the noise information in the time domain.

[0072] S232: Or, adjusting the volume or the sound speed of the voice information to be recognized to obtain the second candidate voice information.

[0073] Exemplarily, the audio volume corresponding to the speech information to be recognized can be adjusted by an amplifier element, or the audio speed can be adjusted by changing the audio sampling rate to obtain one or more second candidate speech information.

[0074] S233: Or, filter the voice information to be recognized to obtain third candidate voice information.

[0075] Exemplarily, the audio representing the speech information to be recognized is filtered, and the filter used may be a filter that does not change the speaker's characteristics, such as a Gaussian filter, a median filter, a mean filter, etc., to obtain one or more third candidate speech information.

[0076] S234: Or, reconstruct the audio of the voice information to be recognized to obtain fourth candidate voice information.

[0077] For example, conventional methods for reconstructing audio signals include Filter-Bank-Summation (FBS) and Overlap-and-Add (OLA). In addition, the original speech information to be recognized can be reconstructed by reducing noise using an audio noise reduction model, or the original speech information to be recognized can be reconstructed using a vocoder, an interpolation algorithm, etc.

[0078] In the above embodiments, one or more candidate voice information is obtained in a variety of ways, which is a preprocessing conversion performed on the voice information to be recognized, so that it can be output to the voiceprint model together with the voice information to be recognized.

[0079] S250: Input the voice information to be recognized and the candidate voice information into the voiceprint model respectively, perform voiceprint verification processing, and obtain a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information.

[0080] In general, such as Figure 4 As shown, voiceprint recognition mainly involves two steps: voiceprint reservation and voiceprint verification. Voiceprint reservation is to convert the user voice of the user to be registered into a speaker representation vector and store it. Voiceprint verification is to determine whether a piece of voice to be recognized comes from a specified speaker. It is mainly achieved by converting the voice to be recognized into a speaker representation vector and comparing it with the speaker representation vector corresponding to the reserved user voice. If the score result is greater than a pre-set threshold, it can be determined that they belong to the same speaker; otherwise, if the score is less than the threshold, it can be determined that they do not belong to the same speaker.

[0081] In an embodiment of the present application, voiceprint verification processing is performed on the voice information to be recognized and one or more candidate voice information, respectively, to obtain a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information, respectively. When there are multiple candidate voice information, there are also multiple second voiceprint verification results. The above-mentioned voiceprint verification result can be a verification score, which represents the similarity with the speaker representation vector corresponding to the reserved user voice, as shown in formula (2), where X represents the set of voice information to be recognized and candidate voice information, s0 represents the first voiceprint verification result, s1, s2, s3…s n The second voiceprint verification results corresponding to the n candidate voice information respectively constitute a score group S.

[0082] S=f(X)=[s0,s1,s2,s3…s n ] (2)

[0083] In one embodiment of the present application, each voice information is subjected to feature extraction and then input into the voiceprint model. Specifically, Figure 5 As shown, the following steps may be included:

[0084] S251: Extract features of the voice information to be recognized to obtain first feature information corresponding to the voice information to be recognized.

[0085] S252: Extract features from the candidate voice information to obtain second feature information corresponding to the candidate voice information.

[0086] It is feasible. Speech information can be considered as a short-term stable signal and a long-term non-stationary signal. Its long-term non-stationary characteristics are caused by the changes in the physical movement process of the vocal organs. However, the movement of the vocal organs has a certain inertia, so the speech signal can still be treated as a stable signal in a short time. This short time is generally between 10 and 30 milliseconds. In digital signal processing, it is generally expected to perform time-frequency analysis on the stable signal to extract features. Therefore, when extracting features from the speech signal, a time window of about 20ms can be set. Within this time window, the speech signal is considered to be stable, and then the speech signal is slid on this window as a unit. Each time window can extract a feature that can characterize the signal within this time window. This feature can characterize the relevant information of the speech signal within this time window, thereby obtaining a feature sequence of the speech signal. In this way, a segment of speech can be converted into a feature sequence in frames as the feature information corresponding to the speech information. Common feature parameters include filter banks, Mel frequency cepstral coefficients (MFCCs), and perceptual linear prediction coefficients (PLPs).

[0087] S253: Input the first feature information into the voiceprint model, and compare it with the reserved registered voice feature information to obtain a first voiceprint verification result corresponding to the voice information to be recognized.

