Identity recognition method, system, storage medium and electronic device based on sound characteristics
Through the training and authenticity testing of the voiceprint model, the problem of reduced accuracy of sound feature recognition in complex environments and insufficient false sound recognition capabilities are solved, achieving higher recognition accuracy and anti-interference ability.
Patent Information
- Application Number
- CN202410263044.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-07
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-03-07
AI Technical Summary
The existing sound feature recognition technology has the problems of poor sound quality and forged sound leading to identity recognition errors, especially in complex environments, which reduce the accuracy of recognition and lack the anti-interference ability of forged sounds.
By acquiring and preprocessing sound data, establishing a voiceprint model, using a convolutional network and deep learning model for training, combining noise suppression and normalization processing, vocalprint similarity is calculated, and authenticity tests are performed to confirm identity information.
It improves the accuracy and anti-interference ability of sound recognition, enhances the recognition ability of forged sounds, and is suitable for real-time authentication scenarios.
Smart Images

Figure CN118212927B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of computer technology, and in particular to a method, system and storage medium for identifying an individual based on sound features. Background Art
[0002] Voice feature-based identity recognition technology is a method of using individual voice features to authenticate and identify an individual. This technology belongs to the field of speech processing and pattern recognition and has a wide range of applications, including but not limited to security, finance, or medical fields.
[0003] Existing recognition technologies still have problems such as poor sound feature quality or forged sound, which can lead to identity errors. Summary of the invention
[0004] In view of this, the purpose of the present disclosure is to propose an identity recognition method, system, storage medium and electronic device based on sound features, which solves the problem of identity recognition errors caused by misrecognition of sound features.
[0005] In order to achieve one of the above disclosed purposes, the present disclosure provides an identity recognition method based on sound features, the method comprising:
[0006] Acquire audio data of multiple speakers;
[0007] Preprocessing the sound data to obtain a sound feature set;
[0008] Based on the sound feature set, training a voiceprint model;
[0009] Acquire recorded voice matching the personal information, and convert the recorded voice into a recorded voiceprint using the voiceprint model;
[0010] Acquire input speech, and convert the input speech into an input voiceprint using the voiceprint model;
[0011] Calculating the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information;
[0012] An authenticity test is performed on the input voice, and in response to the authenticity being within a threshold range, the matched personal information is confirmed to be the recognized identity information.
[0013] As a further improvement of an embodiment of the present disclosure, the authenticity test of the input speech includes:
[0014] Performing time domain analysis on the input speech to generate a speech rate change rule and a tone change rule of the input speech;
[0015] Performing frequency domain analysis on the input speech to generate a frequency domain variation rule of the input speech;
[0016] Using natural language processing technology to identify the input semantics of the input speech;
[0017] A deep learning model is used to determine the speech rate change pattern, intonation change pattern and the authenticity of the input semantics.
[0018] As a further improvement of an embodiment of the present disclosure, the training of the voiceprint model based on the sound feature set includes:
[0019] Acquire the sound features and corresponding voiceprint features in the sound feature set;
[0020] Dividing the sound features and the corresponding voiceprint features into a training set, a validation set and a test set;
[0021] Constructing a convolutional network, wherein the convolutional network includes a layer for extracting voiceprint features;
[0022] Defining a loss function and training parameters, and training the convolutional network using the training set and the validation set;
[0023] According to the performance results of the validation set, the timing to stop training is determined and the training is stopped.
[0024] As a further improvement of an embodiment of the present disclosure, the loss function is:
[0025] L(A,P,N)=max(d(A,P)-d(A,N)+α,0);
[0026] Where A is the anchor point, P is the positive sample, N is the negative sample, d is the distance function between samples, and α is the edge margin;
[0027] The training parameters include at least one of a learning rate, a batch size, a number of iterations, and an optimizer.
[0028] As a further improvement of an embodiment of the present disclosure, the training of the voiceprint model based on the sound feature set includes:
[0029] The voiceprint model is trained using any algorithm of a support vector machine, a Gaussian mixture model or a deep learning model.
