TW201833810A - Method and system of authentication based on voiceprint recognition - Google Patents

Abstract

A method and a system of authentication based on voiceprint recognition are disclosed. The method comprises steps of: after receiving a voice data of a user to be identified, acquiring a voiceprint feature of the voice data, and generating a corresponding voiceprint feature vector based on the voiceprint feature; inputting the voiceprint feature vector into a background channel model generated by training in advance, to construct a current voiceprint identification vector corresponding to the voice data; and calculating a spatial distance between the current voiceprint identification vector and a pre-stored standard voiceprint identification vector of the user, and generating a verification result by authenticating the user based on the spatial distance. Thus, the invention can improve an accuracy and an efficiency of user authentication.

Description

 Method and system for identity verification based on voiceprint recognition  

The present invention relates to the field of communications technologies, and in particular, to a method and system for identity verification based on voiceprint recognition.

At present, the business scope of large financial companies involves insurance, banking, investment and other business areas. Each business category usually needs to communicate with customers, and there are many ways to communicate (such as telephone communication or face-to-face communication). Before communicating, verifying the identity of the customer becomes an important part of ensuring business security. In order to meet the immediate needs of the business, financial companies usually use manual methods to analyze and verify the identity of customers. Due to the large customer base, relying on manual discriminant analysis to verify the identity of the customer is not accurate or efficient.

An object of the present invention is to provide a method and system for identity verification based on voiceprint recognition, which aims to improve the accuracy and efficiency of user identity verification.

In order to achieve the above object, the present invention provides a voiceprint recognition-based identity verification method, and the voiceprint recognition-based identity verification method includes the following steps: S1, after receiving the voice data of the user performing the identity verification, acquiring the voice The voiceprint feature of the data is constructed, and the corresponding voiceprint feature vector is constructed based on the voiceprint feature; S2, the voiceprint feature vector is input into the background channel model generated by the pre-training to construct the current voiceprint discrimination vector corresponding to the voice data. S3, calculating a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user, authenticating the user based on the distance, and generating a verification result.

Preferably, the step S1 includes the sub-steps: S11, pre-weighting, framing, and windowing the voice data; S12, performing Fourier transform on each windowing process to obtain a corresponding spectrum; S13, inputting the spectrum The Meyer filter obtains the Mel spectrum from the output; S14, performs cepstrum analysis on the Mel spectrum to obtain the Mel frequency cepstral coefficient MFCC, and forms a corresponding voiceprint feature vector based on the Mel frequency cepstral coefficient MFCC.

Preferably, the step S3 comprises the substeps: S31, calculating a cosine distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user: , Identify the vector for the standard voiceprint, For the current voiceprint discrimination vector; S32, if the cosine distance is less than or equal to the preset distance threshold, generate verification information; S33, if the cosine distance is greater than the preset distance threshold, generate information that the verification fails.

Preferably, the background channel model is a Gaussian mixture model, and the step S1 includes the steps of: acquiring a preset number of voice data samples, and acquiring voiceprint features corresponding to the voice data samples, and based on the sounds corresponding to the voice data samples. The voice feature vector corresponding to each voice data sample is constructed; the voiceprint feature vector corresponding to each voice data sample is divided into a first proportional training set and a second proportional verification set, and the first ratio and the second ratio are And less than or equal to 1; the Gaussian mixture model is trained by using the voiceprint feature vector in the training set, and after the training is completed, the verification set is used to verify the accuracy of the trained Gaussian mixture model; if the accuracy is greater than the pre- If the threshold is set, the model training ends, and the trained Gaussian mixture model is used as the background channel model of the step S2, or if the accuracy is less than or equal to the preset threshold, the number of the voice data samples is increased, and based on the increased The voice data samples are retrained.

Preferably, the step S3 is replaced by: calculating a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vectors, obtaining a minimum spatial distance, and authenticating the user based on the minimum spatial distance. And generate verification results.

To achieve the above object, the present invention also provides a voiceprint recognition based authentication system, the voiceprint recognition based authentication system comprising: a first acquisition module, configured to receive a voice of a user performing authentication After the data, the voiceprint feature of the voice data is obtained, and the corresponding voiceprint feature vector is constructed based on the voiceprint feature; and the module is configured to input the voiceprint feature vector into the background channel model generated by the pre-training to construct The current voiceprint discrimination vector corresponding to the voice data; the first verification module is configured to calculate a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user, and perform the user on the basis based on the distance Authenticate and generate verification results.