[0088] S254: Input the second feature information into the voiceprint model, and compare it with the reserved registered voice feature information to obtain a second voiceprint verification result corresponding to the candidate voice information.

[0089] The registered voice feature information corresponds to the registered voice information of the registered user, and is the user voiceprint feature information stored in the voiceprint reservation stage.

[0090] It is feasible that the voiceprint model can be a model framework based on traditional methods such as Gaussian mixture model (GMM), Gaussian mixture model-Universal Background Model (GMM-UBM), etc., or it can be a neural network model framework based on deep learning, such as the identity vector (i-vector) model. Taking the identity vector model as an example, by calculating the cosine distance between each feature information and the reserved registered voice feature information, each voiceprint verification result can be obtained, and each voiceprint verification result is a verification score.

[0091] In another embodiment of the present application, an end-to-end neural network model based on deep learning may be used, such as a d-vector deep neural network model. Specifically, the feature extraction layer (hidden layer) of the deep neural network is used to output the speaker features at the frame level and express them as speaker feature vectors, and then the cosine distance between the speaker feature vector corresponding to the speech information to be recognized and the reserved speaker feature vector of the registered user is calculated to obtain a score.

[0092] In one embodiment of the present application, in the voiceprint reservation stage, it is necessary to obtain the registered voice information of the registered user, and input the registered voice information into the voiceprint model to obtain the registered voice feature information of the registered user, wherein the registered voice feature information represents the voiceprint feature of the registered user. That is, in the application of voiceprint confirmation or voiceprint recognition, it is necessary to reserve the voice information and voiceprint features of the target speaker in order to compare with the voice information to be recognized and the candidate voice information.

[0093] S270: Determine a voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result.

[0094] In the embodiment of the present application, for the voiceprint recognition task of the voice information to be recognized, the voice information to be recognized and the corresponding one or more candidate voice information are processed together for voiceprint verification, so multiple voiceprint verification results are obtained, including a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information. Therefore, for multiple output results, the method provided in the embodiment of the present application designs a corresponding decision strategy to obtain the final voiceprint recognition result corresponding to the voice information to be recognized, wherein the voiceprint recognition result indicates whether the current speaker (that is, the provider of the voice information to be recognized) is one of the registered users, or indicates whether the current speaker is a designated target speaker.

[0095] In a feasible implementation, the first voiceprint verification result and the second voiceprint verification result are verification scores, so a judgment can be made based on a certain average result to obtain the final voiceprint recognition result. Specifically, Figure 6 As shown, the method may include:

[0096] S2711: Calculate a first target mean value based on the first voiceprint verification result and the second voiceprint verification result, where the first target mean value represents the mean value of the verification scores.

[0097] S2713: Based on the first preset threshold and the first target mean, determine a voiceprint recognition result for the voice information to be recognized, the voiceprint recognition result representing whether the object to be recognized who provides the voice information to be recognized is a registered user.

[0098] Exemplarily, based on formula (2), as shown in formula (3), the first target mean of the score group S can be calculated as

[0099]

[0100] Then the first target mean The first preset threshold τ1 is compared to determine whether the current speaker is the same speaker as a registered user or whether the current speaker is one of the registered users.

[0101] In another feasible implementation, for multiple voiceprint verification results, a final voiceprint recognition result is obtained based on a majority voting strategy. Specifically, Figure 6 As shown, the method may include:

[0102] S2721: Determine a first voiceprint recognition result corresponding to the first voiceprint verification result and a second voiceprint recognition result corresponding to the second voiceprint verification result according to a second preset threshold, the first voiceprint verification result and the second voiceprint verification result.

[0103] The first voiceprint verification result and the second voiceprint verification result are respectively compared with the second preset threshold to obtain the corresponding first voiceprint recognition result and second voiceprint recognition result, each of which indicates whether the current speaker is the same speaker as a registered user or determines whether the current speaker is one of the registered users. The second preset threshold τ2 may be equal to or different from the first preset threshold τ1 in the aforementioned embodiment.

[0104] S2723: Based on the principle of minority obeys majority, determine the final voiceprint recognition result according to the first voiceprint recognition result and the second voiceprint recognition result; the voiceprint recognition result indicates whether the object to be recognized who provides the voice information to be recognized is a registered user.

