CN103973453A - Vocal print secret key generating method and device and logging-in method and system based on vocal print secret key - Google Patents

Abstract

The invention discloses a method and device for generating a voiceprint key, and a method and system for logging in based on a voiceprint key. key method to generate the first voiceprint key and store it on the authentication server; when not logging into the data management system for the first time, input the second voice information to generate the second voiceprint key, and the authentication server will combine the second voiceprint key with the first The voiceprint key performs authentication matching, and when the authentication matching is completed, the authentication confirmation information is sent to the mobile client for login. The invention ensures the safety of the mobile client accessing the data management system by adopting the voiceprint key method, and is beneficial to real-time remote supervision of the database.

Description

Voiceprint key generation method and device, and voiceprint key-based login method and system

technical field

The invention relates to computer security technology, in particular to a voiceprint key generation method and device, and a voiceprint key-based login method and system.

Background technique

With the continuous introduction of computer technology, a large number of application data platforms have been applied, and cloud data or big data data management platforms such as e-commerce, data service providers, and operators have been developed, such as Amazon's Amazon Web Services (AWS, Amazon Web Services) platform, domestic Alibaba Cloud, Woyun and other platforms. The powerful computing capabilities of these data management platforms have been widely used in the field of national production, such as 12306 train ticket booking website, Alibaba's Taobao platform, etc. To realize the storage and supervision of massive user data, the management and maintenance burden of the database administrator is relatively heavy. In addition, the geographic location of the database data center (IDC) where the data management platform is located often has a certain physical distance from the office area of the database administrator. In order to maintain and manage data resources more conveniently, the data management system is generally mapped to the public network, and the address of the data management system mapped to the public network is logged in through the public network IP to manage and maintain the database of the data management platform.

At present, there are the following defects in mapping the database management system to the public network, logging into the data management system through the public network IP, and managing and maintaining the database of the data management platform: because the database of the data management platform carries a large amount of data resources, it needs The data management platform always pays attention to the situation of the database. When the staff of the data management platform are not in the office area (the computers in the office area have strengthened the firewall defense settings, that is, they are in the security domain), they cannot log in and access the database in real time through computer terminals outside the office area. Management system for real-time maintenance and control of the database.

In order to realize the real-time monitoring of the data management system, the mobile client can be used to log in to the data management system to supervise the data management system. However, if the traditional password is used for security authentication when the mobile client logs in to the database management system, the following problems exist:

1. When the password set by the user is too simple, it is vulnerable to network attacks and the password is easy to be cracked; if it is too complicated, it is inconvenient to remember, and it is easy to cause the password to be stolen with the help of other recording methods.

2. Passwords set in traditional ways are easy to be cracked by attackers using certain means, such as information related to user identity (ID number, birthday, etc.); or dictionary attack method (dictionary attack);

In addition, if the account information is a shared resource, the system cannot distinguish the staff information of the login behavior, which may easily lead to legal disputes.

If biometric identification technology (fingerprint, facial recognition) is used for security authentication, the following problems exist:

1. Confidentiality of biometric samples; for example, attackers can steal user fingerprints from items such as cups touched by users, and hidden image acquisition devices may steal facial feature information of users;

2. Since the biometric feature is a feature based on statistical characteristics, it has fuzziness within the class. Fuzziness means that due to various factors such as position, light, angle, background, etc., there will be certain differences between multiple biometric samples collected by the sensor for the same user, which will easily lead to system misjudgment (especially face recognition has these problems);

3. Most biological templates are not discrete, but real vectors obtained by using some signal processing technology in the continuous domain, which may easily lead to misjudgment deviations.

To sum up, the current existing data management system, if the staff is not in the office area, cannot use the computer in the office area to realize real-time and effective monitoring and management; use the existing mobile client to supervise the data management system, and there is a security authentication problem Therefore, there is an urgent need for a security authentication method, which can be used to implement security authentication of the data management system through the mobile client when leaving the office area, and realize real-time monitoring and management of the data management platform.

Contents of the invention

In order to solve the above technical problems, the present invention provides a voiceprint key generation method and device, and a voiceprint key login method and system, which can use voice information to generate a voiceprint key and ensure the use of mobile clients to log in to the data management system. Security, to achieve real-time supervision of the data management system.