Preferably, the first acquiring module is specifically configured to perform pre-weighting, framing, and windowing processing on the voice data; performing Fourier transform on each window to obtain a corresponding spectrum; and inputting the spectrum into the mel filter The output obtains the Mel spectrum; the cepstrum analysis is performed on the Mel spectrum to obtain the Mel frequency cepstral coefficient MFCC, and the corresponding Moiré feature vector is composed based on the Mel frequency cepstral coefficient MFCC.

Preferably, the first verification module is specifically configured to calculate a cosine distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user: , Identify the vector for the standard voiceprint, A vector for identifying the current voiceprint; if the cosine distance is less than or equal to a preset distance threshold, generating information for verifying the pass; if the cosine distance is greater than the preset distance threshold, generating information that the verification fails.

Preferably, the voiceprint recognition-based identity verification system further includes: a second acquisition module, configured to acquire a preset number of voice data samples, and obtain voiceprint features corresponding to each voice data sample, and based on each voice The voiceprint feature corresponding to the data sample constructs a voiceprint feature vector corresponding to each voice data sample; the partitioning module is configured to divide the voiceprint feature vector corresponding to each voice data sample into a first proportional training set and a second proportional verification. The set, the sum of the first ratio and the second ratio is less than or equal to 1; the training module is configured to train the Gaussian mixture model by using the voiceprint feature vector in the training set, and after the training is completed, use the verification set pair training The accuracy of the subsequent Gaussian mixture model is verified; the processing module is configured to end the model training if the accuracy is greater than a preset threshold, and use the trained Gaussian mixture model as the background channel model, or if the accuracy is If the threshold is less than or equal to the preset threshold, the number of the voice data samples is increased, and the training is performed based on the added voice data samples.

Preferably, the first verification module is replaced with a second verification module, configured to calculate a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vectors, to obtain a minimum spatial distance, based on the The minimum spatial distance authenticates the user and generates a verification result.

The invention has the beneficial effects that the background channel model generated by the pre-training of the invention is obtained by mining and comparing a large amount of speech data, and the model can accurately describe the user's speech while retaining the user's voiceprint feature to the utmost extent. The background voiceprint feature can be removed at the time of recognition, and the intrinsic feature of the user voice can be extracted, which can greatly improve the accuracy of user identity verification and improve the efficiency of identity verification.

1‧‧‧Electronic device

10‧‧‧System for identity verification based on voiceprint recognition

101‧‧‧First acquisition module

102‧‧‧Building module

103‧‧‧First verification module

11‧‧‧ memory

12‧‧‧ Processor

13‧‧‧ display

S1‧‧‧ steps

S11‧‧‧ substeps

S12‧‧‧ substeps

S13‧‧‧ substeps

S14‧‧‧ substeps

S2‧‧‧ steps

S3‧‧‧ steps

S31‧‧‧ substeps

S32‧‧‧ substeps

S33‧‧‧ substeps

FIG. 1 is a schematic flow chart of a preferred embodiment of a method for authenticating voiceprint recognition based authentication according to the present invention.

Fig. 2 is a schematic diagram showing the refinement flow of step S1 shown in Fig. 1.

Fig. 3 is a schematic diagram showing the refinement flow of step S3 shown in Fig. 1.

4 is a schematic diagram of an operating environment of a preferred embodiment of a voiceprint recognition based authentication system according to the present invention.

FIG. 5 is a schematic structural diagram of a system for authenticating a voiceprint recognition based authentication method according to the present invention.

The principles and features of the present invention are described in the following with reference to the accompanying drawings.

As shown in FIG. 1 , FIG. 1 is a schematic flowchart of an embodiment of a voiceprint recognition based identity verification method according to an embodiment of the present invention. The voiceprint recognition based identity verification method may be performed by a voiceprint recognition based identity verification. The system is executed, the system can be implemented by software and/or hardware, and the system can be integrated in the server. The voiceprint recognition-based authentication method includes the following steps: Step S1: After receiving the voice data of the user who performs the authentication, acquire the voiceprint feature of the voice data, and construct a corresponding voiceprint based on the voiceprint feature. Feature vector; in this embodiment, the voice data is collected by the voice collection device (the voice collection device is, for example, a microphone), and the voice collection device sends the collected voice data to the voice recognition-based identity verification system.

When collecting voice data, you should try to prevent environmental noise and interference from voice collection equipment. The voice collection device maintains an appropriate distance from the user, and tries not to use a large voice acquisition device. The power supply preferably uses the commercial power and keeps the current stable; the sensor should be used when recording the telephone. Before extracting the voiceprint features in the voice data, the voice data can be denoised to further reduce interference. In order to extract the voiceprint feature of the voice data, the collected voice data is voice data of a preset data length, or voice data larger than a preset data length.