[0105] Exemplarily, if the number of voiceprint recognition results indicating that the current speaker and a registered user are the same speaker is greater than the number of voiceprint recognition results indicating that the current speaker and a registered user are not the same speaker, the final voiceprint recognition result indicates that the current speaker and a registered user are the same speaker.

[0106] In the above embodiment, the final voiceprint recognition result may be different from the first voiceprint recognition result corresponding to the first voiceprint verification result. If the acquired voice information to be recognized is an adversarial sample with disturbance added, effective adversarial defense can be implemented at this time to effectively prevent misjudgment.

[0107] In a feasible implementation, the difference between the first voiceprint verification result and the second voiceprint verification result can be used to detect whether the voice information to be recognized is an adversarial sample with added disturbance. Figure 7 As shown, the method may include:

[0108] S2731: Calculate a second target mean based on the first voiceprint verification result and the second voiceprint verification result, where the second target mean represents the mean of the verification score differences.

[0109] S2733: Based on the third preset threshold and the second target mean, determine the attribute status information of the speech information to be recognized, the attribute status information representing whether the speech information to be recognized is an adversarial sample.

[0110] Exemplarily, based on formula (2), as shown in formula (4), the second target mean s' of the score group S can be calculated:

[0111]

[0112] Where s0 is the first voiceprint verification result corresponding to the voice information to be recognized, s iThe second voiceprint verification result corresponding to the candidate voice information is obtained, and then the second target mean s' is compared with the third preset threshold τ3. If s' is greater than the preset threshold τ3, the voice information to be recognized is determined to be an adversarial sample, and the final voiceprint recognition result may not be returned; if s' is less than the threshold τ3, the voice information to be recognized is not considered to be an adversarial sample, and the voiceprint recognition result may be output according to s0.

[0113] In another feasible implementation, for multiple voiceprint verification results, the final attribute state information is obtained based on the majority voting strategy to detect whether the voice information to be recognized is an adversarial sample with added disturbance. Specifically, Figure 7 As shown, the method may include:

[0114] S2741: Determine first attribute status information corresponding to the first voiceprint verification result and second attribute status information corresponding to the second voiceprint verification result according to a fourth preset threshold, the first voiceprint verification result and the second voiceprint verification result.

[0115] The first voiceprint verification result and the second voiceprint verification result are respectively compared with the fourth preset threshold to obtain the corresponding first attribute state information and the first attribute state information, each of which indicates whether the corresponding voice information is an adversarial sample. The fourth preset threshold τ4 may be equal to or not equal to the third preset threshold τ3 in the aforementioned embodiment.

[0116] S2743: Based on the principle of minority obeys majority, final attribute state information is determined according to the first attribute state information and the second attribute state information, where the attribute state information represents whether the speech information to be recognized is an adversarial sample.

[0117] Exemplarily, if the amount of attribute state information indicating that the corresponding speech information is an adversarial sample is greater than the amount of attribute state information indicating that the corresponding speech information is not an adversarial sample, the final attribute state information indicates that the speech information to be identified is an adversarial sample.

[0118] Exemplarily, when the attribute status information indicates that the voice information to be recognized is not an adversarial sample, the first voiceprint recognition result corresponding to the first voiceprint verification result is output as the voiceprint recognition result.

[0119] Figure 8 A flow chart of voiceprint recognition is shown. Figure 8As shown, in addition to voiceprint reservation, the voiceprint recognition method provided by the present application mainly performs batch conversion on the voice information to be recognized (i.e., the test voice), and multiple candidate voice information can be obtained, and the voice information to be recognized and the candidate voice information are input into the voiceprint model together for voiceprint verification processing, thereby obtaining a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information, which constitute a score group. Then, a voting decision can be made based on the score group to obtain the final voiceprint recognition result or attribute status information corresponding to the voice information to be recognized. The former can indicate whether the current speaker is the same speaker as the user who registered the voice, and the latter can indicate whether the test voice is an adversarial sample. When applied to adversarial defense and detection, the technical solution provided by the present application makes improvements in the input preprocessing stage. There is no need to conduct adversarial training on the original voiceprint model, nor is there a need to add a defense module for adversarial sample attacks. Instead, it only requires adding conversion processing of the speech to be recognized and judgment processing of multiple voiceprint verification results to the original voiceprint verification stage. The overall solution is simple and easy to implement, and has high versatility. Moreover, the technical solution provided by the present application can implement effective defense and detection against white-box attacks or black-box attacks based on adversarial samples without affecting the performance of the original voiceprint model, thereby improving the security and reliability of voiceprint recognition.