[0030] As a further improvement of an embodiment of the present disclosure, the calculating the similarity between the recorded voiceprint and the input voiceprint includes:
[0031] Selecting the input voiceprint with a clustering structure, calculating the Euclidean distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity;
[0032] Selecting the input voiceprint with high-dimensional features, calculating the Mahalanobis distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity;
[0033] The input voiceprint having a directional feature is selected, and the cosine similarity between the recorded voiceprint and the input voiceprint is calculated.
[0034] As a further improvement of an embodiment of the present disclosure, the preprocessing of the sound data to obtain a sound feature set includes:
[0035] Using a noise suppression algorithm to perform noise reduction processing on the sound data;
[0036] The sound data is normalized:
[0037] The processed sound data is collected to obtain the sound feature set.
[0038] Based on the same inventive concept, the present disclosure also provides an identity recognition system based on sound features, comprising:
[0039] A first acquisition module, used to acquire voice data of multiple speakers;
[0040] A preprocessing module, used for preprocessing the sound data to obtain a sound feature set;
[0041] A first training module, used for training a voiceprint model based on the voice feature set;
[0042] A second acquisition module is used to acquire recorded voice matching the personal information, and convert the recorded voice into a recorded voiceprint using the voiceprint model;
[0043] A third acquisition module is used to acquire input speech and convert the input speech into an input voiceprint using the voiceprint model;
[0044] a calculation module, configured to calculate the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information;
[0045] The testing module is used to perform an authenticity test on the input voice, and in response to the authenticity being within a threshold range, confirm that the matched personal information is the recognized identity information.
[0046] Based on the same inventive concept, the present disclosure also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute any of the above-mentioned identity recognition methods based on sound features.
[0047] Based on the same inventive concept, the present disclosure also provides an electronic device, including: a processor and a memory; the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any of the above-mentioned identity recognition methods based on sound features.
[0048] Compared with the prior art, the technical effect of the present invention is that the present invention utilizes sound features for identity recognition, and pre-processes the sound data to obtain consistent sound features, trains the voiceprint model, improves the accuracy of recognizing input voice, and has stronger resistance to interference from environmental noise. The present invention also confirms the authenticity of the input voice through authenticity testing, through multiple feature recognition and strategies, and improves the recognition ability of forged voices and deceptive attacks. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions in the present disclosure or related technologies, the drawings required for use in the implementation methods or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only implementation methods of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0050] Figure 1 A flow chart of an identity recognition method based on sound features provided in one embodiment of the present disclosure;
[0051] Figure 2 A flowchart of preprocessing sound data provided by another embodiment of the present disclosure;
[0052] Figure 3 A flowchart of training a voiceprint model provided in another embodiment of the present disclosure;
[0053] Figure 4 A flowchart of calculating similarity provided for another embodiment of the present disclosure;
[0054] Figure 5 A flowchart of an authenticity test provided for another embodiment of the present disclosure;
[0055] Figure 6 A schematic diagram of an identity recognition system based on sound features provided by another embodiment of the present disclosure;
[0056] Figure 7 A schematic diagram of the hardware structure of an electronic device provided in another embodiment of the present disclosure. DETAILED DESCRIPTION
[0057] The present invention will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by a person skilled in the art based on these embodiments are all within the scope of protection of the present invention.
[0058] Embodiments It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprising" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects.
[0059] Voice feature-based identity recognition technology is a method of using individual voice features for identity verification and identification. This technology belongs to the field of speech processing and pattern recognition, and has a wide range of applications, including but not limited to the following: Security field: Voiceprint recognition can be applied to voice recognition access control systems, telephone customer service identity verification, etc., providing a safer and more convenient way of identity verification. Financial field: Voice feature-based identity recognition can be used in scenarios such as telephone banking and transaction authorization to improve transaction security. Medical field: Voiceprint recognition technology can be used for patient identity identification to ensure correct medical records and personal information.