The voiceprint feature includes a plurality of types, such as a wide-frequency voiceprint, a narrow-frequency voiceprint, an amplitude voiceprint, etc., and the voiceprint feature of the present embodiment is preferably a Mel Frequency Cepstrum Coefficient (MFCC). ). When constructing the corresponding voiceprint feature vector, the voiceprint feature of the voice data is composed into a feature data matrix, which is the voiceprint feature vector of the voice data.

Step S2, the voiceprint feature vector is input into the background channel model generated by the pre-training to construct a current voiceprint discrimination vector corresponding to the voice data; wherein the voiceprint feature vector is input into the background channel model generated by the pre-training, preferably The background channel model is a Gaussian mixture model, and the background channel model is used to calculate the voiceprint feature vector to obtain a corresponding current voiceprint discrimination vector (ie, i-vector).

Specifically, the calculation process includes: 1) selecting a Gaussian model: first, using parameters in the general background channel model to calculate the likelihood value of each frame of data in different Gaussian models, by using a column of likelihood logarithmic values Parallel sorting, select the first N Gaussian models, and finally obtain a matrix of values per frame of data in the mixed Gaussian model: Loglike = E ( X ) * D ( X ) ^-1 * X ^T -0.5* D ( X ) ^-1 *( X . ² ) ^T , where Loglike is a likelihood logarithmic matrix, E(X) is a mean matrix trained by a general background channel model, D(X) is a covariance matrix, and X is a data matrix, X ² Square the value of each matrix.

2) Calculate the posterior probability: X * XT calculation is performed for each frame of data X to obtain a symmetric matrix, which can be simplified into a lower triangular matrix, and the elements are arranged in order of 1 row, and become an N frame multiplied by the lower triangle. A vector of latitudes of the matrix is calculated, and the vectors of all frames are combined into a new data matrix, and the matrix of the common variability of the computer rate in the general background model is simplified, and each matrix is also reduced to a lower triangular matrix. A matrix similar to the new data matrix, the likelihood log value of each frame of data in the selected Gaussian model is calculated by using the mean matrix and the common variance matrix in the universal background channel model, and then Softmax regression is performed, and finally the positive return is performed. The operation results in a posterior probability distribution of each frame in the mixed Gaussian model, and the probability distribution vector of each frame is composed into a probability matrix.

3) Extract the current voiceprint discrimination vector: first calculate the first-order and second-order coefficients, and the first-order coefficient calculation can be obtained by summing the probability matrix columns: Where Gamma _i is the ith element of the first-order coefficient vector, and loglikes _ji is the j-th row of the probability matrix, the ith element.

The second-order coefficients can be obtained by multiplying the transposition of the probability matrix by the data matrix: X = Loglike ^T * feats , where X is a second-order coefficient matrix, loglike is a probability matrix, and feats is a feature data matrix.

After the first-order and second-order coefficients are calculated, the primary term and the quadratic term are calculated in parallel, and then the current voiceprint discrimination vector is calculated by the primary term and the quadratic term.

Preferably, the background channel model is a Gaussian mixture model, and before the step S1, the method includes: acquiring a preset number of voice data samples, and acquiring voiceprint features corresponding to each voice data sample, and based on the voiceprint corresponding to each voice data sample The feature maps the voiceprint feature vector corresponding to each voice data sample; the voiceprint feature vector corresponding to each voice data sample is divided into a first proportional training set and a second proportional verification set, and the sum of the first ratio and the second ratio Less than or equal to 1; the Gaussian mixture model is trained by using the voiceprint feature vector in the training set, and after the training is completed, the verification set is used to verify the accuracy of the trained Gaussian mixture model; if the accuracy is greater than the preset Threshold, the model training ends, the trained Gaussian mixture model is used as the background channel model of the step S2, or if the accuracy is less than or equal to the preset threshold, the number of the voice data samples is increased, and based on the increased The voice data samples are retrained.

When the Gaussian mixture model is trained by using the voiceprint feature vector in the training set, the likelihood probability of the extracted D-dimensional voiceprint feature can be expressed by K Gaussian components: , where P(x) is the probability that the speech data samples are generated by the Gaussian mixture model (mixed Gaussian model), w _k is the weight of each Gaussian model, and p(x|k) is the sample generated by the kth Gaussian model. Probability, K is the number of Gaussian models.

The parameters of the entire Gaussian mixture model can be expressed as: { w _i , μ _i , Σ _i }, w _i is the weight of the i-th Gaussian model, μ _i is the mean of the i-th Gaussian model, and Σ _i is the i-th Gaussian The covariance of the model. Training the Gaussian mixture model can use an unsupervised EM algorithm. After the training is completed, the Gaussian mixture model weight vector, constant vector, N common variance matrix, and the mean multiplied by the common variance matrix are obtained, which is a trained Gaussian mixture model.