[0120] The present application also provides a voiceprint recognition device 900. Fig. 9 As shown, the device 900 may include:

[0121] An acquisition module 910 is used to acquire voice information to be recognized;

[0122] The conversion module 920 is used to convert the voice information to be recognized to obtain candidate voice information;

[0123] The voiceprint verification module 930 is used to input the voice information to be recognized and the candidate voice information into the voiceprint model respectively, perform voiceprint verification processing, and obtain a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information;

[0124] The recognition module 940 is used to determine the voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result.

[0125] In one embodiment of the present application, the voiceprint verification module 930 may include:

[0126] A feature extraction unit, used to extract features from the voice information to be recognized and the candidate voice information respectively, to obtain first feature information corresponding to the voice information to be recognized and second feature information corresponding to the candidate voice information;

[0127] A feature comparison unit is used to input the first feature information and the second feature information into the voiceprint model respectively, and compare them with the reserved registered voice feature information to obtain a first voiceprint verification result corresponding to the voice information to be identified and a second voiceprint verification result corresponding to the candidate voice information; the registered voice feature information corresponds to the registered voice information of a registered user.

[0128] In one embodiment of the present application, the identification module 940 may include:

[0129] a first mean value calculation unit, configured to calculate a first target mean value according to the first voiceprint verification result and the second voiceprint verification result, wherein the first target mean value represents a mean value of verification scores;

[0130] The first determination unit is used to determine the voiceprint recognition result for the voice information to be recognized based on a first preset threshold and the first target mean, wherein the voiceprint recognition result represents whether the object to be recognized who provides the voice information to be recognized is a registered user.

[0131] In one embodiment of the present application, the identification module 940 may further include:

[0132] A second determining unit, configured to determine a first voiceprint recognition result corresponding to the first voiceprint verification result and a second voiceprint recognition result corresponding to the second voiceprint verification result according to a second preset threshold, the first voiceprint verification result and the second voiceprint verification result;

[0133] The first judgment unit is used to determine the final voiceprint recognition result based on the principle of minority obeys majority and according to the first voiceprint recognition result and the second voiceprint recognition result; the voiceprint recognition result represents whether the object to be identified who provides the voice information to be identified is a registered user.

[0134] In one embodiment of the present application, the identification module 940 may further include:

[0135] A second mean value calculation unit, configured to calculate a second target mean value according to the first voiceprint verification result and the second voiceprint verification result, wherein the second target mean value represents a mean value of the verification score differences;

[0136] The third determination unit is used to determine the attribute status information of the speech information to be recognized based on a third preset threshold and the second target mean, wherein the attribute status information represents whether the speech information to be recognized is an adversarial sample.

[0137] In one embodiment of the present application, the identification module 940 may further include:

[0138] a fourth determining unit, configured to determine first attribute state information corresponding to the first voiceprint verification result and second attribute state information corresponding to the second voiceprint verification result according to a fourth preset threshold, the first voiceprint verification result and the second voiceprint verification result;

[0139] The second decision unit is used to determine final attribute state information based on the principle of minority obeys majority and according to the first attribute state information and the second attribute state information, wherein the attribute state information represents whether the speech information to be identified is an adversarial sample.

[0140] In one embodiment of the present application, the third determining unit or the fourth determining unit may include:

[0141] The result determination subunit is used to use the first voiceprint recognition result corresponding to the first voiceprint verification result as the voiceprint recognition result when the attribute status information indicates that the voice information to be recognized is not an adversarial sample.

[0142] In one embodiment of the present application, the device 900 may further include:

[0143] A registered voice acquisition unit, used to acquire the registered voice information of the registered user;

[0144] The voiceprint feature reserving unit is used to input the registered voice information into the voiceprint model to obtain the registered voice feature information of the registered user, wherein the registered voice feature information represents the voiceprint feature of the registered user.