[0060] Although there are many potential applications for voice feature-based identity recognition technology, there are still some problems and shortcomings in the existing background technology, including the following aspects: Environmental influence: Sound is disturbed by environmental factors, such as noise, echo, and room anisotropy, which may reduce the accuracy of voiceprint recognition, especially in complex environments. Individual differences: Everyone's voice features are unique, but individual differences may lead to reduced recognition accuracy. For example, factors such as age, mood, health status, and pronunciation habits may cause changes in voice features. Data scarcity: For voice feature-based identity recognition technology, large-scale, high-quality voiceprint datasets are crucial to the training and performance of the algorithm. However, obtaining large-scale individual voice feature datasets may face privacy and legal restrictions. Real-time requirements: Some application scenarios require fast identity authentication and recognition, such as telephone banking services. However, existing voiceprint recognition technology may have certain challenges in real-time, requiring higher algorithm efficiency and processing speed. Spoofing attacks: Voiceprint recognition technology may be subject to spoofing attacks, such as using forged voice samples or simulating the voice of a specific individual. The recognition system needs to have the ability to resist spoofing attacks to improve the security of the system.
[0061] The present disclosure provides a method for identifying an individual based on sound features. Figure 1 As shown, the identity recognition method based on sound features includes the following steps:
[0062] Step S100: Acquire voice data of multiple speakers.
[0063] Specifically, use high-quality microphones or sound collection equipment to ensure quality and frequency response. In addition, the speaker should try to collect sound in a relatively quiet environment to reduce the impact of background noise.
[0064] Step S200, pre-processing the sound data to obtain a sound feature set.
[0065] Specifically, sound data with complex sources or changing environments are preprocessed to obtain sound features with a unified format, and sound features of multiple speakers and different sources are combined into a sound feature set to facilitate the use and processing of the sound features in the next step.
[0066] In one possible implementation of the present disclosure, Figure 2 As shown, in step S200, it also includes:
[0067] Step S210, using a noise suppression algorithm to perform noise reduction processing on the sound data;
[0068] Step S220, normalizing the sound data;
[0069] Step S230, collect the processed sound data to obtain the sound feature set.
[0070] Specifically, noise suppression algorithms, such as spectral subtraction or wavelet denoising, are used on sound data to remove noise from the collected sound data. Spectral subtraction achieves noise reduction by subtracting the spectrum of noise from the spectrum of the original signal. Wavelet denoising uses wavelet transform to decompose and reconstruct the signal and removes noise from the wavelet coefficients. The sound data is then normalized to unify the formats of sound data from different sources, which is convenient for subsequent processing and ensures stability and consistency in the feature extraction process of the sound data.
[0071] Step S300: training a voiceprint model based on the sound feature set.
[0072] Specifically, the voiceprint model is trained using sound data in a sound feature set so that the voiceprint model has the ability to convert sound data into voiceprints, wherein the sound data is the speaker's voice, and the voiceprint refers to the speaker's voice features, which can reflect the speaker's physiological characteristics. Theoretically, voiceprints are like fingerprints, and it is rare for two people to have the same voiceprint features. Therefore, the characteristics of voiceprints can be used for identity recognition. The features extracted from voiceprints are usually vectorized into one or more mathematical vectors, such as 512-dimensional or 1024-dimensional vectors. Voiceprint features are usually represented in the form of multi-dimensional vectors.
[0073] In one possible implementation of the present disclosure, Figure 3 As shown, in step S300, it also includes:
[0074] Step S310, obtaining the sound features and corresponding voiceprint features in the sound feature set.
[0075] Step S320, dividing the sound features and the corresponding voiceprint features into a training set, a verification set and a test set.
[0076] Specifically, the voiceprint model is trained using the sound features and voiceprint features with existing corresponding relationships as data sources, and the data source is divided into a training set, a validation set, and a test set.
[0077] Step S330, constructing a convolutional network, wherein the convolutional network includes a layer for extracting voiceprint features;
[0078] Step S340, defining a loss function and training parameters, and training the convolutional network using the training set and the validation set.
[0079] Specifically, the basic framework for establishing the voiceprint model is a deep learning model, which is specifically set to a Siamese network structure. The Siamese network structure includes layers for extracting voiceprint features, such as convolutional layers or pooling layers.
[0080] In one possible implementation of the present disclosure, the loss function is:
[0081] L(A,P,N)=max(d(A,P)-d(A,N)+α,0);
[0082] Where A is the anchor point, P is the positive sample, N is the negative sample, d is the distance function between samples, and α is the edge margin.