Step S3: Calculate a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user, perform identity verification on the user based on the distance, and generate a verification result.

There are various distances between the vector and the vector, including the cosine distance and the Euclidean distance, etc. Preferably, the spatial distance of the present embodiment is a cosine distance, and the cosine distance is measured by using the cosine of the angles of the two vectors in the vector space. A measure of the magnitude of the difference between two individuals.

The standard voiceprint discriminant vector is a voiceprint discriminant vector obtained and stored in advance, and the standard voiceprint discriminant vector carries the identifier information of the corresponding user when stored, which can accurately represent the identity of the corresponding user. The stored voiceprint discrimination vector is obtained according to the identification information provided by the user before calculating the spatial distance.

Wherein, when the calculated spatial distance is less than or equal to the preset distance threshold, the verification passes, and vice versa, the verification fails.

Compared with the prior art, the background channel model generated by the pre-training in this embodiment is obtained by mining and comparing a large amount of voice data, and the model can accurately describe the user's voice while retaining the user's voiceprint feature to the utmost extent. The background voiceprint feature, and can remove this feature at the time of recognition, and extract the inherent features of the user voice, can greatly improve the accuracy of the user identity verification, and improve the efficiency of the identity verification; in addition, this embodiment is sufficient It utilizes the voiceprint features associated with the vocal vocal in the human voice. This voiceprint feature does not require restrictions on the text, and thus has greater flexibility in the process of identification and verification.

In a preferred embodiment, as shown in FIG. 2, based on the embodiment of FIG. 1 above, the step S1 includes: sub-step S11, pre-weighting, framing, and windowing the voice data. Processing; in this embodiment, after receiving the voice data of the user who performs identity verification, the voice data is processed. Wherein, the pre-weighting process is actually a high-pass filtering process, filtering out the low-frequency data, so that the high-frequency characteristics in the voice data are more prominent. Specifically, the transfer function of the high-pass filter is: H ( Z )=1- αZ ^-1 , wherein Z is the speech data, α is a constant coefficient, preferably, the value of α is 0.97; since the sound signal exhibits smoothness only in a short time, the sound signal is divided into N segments of short time signals (ie, N frames) ), and in order to avoid the loss of the continuity feature of the sound, there is a repeating area between adjacent frames, and the repeating area is generally 1/2 of the length of each frame; after framing the voice data, each frame signal is regarded as stable The signal is processed, but the existence of the Gibbs effect, the start frame and the end frame of the speech data are discontinuous, and after the framing, the original speech is further deviated. Therefore, the speech data needs to be windowed.

Sub-step S12, performing Fourier transform on each window to obtain a corresponding spectrum; sub-step S13, inputting the spectrum into the mel filter to output a mel spectrum; sub-step S14, performing cepst analysis on the mel spectrum A Mel frequency cepstral coefficient MFCC is obtained, and a corresponding voiceprint feature vector is formed based on the Mel frequency cepstral coefficient MFCC. The cepstrum analysis is, for example, taking logarithm and inverse transform. The inverse transform is generally implemented by DCT discrete cosine transform, and the second to thirteenth coefficients after DCT are taken as MFCC coefficients. The Mel frequency cepstral coefficient MFCC is the voiceprint feature of the speech data of this frame. The Mel frequency cepstral coefficient MFCC of each frame is composed into a feature data matrix, which is the voiceprint feature vector of the speech data.

In a preferred embodiment, as shown in FIG. 3, based on the embodiment of FIG. 1 above, the step S3 includes: sub-step S31, calculating the current voiceprint discrimination vector and the pre-stored user's The cosine distance between standard voiceprint discrimination vectors: , Identify the vector for the standard voiceprint, For the current voiceprint discrimination vector; sub-step S32, if the cosine distance is less than or equal to the preset distance threshold, generate verification pass information; sub-step S33, if the cosine distance is greater than the preset distance threshold, generate verification Information passed.

In a preferred embodiment, based on the embodiment of FIG. 1 above, the step S3 is replaced by: calculating a spatial distance between the current voiceprint discrimination vector and each of the pre-stored standard voiceprint discrimination vectors, Get the minimum spatial distance, authenticate the user based on the minimum spatial distance, and generate the verification result.

This embodiment is different from the embodiment of FIG. 1 in that the present embodiment does not carry the user's identification information when storing the standard voiceprint authentication vector, and calculates the current voiceprint identification vector and the pre-stored each when verifying the identity of the user. The standard voiceprint discriminates the spatial distance between the vectors and obtains a minimum spatial distance. If the minimum spatial distance is less than a preset distance threshold (the distance threshold is the same as or different from the distance threshold of the above embodiment), the verification passes, Otherwise the verification fails.