In order to achieve the purpose of the above invention, the present invention discloses a voiceprint key generation method, including:

Two voiceprint feature extraction methods are used to extract the voiceprint feature of the voice information respectively, and two voiceprint feature vector sets are obtained;

Calculating each set of voiceprint feature vectors to obtain a set of matching score vectors of the first preset number;

Build and fuse corresponding elements in each matching score vector set into a binary group;

The constructed and fused binary group is used as the coordinate point to fit the curve analytical formula, and the coefficients of the curve analytical formula are sorted and discretized as the voiceprint key.

Further, the two voiceprint feature extraction methods include any two of the following methods: Mel frequency cepstral coefficient MFCC, Mel linear spectral frequency MFSL, linear predictive cepstral coefficient LPCC, and residual phase.

Further, calculating each set of voiceprint feature vectors, and respectively obtaining a set of matching score vectors with a first preset number includes: inputting the vectors of each set of voiceprint feature vectors into the Gaussian mixture model-universal background model GMM-UBM or a GMM model to respectively obtain a set of matching score vectors of the first preset value.

On the other hand, this application also provides a voiceprint-based key login method, including:

When the mobile client logs into the data management system for the first time, it inputs the first voice information to generate the first voiceprint key and stores it on the authentication server; when it is not the first login to the data management system, it inputs the second voice information to generate the second voice print key. The authentication server authenticates and matches the second voiceprint key with the first voiceprint key, and when the authentication matching is completed, sends authentication confirmation information to the mobile client for login;

The generation method of the first voiceprint key/second voiceprint key is:

Using two voiceprint feature extraction methods to extract the voiceprint feature vector sets of the first voice information/second voice information respectively;

calculating each voiceprint feature vector set of the first sound information/second sound information, and respectively obtaining a matching score vector set of a first preset value number of the first sound information/second sound information;

Constructing and fusing corresponding elements in each matching score vector set of the first sound information/second sound information into a binary group;

Use the binary group after the construction and fusion of the first sound information/the second sound information as the coordinate point, carry out the fitting of the curve analytical formula of the first sound information/the second sound information, and combine the first sound information/the second sound information The coefficients of the curve analytical formula are sorted and discretized as the voiceprint key.

Further, the two voiceprint feature extraction methods include any two of the following methods: Mel frequency cepstral coefficient MFCC, Mel linear spectral frequency MFSL, linear predictive cepstral coefficient LPCC, and residual phase.

Further, calculating each set of voiceprint feature vectors of the first sound information/second sound information, respectively obtaining a set of matching score vector sets of a first preset number of first sound information/second sound information includes: The vectors of each voiceprint feature vector set of a sound information/second sound information are input to the GMM-UBM or GMM model to obtain the matching score vectors of the first sound information/second sound information of the first preset number respectively gather.

Further, the authentication server presets the content of the second voice information for each authentication matching, and before completing the authentication matching of the voiceprint key for login, the method further includes: the authentication server determines that the content of the second voice information is preset If the content is specified, log in; otherwise, refuse to log in.

In another aspect, the present application also provides a voiceprint key generation device, including: an extraction unit, a calculation unit, a construction fusion unit, and a fitting generation unit; wherein,

The extraction unit is used to perform voiceprint feature extraction on the sound information by using two voiceprint feature extraction methods to obtain two voiceprint feature vector sets;

A calculation unit, configured to calculate each set of voiceprint feature vectors, and respectively obtain a set of matching score vectors of the first preset number;

Build a fusion unit, which is used to construct and fuse corresponding elements in each matching score vector set into a binary group;

The fitting generation unit is used to use the fused binary group as the coordinate point to perform curve fitting, and sort and discretize the coefficients of the curve analysis formula as the voiceprint key.

Further, the two voiceprint feature extraction methods include any two of the following methods: Mel frequency cepstral coefficient MFCC, Mel linear spectral frequency MFSL, linear predictive cepstral coefficient LPCC, and residual phase.

Further, the calculation unit is specifically configured to input the vectors of each voiceprint feature vector set into the GMM-UBM or the GMM model, so as to respectively obtain a set of matching score vectors of a first preset number.