[0083] The training parameters include at least one of a learning rate, a batch size, a number of iterations, and an optimizer.
[0084] Specifically, the learning rate is used to control the step size of parameter updates; an appropriate batch size is selected to balance computational efficiency and model stability; the number of iterations in the process is defined, and training is stopped when the number of iterations is reached to prevent overfitting; Adam or other optimizers are chosen to minimize the triplet loss.
[0085] Step S350: confirming the timing to stop training and stopping training according to the performance result of the verification set.
[0086] Specifically, an early stopping strategy is adopted in this step. The performance of the validation set reflects the functional performance of the voiceprint model. When the accuracy of the voiceprint model in converting sound data into voiceprints meets the requirements, it is confirmed that it is time to stop training. Stopping training when the functional requirements are met can prevent overfitting.
[0087] In one possible implementation of the present disclosure, step S300 further includes: training the voiceprint model using any algorithm of a support vector machine, a Gaussian mixture model or a deep learning model.
[0088] Specifically, the support vector machine (SVM) is a binary classification model that divides samples into two categories by finding a hyperplane that maximizes the distance from the hyperplane to the two categories of samples, thereby achieving optimal classification. During the training process, the SVM calculates the distance from each sample to the hyperplane, and calls samples with a distance greater than 1 support vectors. The SVM has good generalization performance and robustness. The Gaussian mixture model (GMM) is a probabilistic model used to describe the distribution of a set of data. During the training process, the GMM determines the appropriate number and parameters of Gaussian distributions based on the characteristics of the data, and improves the model's fitting effect through iterative optimization methods. The deep learning model is a machine learning model based on a neural network that can automatically learn features and patterns in data.
[0089] Step S400, obtaining recorded speech matching the personal information, and converting the recorded speech into a recorded voiceprint using the voiceprint model.
[0090] Specifically, firstly, the recorded voice matching the personal information is obtained and converted into a recorded voiceprint, so that the database contains the corresponding relationship between the characteristics of the recorded voiceprint and the identity information.
[0091] Step S500: obtaining input speech, and converting the input speech into an input voiceprint using the voiceprint model.
[0092] Specifically, the input voice is the voice to be recognized, and corresponding personal information needs to be selected for the input voice to achieve identity recognition.
[0093] Step S600, calculating the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information.
[0094] Specifically, when the similarity between the recorded voiceprint and the input voiceprint meets the threshold requirement, it is confirmed that the personal information corresponding to the recorded voiceprint is the personal information that needs to be matched with the input voiceprint, that is, the personal information that matches the input voice.
[0095] In one implementation of the present disclosure, in step S600, Figure 4 As shown, it also includes:
[0096] Step S610, selecting the input voiceprint with a clustering structure, calculating the Euclidean distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity.
[0097] Specifically, using Euclidean distance to calculate similarity is suitable for data with obvious clustering structure and has better effects on voiceprint features with relatively even distribution.
[0098] Step S620, selecting the input voiceprint with high-dimensional features, calculating the Mahalanobis distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity.
[0099] Specifically, the Mahalanobis distance is used to calculate the similarity, which is applicable to high-dimensional data and takes into account the correlation between feature dimensions. The Mahalanobis distance is more suitable for handling situations where there is correlation in voiceprint features.
[0100] Step S630, selecting the input voiceprint with directional characteristics, and calculating the cosine similarity between the recorded voiceprint and the input voiceprint.
[0101] Specifically, cosine similarity is used to represent the similarity between the recorded voiceprint and the input voiceprint. It is suitable for situations that focus on the directional information of voiceprint features. It measures the angle between vectors, is more sensitive to the direction of feature vectors, is not affected by scale, and can better capture the relative relationship between features.
[0102] Step S700, performing an authenticity test on the input voice, and in response to the authenticity being within a threshold range, confirming that the matched personal information is the recognized identity information.
[0103] Specifically, in the existing technology, the input voice can be forged, so authenticity testing is needed to avoid the use of technically forged voice data for identity recognition, and to use authenticity testing to identify forged voice data to improve the security of identity recognition. After confirming that the voice data is authentic, the recognized identity information is confirmed.