Please refer to FIG. 4, which is a schematic diagram of an operating environment of a preferred embodiment of the voiceprint recognition based authentication system 10 of the present invention.

In the present embodiment, the voiceprint recognition based authentication system 10 is installed and operates in the electronic device 1. The electronic device 1 may be a computing device such as a desktop computer, a notebook, a palmtop computer, and a server. The electronic device 1 may include, but is not limited to, a memory 11, a processor 12, and a display 13. The first figure shows only the electronic device 1 with the elements 11-13, but it should be understood that not all illustrated elements are required to be implemented, and more or fewer elements may be implemented instead.

The memory 11 may be an internal memory unit of the electronic device 1, such as a hard disk or a memory of the electronic device 1, in some embodiments. The memory 11 may also be an external memory device of the electronic device 1 in other embodiments, such as a plug-in hard disk equipped on the electronic device 1, a smart media card (SMC), and a secure digital device (Secure Digital, SD) card, flash card, etc. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external memory device. The memory 11 is used to store application software installed on the electronic device 1 and various types of materials, such as program code of the system 10 for voiceprint recognition based authentication. The memory 11 can also be used to temporarily store data that has been output or is about to be output.

The processor 12, in some embodiments, may be a central processing unit (CPU), a microprocessor or other data processing chip for running program code or processing data stored in the memory 11, for example, performing sound based Pattern recognition for the authentication system 10 etc.

The display 13 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch sensor, or the like in some embodiments. The display 13 is for displaying information processed in the electronic device 1 and a user interface for displaying visualization, such as a voiceprint recognition interface or the like. The components 11-13 of the electronic device 1 communicate with each other through a system bus bar.

Please refer to FIG. 5, which is a functional block diagram of a preferred embodiment of the voiceprint recognition based authentication system 10 of the present invention. In this embodiment, the voiceprint recognition based authentication system 10 can be divided into one or more modules, one or more modules being stored in the memory 11 and being processed by one or more processors ( This embodiment is executed by the processor 12) to complete the present invention. For example, in FIG. 5, the voiceprint recognition based authentication system 10 can be divided into a detection module 21, an identification module 22, a replication module 23, an installation module 24, and a startup module 25. The module referred to in the present invention refers to a series of computer program instruction segments capable of performing a specific function, and is more suitable than the program for describing the execution process of the voiceprint recognition-based authentication system 10 in the electronic device 1, wherein: The acquiring module 101 is configured to acquire a voiceprint feature of the voice data after receiving the voice data of the user who performs the authentication, and construct a corresponding voiceprint feature vector based on the voiceprint feature; in this embodiment, the voice data It is collected by the voice collection device (the voice collection device is, for example, a microphone), and the voice collection device sends the collected voice data to the system based on the voiceprint recognition identity verification.

When collecting voice data, you should try to prevent environmental noise and interference from voice collection equipment. The voice collection device maintains an appropriate distance from the user, and tries not to use a large voice acquisition device. The power supply preferably uses the commercial power and keeps the current stable; the sensor should be used when recording the telephone. Before extracting the voiceprint features in the voice data, the voice data can be denoised to further reduce interference. In order to extract the voiceprint feature of the voice data, the collected voice data is voice data of a preset data length, or voice data larger than a preset data length.

The voiceprint features include a plurality of types, such as a wide-band voiceprint, a narrow-band voiceprint, an amplitude voiceprint, etc., and the voiceprint feature of the present embodiment is preferably a Mel Frequency Cepstrum Coefficient (MFCC). . When constructing the corresponding voiceprint feature vector, the voiceprint feature of the voice data is composed into a feature data matrix, which is the voiceprint feature vector of the voice data.

The building module 102 is configured to input the voiceprint feature vector into the background channel model generated by the pre-training to construct a current voiceprint discrimination vector corresponding to the voice data; wherein the voiceprint feature vector is input into the background channel generated by the pre-training The model, preferably, the background channel model is a Gaussian mixture model, and the background channel model is used to calculate the voiceprint feature vector to obtain a corresponding current voiceprint discrimination vector (ie, i-vector).

Specifically, the calculation process includes:

1) Select the Gaussian model: First, use the parameters in the general background channel model to calculate the likelihood value of each frame of data in different Gaussian models. By sorting the columns of the likelihood logarithmic matrix in parallel, select the first N Gaussian models. Finally, obtain a matrix of values per frame of data in the mixed Gaussian model: Loglike = E ( X ) * D ( X ) ^-1 * X ^T -0.5* D ( X ) ^-1 *( X . ² ) ^T , where Loglike is the likelihood logarithmic matrix, E(X) is the mean matrix trained by the general background channel model, D(X) is the covariance matrix, X is the data matrix, and X. ² is the square of each value of the matrix.