In another aspect, the present application also provides a voiceprint-based key login system, including: a mobile client, an authentication server, and the above-mentioned voiceprint key generation device; wherein,

The mobile client is used to input the first voice information when logging into the data management system for the first time, and send it to the voiceprint key generation device to generate the first voiceprint key information; when it is not the first time to log into the data management system, input the second voice information , sent to the voiceprint key generating device to generate a second voiceprint key; receiving the login confirmation information from the authentication server to log in;

The authentication server is used to receive and store the first voiceprint key generated by the voiceprint key generating device when logging into the data management system for the first time; to receive the first voiceprint key generated by the voiceprint key generating device when logging into the data management system for the first time; Two voiceprint keys, the second voiceprint key is authenticated and matched with the first voiceprint key, and when the authentication matching is completed, an authentication confirmation message is sent to the mobile client;

The voiceprint key generation device is used to receive the first voice information/second voice information to generate the first voiceprint key/second voiceprint key, and send it to the authentication server service for storage/authentication matching.

Further, the authentication server also includes a preset unit and a content matching unit; wherein,

The preset unit is used to preset the content of the second voice information for each authentication match, and send it to the client for second voice input;

The content matching unit is used to log in when it is determined that the content of the second voice information is the preset content before the authentication matching of the voiceprint key is completed; otherwise, the log-in is refused.

The technical solution of the present application includes: when the mobile client logs in to the data management system for the first time, input the first voice information to generate the first voiceprint key and store it on the authentication server; when not logging in to the data management system for the first time, input the second voice information , to generate the second voiceprint key, the authentication server authenticates the second voiceprint key and the first voiceprint key, and when the authentication matching is completed, sends authentication confirmation information to the mobile client for login; the first The method for generating the voiceprint key/second voiceprint key is: using two voiceprint feature extraction methods to extract the voiceprint feature vector set of the first voice information/second voice information; Each set of voiceprint feature vectors of the sound information obtains a set of matching score vectors of the first preset number of first sound information/second sound information respectively; the matching score vectors of the first sound information/second sound information The corresponding elements in the set are constructed and fused into a binary group; the binary group after the construction and fusion of the first sound information/second sound information is used as a coordinate point to simulate the curve analysis formula of the first sound information/second sound information Combined, the coefficients of the curve analysis formula of the first sound information/second sound information are sorted and discretized as the voiceprint key. The invention ensures the safety of the mobile client accessing the data management system by adopting the voiceprint key method, and is beneficial to real-time supervision of the data management system.

Description of drawings

The accompanying drawings described here are used to provide a further understanding of the present invention and constitute a part of the application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention. In the attached picture:

Fig. 1 is the flow chart that the present invention realizes voiceprint key generation method;

Fig. 2 is a block diagram of the structure of the device for generating a voiceprint key according to the present invention;

Fig. 3 is a structural block diagram of the voiceprint-based key login system according to the present invention.

Detailed ways

Fig. 1 is the flow chart that realizes voiceprint key generation method of the present invention, as shown in Fig. 1, comprises:

Step 100, using two voiceprint feature extraction methods to perform voiceprint feature extraction on the voice information respectively, to obtain two voiceprint feature vector sets.

In this step, the collection of voiceprint feature vectors for extracting voice information includes: using any of the following two voiceprint feature extraction methods: Mel Frequency Cepstral Coefficient (MFCC), Mel Linear Spectral Frequency (MFSL), Linear Prediction Cepstral Coefficient ( LPCC), residual phase angle (residual phase).

It should be noted that the method for extracting voiceprint features in the present invention is mainly the spectral frequency extraction method, generally, as long as the spectral frequency extraction can be realized, it can be applied to the present invention.

Step 101. Calculate each set of voiceprint feature vectors, and respectively obtain a set of matching score vectors of a first preset number.

Here, according to conventional experiments, the number of matching score vector sets obtained is about 120-150, and the number of the first preset value is mainly related to the interference of sound information.

In this step, calculating each set of voiceprint feature vectors, and respectively obtaining a set of matching score vectors of the first preset number includes: inputting the vectors of each set of voiceprint feature vectors into a Gaussian mixture model-general background model (GMM - UBM) or (GMM) model, to respectively obtain a set of matching score vectors of the first preset value number. Here it refers to the vector input of a single set of voiceprint feature vectors to the Gaussian Mixture Model-Universal Background Model (GMM-UBM) or (GMM) model.

It should be noted that the GMM-UBM or GMM model is a calculation model well known to those skilled in the art.

Step 102, constructing and merging corresponding elements in each matching score vector set into a binary group.

Here, assuming the two sets of voiceprint feature vectors calculated in step 101, the obtained first preset value is 6 sets of matching score vectors respectively A=(a, b, c, d, e, f), B= (g, h, i, j, k, l), then the two-tuples to construct fusion are <a, g>, <b, h>, <c, i>, <d, j>, <e, k>, <f,l>;

Use the first value of the above two-tuple as the abscissa to correspond to the x value, and use the second value of the two-tuple as the y-coordinate to correspond to the y value to construct a two-dimensional coordinate system.