[0104] In one possible implementation of the present disclosure, Figure 5As shown, step S700 also includes:
[0105] Step S710, performing time domain analysis on the input speech to generate a speech rate change rule and a tone change rule of the input speech;
[0106] Step S720, performing frequency domain analysis on the input speech to generate a frequency domain variation rule of the input speech;
[0107] Step S730, using natural language processing technology to identify the input semantics of the input speech;
[0108] Step S740: Use a deep learning model to determine the speech rate change rules, intonation change rules and the authenticity of the input semantics.
[0109] Specifically, time domain analysis is used to consider the instantaneous changes in speech rate and intonation in sound data; frequency analysis is used to analyze the frequency components of sound and understand the changes in sound in the frequency domain. The dynamic characteristics of sound data are obtained by combining time domain analysis and frequency domain analysis. It is difficult for forged sound data to forge the dynamic characteristics of sound, and forged sound data can be distinguished more clearly. In addition, natural speech processing technology is used to identify the semantics of speech data, and the authenticity of sound data is judged through semantic understanding. Deep learning modules, such as convolutional neural networks or recurrent neural networks, are then used to determine the law of speech rate changes, the law of intonation changes, and the authenticity of the input semantics.
[0110] This solution provides an identity recognition method based on sound features. This method has high accuracy. By using sound features for identity recognition, the voiceprint model can more accurately capture the individual's voice features, thereby improving recognition accuracy. Environmental impact is also considered in the sound feature collection and preprocessing stages, and technologies such as noise removal and normalization are used to make the recognition system more resistant to environmental noise. By introducing authenticity testing, introducing multiple features and strategies for deception detection, the system's ability to recognize fake sounds or deception attacks is improved. The efficiency of sound data collection and processing is high, and rapid identity recognition can be achieved, which is suitable for various real-time application scenarios. Compared with other biometric recognition technologies, identity recognition based on sound features does not require contact with the human body, is non-invasive and comfortable, and can be widely used in fields such as personal authentication and security access control. The voiceprint model is established using methods such as deep learning. The model can be adapted to different individuals and application scenarios through further training and adjustment, and has good scalability.
[0111] like Figure 6 As shown, another embodiment of the present disclosure discloses an identity recognition system based on sound features, comprising:
[0112] A first acquisition module, used to acquire voice data of multiple speakers;
[0113] A preprocessing module, used for preprocessing the sound data to obtain a sound feature set;
[0114] A first training module, used for training a voiceprint model based on the voice feature set;
[0115] A second acquisition module is used to acquire recorded voice matching the personal information, and convert the recorded voice into a recorded voiceprint using the voiceprint model;
[0116] A third acquisition module is used to acquire input speech and convert the input speech into an input voiceprint using the voiceprint model;
[0117] a calculation module, configured to calculate the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information;
[0118] The testing module is used to perform an authenticity test on the input voice, and in response to the authenticity being within a threshold range, confirm that the matched personal information is the recognized identity information.
[0119] In one possible implementation of the present disclosure, the test module further includes:
[0120] A first analysis module is used to perform time domain analysis on the input speech to generate a speech rate change rule and a tone change rule of the input speech;
[0121] A second analysis module is used to perform frequency domain analysis on the input speech to generate a frequency domain variation rule of the input speech;
[0122] A recognition module, used to recognize the input semantics of the input speech using natural language processing technology;
[0123] The judgment module is used to use a deep learning model to judge the law of speech rate change, the law of intonation change and the authenticity of the input semantics.
[0124] In one possible implementation of the present disclosure, the first training module further includes:
[0125] A fourth acquisition module, used to acquire the sound features and corresponding voiceprint features in the sound feature set;
[0126] A division module, used for dividing the sound features and the corresponding voiceprint features into a training set, a verification set and a test set;
[0127] A construction module, used to construct a convolutional network, wherein the convolutional network includes a layer for extracting voiceprint features;
[0128] A second training module is used to define a loss function and training parameters, and train the convolutional network using the training set and the validation set;
[0129] The confirmation module is used to confirm the timing of stopping training and stop training according to the performance results of the verification set.