[0145] In one embodiment of the present application, the conversion module 920 may include:

[0146] A first conversion unit, configured to add noise information to the voice information to be recognized to obtain first candidate voice information;

[0147] A second conversion unit, used for adjusting the volume or speed of the voice information to be recognized to obtain second candidate voice information;

[0148] a third conversion unit, configured to filter the voice information to be recognized to obtain third candidate voice information;

[0149] The fourth conversion unit is used to perform audio reconstruction on the voice information to be recognized to obtain fourth candidate voice information.

[0150] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0151] An embodiment of the present application provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement a voiceprint recognition method provided in the above method embodiment.

[0152] Fig.10 A schematic diagram of the hardware structure of a device for implementing a voiceprint recognition method provided in an embodiment of the present application is shown. The device may participate in or include the apparatus or system provided in an embodiment of the present application. Fig.10 As shown, the device 10 may include one or more (1002a, 1002b, ..., 1002n are used to illustrate) processors 1002 (the processor 1002 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 1004 for storing data, and a transmission device 1006 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Fig.10 The structure shown is for illustration only and does not limit the structure of the electronic device. Fig.10 More or fewer components as shown, or with Fig.10 Different configurations are shown.

[0153] It should be noted that the one or more processors 1002 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the device 10 (or mobile device). As involved in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0154] The memory 1004 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the methods described in the embodiments of the present application. The processor 1002 executes various functional applications and data processing by running the software programs and modules stored in the memory 1004, that is, to implement the above-mentioned voiceprint recognition method. The memory 1004 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 1004 may further include a memory remotely arranged relative to the processor 1002, and these remote memories may be connected to the device 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0155] The transmission device 1006 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the device 10. In one example, the transmission device 1006 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 1006 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0156] The display may be, for example, a touch screen liquid crystal display (LCD) that may enable a user to interact with a user interface of device 10 (or mobile device).

[0157] An embodiment of the present application also provides a computer-readable storage medium, which can be set in a server to store at least one instruction or at least one program related to a voiceprint recognition method in a method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement a voiceprint recognition method provided by the above method embodiment.

[0158] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers of the computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk, or an optical disk.

[0159] The embodiment of the present invention further provides a computer program product or a computer program, wherein the computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes a voiceprint recognition method provided in the above various optional implementations.

[0160] It can be seen from the embodiments of the voiceprint recognition method, device, medium and equipment provided by the present application that:

[0161] The solution provided by the present application converts the voice information to be recognized to obtain candidate voice information, and inputs the voice information to be recognized and the candidate voice information into the voiceprint model together for voiceprint verification processing, thereby obtaining the final voiceprint recognition result corresponding to the voice information to be recognized based on the first voiceprint verification result corresponding to the voice information to be recognized and the second voiceprint verification result corresponding to the candidate voice information. When applied to adversarial defense and detection, the technical solution provided by the present application is improved in the input preprocessing stage, and there is no need to conduct adversarial training on the original voiceprint model, nor is there any need to add a defense module for adversarial sample attacks. It only needs to add conversion processing of the voice to be recognized and judgment processing of multiple voiceprint verification results in the original voiceprint verification stage. The overall solution is simple and easy to implement, and has high versatility; and the technical solution provided by the present application can implement effective defense and detection against white-box attacks or black-box attacks based on adversarial samples without affecting the performance of the original voiceprint model, thereby improving the security and reliability of voiceprint recognition.

[0162] It should be noted that the above-mentioned sequence of the embodiments of the present application is for description only and does not represent the advantages and disadvantages of the embodiments. The above-mentioned specific embodiments of the present application are described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0163] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment and storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0164] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0165] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A voiceprint recognition method, characterized in that: The method comprises: Acquire voice information to be recognized; Converting the voice information to be recognized to obtain candidate voice information; Inputting the voice information to be recognized and the candidate voice information into a voiceprint model respectively, performing voiceprint verification processing, and obtaining a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information; Determining a voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result; The converting of the voice information to be recognized to obtain candidate voice information includes: Adding noise information to the voice information to be recognized to obtain first candidate voice information; or, adjusting the volume or speed of the voice information to be recognized to obtain second candidate voice information; or, filtering the voice information to be recognized to obtain third candidate voice information; Alternatively, audio reconstruction is performed on the voice information to be recognized to obtain fourth candidate voice information.