2) Calculate the posterior probability: X * XT calculation is performed for each frame of data X to obtain a symmetric matrix, which can be simplified into a lower triangular matrix, and the elements are arranged in order of 1 row, and become an N frame multiplied by the lower triangle. A vector of latitudes of the matrix is calculated, and the vectors of all frames are combined into a new data matrix. At the same time, the computer-wide covariance matrix of the general background model is simplified, and each matrix is also reduced to a lower triangular matrix. A matrix similar to the data matrix, the likelihood logarithm of each frame of data in the selected Gaussian model is calculated by the mean matrix and the common variance matrix in the general background channel model, and then Softmax regression is performed, and finally the normalization operation is performed. Obtain a posterior probability distribution of each frame in the mixed Gaussian model, and construct a probability matrix of the probability distribution vector of each frame.

3) Extract the current voiceprint discrimination vector: first calculate the first-order and second-order coefficients, and the first-order coefficient calculation can be obtained by summing the probability matrix columns: Where Gamma _i is the ith element of the first-order coefficient vector, and loglikes _ji is the j-th row of the probability matrix, the ith element.

The second-order coefficients can be obtained by multiplying the transposition of the probability matrix by the data matrix: X = Loglike ^T * feats , where X is a second-order coefficient matrix, loglike is a probability matrix, and feats is a feature data matrix.

After the first-order and second-order coefficients are calculated, the primary term and the quadratic term are calculated in parallel, and then the current voiceprint discrimination vector is calculated by the primary term and the quadratic term.

Preferably, the background channel model is a Gaussian mixture model, and the voiceprint recognition based authentication system further includes: a second acquisition module, configured to acquire a preset number of voice data samples, and obtain sounds corresponding to the voice data samples. The pattern features, and the voiceprint feature vector corresponding to each voice data sample is constructed based on the voiceprint features corresponding to each phonetic data sample; the segmentation module is configured to divide the voiceprint feature vector corresponding to each phonetic data sample into the first ratio training. And the second ratio of the verification set, the sum of the first ratio and the second ratio is less than or equal to 1; the training module is configured to train the Gaussian mixture model by using the voiceprint feature vector in the training set, and after the training is completed The verification set is used to verify the accuracy of the trained Gaussian mixture model; the processing module is configured to end the model training if the accuracy is greater than a preset threshold, and the trained Gaussian mixture model is used as the background channel model. Or, if the accuracy is less than or equal to the preset threshold, increase the number of the voice data samples, and based on the increased voice data sample Retrain.

When the Gaussian mixture model is trained by using the voiceprint feature vector in the training set, the likelihood probability of the extracted D-dimensional voiceprint feature can be expressed by K Gaussian components: , where P(x) is the probability that the speech data samples are generated by the Gaussian mixture model (mixed Gaussian model), w _k is the weight of each Gaussian model, and p(x|k) is the sample generated by the kth Gaussian model. Probability, K is the number of Gaussian models.

The parameters of the entire Gaussian mixture model can be expressed as: { w _i , μ _i , Σ _i }, w _i is the weight of the i-th Gaussian model, μ _i is the mean of the i-th Gaussian model, and Σ _i is the i-th Gaussian The covariance of the model. Training the Gaussian mixture model can use an unsupervised EM algorithm. After the training is completed, the Gaussian mixture model weight vector, constant vector, N common variance matrix, and the mean multiplied by the common variance matrix are obtained, which is a trained Gaussian mixture model.

The first verification module 103 is configured to calculate a spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user, perform identity verification on the user based on the distance, and generate a verification result.

There are various distances between the vector and the vector, including the cosine distance and the Euclidean distance, etc. Preferably, the spatial distance of the present embodiment is a cosine distance, and the cosine distance is measured by using the cosine of the angles of the two vectors in the vector space. A measure of the magnitude of the difference between two individuals.

The standard voiceprint discriminant vector is a voiceprint discriminant vector obtained and stored in advance, and the standard voiceprint discriminant vector carries the identifier information of the corresponding user when stored, which can accurately represent the identity of the corresponding user. The stored voiceprint discrimination vector is obtained according to the identification information provided by the user before calculating the spatial distance.

Wherein, when the calculated spatial distance is less than or equal to the preset distance threshold, the verification passes, and vice versa, the verification fails.