[0130] In one possible implementation of the present disclosure, the first training module further includes:
[0131] The loss function is:
[0132] L(A,P,N)=max(d(A,P)-d(A,N)+α,0);
[0133] Where A is the anchor point, P is the positive sample, N is the negative sample, d is the distance function between samples, and α is the edge margin;
[0134] The training parameters include at least one of a learning rate, a batch size, a number of iterations, and an optimizer.
[0135] In one possible implementation of the present disclosure, the first training module further includes:
[0136] The voiceprint model is trained using any algorithm of a support vector machine, a Gaussian mixture model or a deep learning model.
[0137] In one possible implementation of the present disclosure, the calculation module further includes:
[0138] A first selection module is used to select the input voiceprint with a clustering structure, calculate the Euclidean distance between the recorded voiceprint and the input voiceprint, and obtain the similarity;
[0139] A second selection module is used to select the input voiceprint with high-dimensional features, calculate the Mahalanobis distance between the recorded voiceprint and the input voiceprint, and obtain the similarity;
[0140] The third selection module is used to select the input voiceprint with directional characteristics and calculate the cosine similarity between the recorded voiceprint and the input voiceprint.
[0141] In one possible implementation of the present disclosure, the preprocessing module further includes:
[0142] A first processing module, configured to perform noise reduction processing on the sound data using a noise suppression algorithm;
[0143] A second processing module, used for normalizing the sound data;
[0144] The collection module is used to collect the processed sound data to obtain the sound feature set.
[0145] Based on the same inventive concept, corresponding to any of the above-mentioned implementation methods, the present disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for generating an order request described in any of the above implementation methods is implemented.
[0146] Figure 7 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.
[0147] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the implementation methods of this specification.
[0148] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1020 and are called and executed by the processor 1010.
[0149] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0150] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).
[0151] The bus 1050 includes pathways that transmit information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).
[0152] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the implementation scheme of this specification, and does not necessarily include all the components shown in the figure.
[0153] The electronic device of the above-mentioned embodiment is used to implement the corresponding identity recognition method based on sound features in any of the above-mentioned embodiments, and has the beneficial effects of the corresponding method implementation method, which will not be repeated here.
[0154] Based on the same inventive concept, corresponding to any of the above-mentioned implementation methods, the present disclosure also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the sound feature-based identity recognition method as described in any of the above implementation methods.
[0155] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
[0156] The computer instructions stored in the storage medium of the above-mentioned implementation are used to enable the computer to execute the identity recognition method based on sound features as described in any of the above implementations, and have the beneficial effects of the corresponding method implementation methods, which will not be repeated here.
[0157] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present disclosure (including the claims) is limited to these examples. This narrative style of the specification is only for the sake of clarity, and those skilled in the art should take the specification as a whole. Under the concept of the present disclosure, the above embodiments or the technical features in different embodiments may be appropriately combined, the steps may be implemented in any order, and there are many other variations of different aspects of the embodiments of the present disclosure as described above, which are not provided in detail for the sake of simplicity.
[0158] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present disclosure difficult to understand, the known power / ground connections to the integrated circuit (IC) chips and other components may or may not be shown in the provided figures. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present disclosure difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present disclosure will be implemented (that is, these details should be fully within the scope of understanding of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present disclosure, it is obvious to those skilled in the art that the embodiments of the present disclosure can be implemented without these specific details or with changes in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.
[0159] Although the present disclosure has been described in conjunction with specific embodiments of the present disclosure, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
[0160] The embodiments of the present disclosure are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made without departing from the spirit and principles of the embodiments of the present disclosure shall be included in the scope of protection of the present disclosure.