2. The method according to claim 1, characterized in that The step of inputting the voice information to be recognized and the candidate voice information into a voiceprint model respectively, performing voiceprint verification processing, and obtaining a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information includes: Extracting features of the voice information to be recognized and the candidate voice information respectively to obtain first feature information corresponding to the voice information to be recognized and second feature information corresponding to the candidate voice information; The first feature information and the second feature information are respectively input into the voiceprint model, and compared with the reserved registered voice feature information to obtain a first voiceprint verification result corresponding to the voice information to be identified and a second voiceprint verification result corresponding to the candidate voice information; the registered voice feature information corresponds to the registered voice information of a registered user.

3. The method according to claim 1, characterized in that The first voiceprint verification result and the second voiceprint verification result are verification scores, and determining the voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result includes: Calculating a first target mean value according to the first voiceprint verification result and the second voiceprint verification result, where the first target mean value represents a mean value of the verification scores; Based on a first preset threshold and the first target mean, the voiceprint recognition result for the voice information to be recognized is determined, and the voiceprint recognition result represents whether the object to be recognized who provides the voice information to be recognized is a registered user.

4. The method according to claim 1, characterized in that: The step of determining the voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result further includes: Determine, according to a second preset threshold, the first voiceprint verification result, and the second voiceprint verification result, a first voiceprint recognition result corresponding to the first voiceprint verification result, and a second voiceprint recognition result corresponding to the second voiceprint verification result; Based on the principle of minority obeys majority, the final voiceprint recognition result is determined according to the first voiceprint recognition result and the second voiceprint recognition result; the voiceprint recognition result indicates whether the object to be identified who provides the voice information to be identified is a registered user.

5. The method according to claim 1, characterized in that The method further comprises: Calculate a second target mean value according to the first voiceprint verification result and the second voiceprint verification result, where the second target mean value represents the mean of the verification score differences; Based on a third preset threshold and the second target mean, attribute state information of the speech information to be recognized is determined, wherein the attribute state information represents whether the speech information to be recognized is an adversarial sample.

6. The method according to claim 1, characterized in that The method further comprises: Determine, according to a fourth preset threshold, the first voiceprint verification result, and the second voiceprint verification result, first attribute state information corresponding to the first voiceprint verification result, and second attribute state information corresponding to the second voiceprint verification result; Based on the principle of minority obeys majority, final attribute state information is determined according to the first attribute state information and the second attribute state information, and the attribute state information represents whether the speech information to be recognized is an adversarial sample.

7. The method according to claim 5 or 6, characterized in that: The method further comprises: When the attribute status information indicates that the voice information to be recognized is not an adversarial sample, the first voiceprint recognition result corresponding to the first voiceprint verification result is used as the voiceprint recognition result.

8. The method according to claim 2, characterized in that: Before acquiring the voice information to be recognized, the method further includes: Acquire the registered voice information of the registered user; The registered voice information is input into the voiceprint model to obtain the registered voice feature information of the registered user, where the registered voice feature information represents the voiceprint feature of the registered user.

9. A voiceprint recognition device, characterized in that: The device comprises: An acquisition module, used for acquiring voice information to be recognized; A conversion module, used for converting the voice information to be recognized to obtain candidate voice information; A voiceprint verification module, used to input the voice information to be recognized and the candidate voice information into a voiceprint model respectively, perform voiceprint verification processing, and obtain a first voiceprint verification result corresponding to the voice information to be recognized and a second voiceprint verification result corresponding to the candidate voice information; an identification module, configured to determine a voiceprint recognition result for the voice information to be recognized according to the first voiceprint verification result and the second voiceprint verification result; The conversion module comprises: A first conversion unit, configured to add noise information to the voice information to be recognized to obtain first candidate voice information; A second conversion unit, used for adjusting the volume or speed of the voice information to be recognized to obtain second candidate voice information; a third conversion unit, configured to filter the voice information to be recognized to obtain third candidate voice information; The fourth conversion unit is used to perform audio reconstruction on the voice information to be recognized to obtain fourth candidate voice information.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement a voiceprint recognition method as described in any one of claims 1 to 8.

11. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement a voiceprint recognition method according to any one of claims 1 to 8.

12. A computer program product comprising computer instructions, characterized in that: When the computer instruction is executed by a processor, a voiceprint recognition method according to any one of claims 1 to 8 is implemented.

Images