In a preferred embodiment, based on the foregoing embodiment of FIG. 5, the first acquiring module 101 is specifically configured to perform pre-weighting, framing, and windowing on the voice data; Perform Fourier transform to obtain the corresponding spectrum; input the spectrum into the Meyer filter to output the Mel spectrum; perform cepstral analysis on the Mel spectrum to obtain the Mel frequency cepstral coefficient MFCC, based on the Mel frequency cepstral coefficient The MFCC constitutes a corresponding voiceprint feature vector.

Wherein, the pre-weighting process is actually a high-pass filtering process, filtering out the low-frequency data, so that the high-frequency characteristics in the voice data are more prominent. Specifically, the transfer function of the high-pass filter is: H ( Z )=1- αZ ^-1 , wherein Z is the speech data, α is a constant coefficient, preferably, the value of α is 0.97; since the sound signal exhibits smoothness only in a short time, the sound signal is divided into N segments of short time signals (ie, N frames) ), and in order to avoid the loss of the continuity feature of the sound, there is a repeating area between adjacent frames, and the repeating area is generally 1/2 of the length of each frame; after framing the voice data, each frame signal is regarded as stable The signal is processed, but the existence of the Gibbs effect, the start frame and the end frame of the speech data are discontinuous, and after the framing, the original speech is further deviated. Therefore, the speech data needs to be windowed.

Among them, the cepstrum analysis is, for example, taking the logarithm and performing the inverse transform. The inverse transform is generally realized by DCT discrete cosine transform, and the second to thirteenth coefficients after the DCT are taken as the MFCC coefficients. The Mel frequency cepstral coefficient MFCC is the voiceprint feature of the speech data of this frame. The Mel frequency cepstral coefficient MFCC of each frame is composed into a feature data matrix, which is the voiceprint feature vector of the speech data.

In a preferred embodiment, based on the foregoing embodiment of FIG. 5, the first verification module 103 is specifically configured to calculate between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user. Cosine distance: , Identify the vector for the standard voiceprint, A vector for identifying the current voiceprint; if the cosine distance is less than or equal to a preset distance threshold, generating information for verifying the pass; if the cosine distance is greater than the preset distance threshold, generating information that the verification fails.

In a preferred embodiment, based on the embodiment of FIG. 4, the first verification module is replaced with a second verification module for calculating the current voiceprint identification vector and the pre-stored standards. The spatial distance between the voiceprint discrimination vectors obtains a minimum spatial distance, and the user is authenticated based on the minimum spatial distance, and a verification result is generated.

This embodiment is different from the embodiment of FIG. 5 in that the present embodiment does not carry the user's identification information when storing the standard voiceprint authentication vector, and calculates the current voiceprint identification vector and the pre-stored each when verifying the identity of the user. The standard voiceprint discriminates the spatial distance between the vectors and obtains a minimum spatial distance. If the minimum spatial distance is less than a preset distance threshold (the distance threshold is the same as or different from the distance threshold of the above embodiment), the verification passes, Otherwise the verification fails.

The above is only the preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalents, improvements, etc., which are within the spirit and scope of the present invention, should be included in the scope of the present invention. within.

Claims (10)