Claims
1. A method for identity recognition based on sound features, characterized in that: The method comprises: Acquire audio data of multiple speakers; Preprocessing the sound data to obtain a sound feature set; Based on the sound feature set, training a voiceprint model; Acquire recorded voice matching the personal information, and convert the recorded voice into a recorded voiceprint using the voiceprint model; Acquire input speech, and convert the input speech into an input voiceprint using the voiceprint model; Calculating the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information; Performing an authenticity test on the input voice, and in response to the authenticity being within a threshold range, confirming that the matched personal information is the identified identity information; The calculating the similarity between the recorded voiceprint and the input voiceprint includes: Selecting the input voiceprint with a clustering structure, calculating the Euclidean distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity; Selecting the input voiceprint with high-dimensional features, calculating the Mahalanobis distance between the recorded voiceprint and the input voiceprint, and obtaining the similarity; The input voiceprint having a directional feature is selected, and the cosine similarity between the recorded voiceprint and the input voiceprint is calculated.
2. The method for identifying an individual based on sound features according to claim 1, characterized in that: The authenticity test of the input speech comprises: Performing time domain analysis on the input speech to generate a speech rate change rule and a tone change rule of the input speech; Performing frequency domain analysis on the input speech to generate a frequency domain variation rule of the input speech; Using natural language processing technology to identify the input semantics of the input speech; A deep learning model is used to determine the speech rate change pattern, intonation change pattern and the authenticity of the input semantics.
3. The method for identifying an individual based on sound features according to claim 1, characterized in that: The training of the voiceprint model based on the sound feature set includes: Acquire the sound features and corresponding voiceprint features in the sound feature set; Dividing the sound features and the corresponding voiceprint features into a training set, a validation set and a test set; Constructing a convolutional network, wherein the convolutional network includes a layer for extracting voiceprint features; Defining a loss function and training parameters, and training the convolutional network using the training set and the validation set; According to the performance results of the validation set, the timing to stop training is determined and the training is stopped.
4. The method for identifying an individual based on sound features according to claim 3, characterized in that: The loss function is: L(A,P,N)=max(d(A,P)-d(A,N)+α,0); Where A is the anchor point, P is the positive sample, N is the negative sample, d is the distance function between samples, and α is the edge margin; The training parameters include at least one of a learning rate, a batch size, a number of iterations, and an optimizer.
5. The method for identifying an individual based on sound features according to claim 1, characterized in that: The training of the voiceprint model based on the sound feature set includes: The voiceprint model is trained using any algorithm of a support vector machine, a Gaussian mixture model or a deep learning model.
6. The method for identifying an individual based on sound features according to claim 1, characterized in that: The preprocessing of the sound data to obtain a sound feature set includes: Using a noise suppression algorithm to perform noise reduction processing on the sound data; Performing normalization processing on the sound data; The processed sound data is collected to obtain the sound feature set.
7. An identity recognition system based on sound features, characterized in that: The system comprises: A first acquisition module, used to acquire voice data of multiple speakers; A preprocessing module, used for preprocessing the sound data to obtain a sound feature set; A first training module, used for training a voiceprint model based on the voice feature set; A second acquisition module is used to acquire recorded voice matching the personal information, and convert the recorded voice into a recorded voiceprint using the voiceprint model; A third acquisition module is used to acquire input speech and convert the input speech into an input voiceprint using the voiceprint model; a calculation module, configured to calculate the similarity between the recorded voiceprint and the input voiceprint, and in response to the similarity being within a threshold range, the input voiceprint matches the personal information; A testing module, configured to perform an authenticity test on the input voice, and in response to the authenticity being within a threshold range, confirm that the matched personal information is the identified identity information; The calculation module also includes: A first selection module is used to select the input voiceprint with a clustering structure, calculate the Euclidean distance between the recorded voiceprint and the input voiceprint, and obtain the similarity; A second selection module is used to select the input voiceprint with high-dimensional features, calculate the Mahalanobis distance between the recorded voiceprint and the input voiceprint, and obtain the similarity; The third selection module is used to select the input voiceprint with directional characteristics and calculate the cosine similarity between the recorded voiceprint and the input voiceprint.
8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the identity recognition method based on sound features as described in any one of claims 1 to 6.
9. An electronic device, characterized in that: include: Processor and memory; The memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the identity recognition method based on sound features as claimed in any one of claims 1 to 6.
Citation Information
Cited By
Group robot multi-sound-source individual identification and hierarchical cooperative positioning method and system
CN121254197A