 A method for authentication based on voiceprint recognition includes the steps of: after receiving a voice material of a user performing an identity verification, acquiring a voiceprint feature of the voice data, and constructing a corresponding one based on the voiceprint feature a voiceprint feature vector; inputting the voiceprint feature vector into a background channel model generated by pre-training to construct a current voiceprint discrimination vector corresponding to the voice data; and calculating the current voiceprint discrimination vector and the pre-stored user A spatial distance between a standard voiceprint discrimination vector, the identity verification is performed on the user based on the spatial distance, and a verification result is generated.    The method for authenticating voiceprint recognition based on claim 1, wherein the voiceprint feature of the voice material is obtained after receiving the voice data of the user who performs the identity verification, and based on the The step of constructing the corresponding voiceprint feature vector of the voiceprint feature comprises the substeps of: performing a pre-weighting, a framing and a windowing process on the voice data; performing a Fourier transform on each windowing process to obtain a corresponding spectrum. Entering the spectrum into a mel filter to output a mel spectrum; and performing a cepstrum analysis on the mel spectrum to obtain a Mel frequency cepstral coefficient, based on the Mel frequency cepstral coefficient composition A striated feature vector.   The method for authenticating voiceprint recognition based on claim 1, wherein the calculating the spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user is based on the space The step of authenticating the user and generating the verification result includes the substep of calculating a cosine distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user: , Identify the vector for the standard voiceprint, And identifying a vector for the current voiceprint; if the cosine distance is less than or equal to a preset distance threshold, generating a verification pass information; and if the cosine distance is greater than the preset distance threshold, generating a verification failure News.  The method of voiceprint recognition based authentication according to any one of claims 1 to 3, wherein the sound of the voice material is acquired after the voice material of the user who has performed the identity verification is received The step of constructing the corresponding voiceprint feature vector based on the voiceprint feature includes the steps of: acquiring a preset number of plurality of voice data samples, and acquiring a voiceprint feature corresponding to each voice data sample, and based on each The voiceprint feature corresponding to the voice data sample constructs a voiceprint feature vector corresponding to each voice data sample; and the voiceprint feature vector corresponding to each voice data sample is divided into a training set of a first ratio and a second ratio a verification set, the sum of the first ratio and the second ratio is less than or equal to 1; using the voiceprint feature vector in the training set to train a Gaussian mixture model, and after the training is completed, using the verification set pair after training The accuracy of the Gaussian mixture model is verified; and if the accuracy is greater than a predetermined threshold, the Gaussian mixture model is trained to end the Gaussian after training And combining the voiceprint feature vector into the background channel model generated by the pre-training to construct the background channel model in the step of the current voiceprint discrimination vector corresponding to the voice data, or if the accuracy is less than Or equal to the preset threshold, the number of the voice data samples is increased, and the training is resumed based on the added voice data samples.    The method for authenticating voiceprint recognition based on claim 1 or 2, wherein the calculating the spatial distance between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user is based on The spatial distance is used to perform the identity verification on the user, and the step of generating the verification result is replaced by: calculating a plurality of spatial distances between the current voiceprint discrimination vector and the pre-stored plurality of standard voiceprint discrimination vectors, and obtaining a minimum The spatial distance, an identity verification is performed on the user based on the minimum spatial distance, and a verification result is generated.    A voiceprint recognition-based authentication system, comprising: a first acquisition module, configured to acquire a voiceprint feature of a voice data after receiving a voice data of a user performing an identity verification, and based on The voiceprint feature constructs a corresponding voiceprint feature vector; a building module is configured to input the voiceprint feature vector into a background channel model generated by pre-training to construct a current voiceprint discrimination vector corresponding to the voice data. a first verification module, configured to calculate a spatial distance between the current voiceprint discrimination vector and a pre-stored standard voiceprint discrimination vector of the user, perform the identity verification on the user based on the spatial distance, and generate A verification result.    The system for authenticating voiceprint recognition based on claim 6, wherein the first obtaining module is specifically configured to perform a pre-weighting, a framing, and a windowing process on the voice data; Windowing processing performs a Fourier transform to obtain a corresponding spectrum; inputting the spectrum into a Meyer filter to output a Merr spectrum; performing a cepstrum analysis on the Mel spectrum to obtain a Mel frequency cepstral coefficient And corresponding to a voicing feature vector based on the Mel frequency cepstral coefficient.   The system for authenticating voiceprint recognition based on claim 6, wherein the first verification module is specifically configured to calculate between the current voiceprint discrimination vector and the pre-stored standard voiceprint discrimination vector of the user. One cosine distance: , Identify the vector for the standard voiceprint, And identifying a vector for the current voiceprint; if the cosine distance is less than or equal to a preset distance threshold, generating a verification pass information; if the cosine distance is greater than the preset distance threshold, generating a verification failure message .  The system for voiceprint recognition based authentication according to any one of claims 6 to 8, further comprising: a second acquisition module, configured to acquire a preset number of multiple voice data samples, and obtain each a voiceprint feature corresponding to the voice data sample, and constructing a voiceprint feature vector corresponding to each voice data sample based on the voiceprint feature corresponding to each voice data sample; a partitioning module for corresponding to each voice data sample The voiceprint feature vector is divided into a training set of a first ratio and a verification set of a second ratio, the sum of the first ratio and the second ratio being less than or equal to 1; a training module for utilizing the training set The voiceprint feature vector trains a Gaussian mixture model, and after the training is completed, uses the verification set to verify an accuracy rate of the trained Gaussian mixture model; a processing module is used for the accuracy rate If the accuracy is less than or equal to the preset threshold, the Gaussian mixture model is trained to end the training, and the Gaussian mixture model is used as the background channel model, or if the accuracy is less than or equal to the preset threshold, The number of the voice data samples is increased, and the training is resumed based on the added voice data samples.    The system according to claim 6 or 7, wherein the first verification module is replaced with a second verification module for calculating the current voiceprint discrimination vector and the pre-stored number. A plurality of spatial distances between the standard voiceprint discrimination vectors obtain a minimum spatial distance, perform an identity verification on the user based on the minimum spatial distance, and generate a verification result.