Step 103 , using the constructed and fused binary as the coordinate points, performing curve analysis equation fitting, and sorting and discretizing the coefficients of the curve analysis equation as the voiceprint key.

It should be noted that the analytical curve fitting can be fitted by the existing Lagrangian interpolation method, least squares curve fitting and other methods, sorted after the fitting process, and discretized by the median Get the voiceprint key.

Here, the fused pair is used as the coordinate point, and the curve analytical formula of the fitted polynomial is assumed to be:

p(x)＝a ₀ +a ₁ x+…+a _nk-1 x ^kn-1 +a _nk x ^kn

Then sort the coefficients of the curve analytical formula to get A={a ₀ ,a ₁ ,…a _kn }

The coefficient A is sorted by size and then the median am is taken out, then the discretized S includes:

Here, if S is to be applied, it is necessary to use the basic processing method of biometrics and organize it into a binary string form. Since this part is common knowledge in the field, it will not be repeated here.

A method for logging in based on a voiceprint key, comprising: when a mobile client logs in to a data management system for the first time, inputting first voice information to generate a first voiceprint key and storing it on an authentication server; when not logging in to the data management system for the first time , input the second voice information to generate the second voiceprint key, the authentication server will authenticate and match the second voiceprint key with the first voiceprint key, and send authentication confirmation information to the mobile client when the authentication matching is completed , to log in;

The method for generating the first voiceprint key/second voiceprint key is:

Two voiceprint feature extraction methods are used to extract the voiceprint feature vector sets of the first voice information/second voice information respectively. Here, the two voiceprint feature extraction methods include any two of the following methods: MFCC, MFSL, residual phase, and LPCC.

calculating each voiceprint feature vector set of the first sound information/second sound information, and respectively obtaining a matching score vector set of a first preset value number of the first sound information/second sound information;

Calculating each voiceprint feature vector set of the first sound information/second sound information, respectively obtaining the matching score vector set of the first preset value number of the first sound information/second sound information includes: by combining the first sound information The vectors of each set of voiceprint feature vectors of the second sound information are input to the GMM-UBM or GMM model to obtain a first preset number of matching score vector sets of the first sound information/second sound information.

The corresponding elements in each matching score vector set of the first sound information/the second sound information are constructed and fused into a binary group.

Use the binary group after the construction and fusion of the first sound information/the second sound information as the coordinate point, carry out the fitting of the curve analytical formula of the first sound information/the second sound information, and combine the first sound information/the second sound information The coefficients of the curve analytical formula are sorted and discretized as the voiceprint key.

The authentication server presets the content of the second voice information for each authentication matching, and before completing the authentication matching of the voiceprint key to log in, the method of the present invention further includes: the authentication server determines that the content of the second voice information is the preset content , log in; otherwise, refuse to log in.

It should be noted that by presetting the content of the second voice information for each login, when performing authentication and login, the staff must complete the voice information consistent with the preset content before the login can be realized. The determination of the content of the second voice information can be realized through the existing voice recognition method, and the content of the determined second voice information is matched with the preset content. system login.

Fig. 2 is a structural block diagram of the voiceprint key generation device of the present invention; as shown in Fig. 2, it includes: an extraction unit, a calculation unit, a construction fusion unit and a fitting generation unit; wherein,

The extraction unit is configured to use two voiceprint feature extraction methods to perform voiceprint feature extraction on the voice information to obtain two voiceprint feature vector sets.

Here, the two voiceprint feature extraction methods include any two of the following methods: MFCC, MFSL, residual phase, and LPCC.

The calculation unit is used to calculate each set of voiceprint feature vectors, and respectively obtain a set of matching score vectors with a first preset number of values.

The calculation unit is specifically configured to input the vectors of each voiceprint feature vector set into the GMM-UBM or the GMM model, so as to respectively obtain a first preset number of matching score vector sets.

A fusion unit is constructed for constructing and fusing corresponding elements in each matching score vector set into a binary group.

The fitting generation unit is used to use the fused binary group as the coordinate point to perform curve fitting, and sort and discretize the coefficients of the curve analysis formula as the voiceprint key.

Fig. 3 is a structural block diagram of the present invention based on voiceprint key login system, as shown in Fig. 3, including: a mobile client, an authentication server and a voiceprint key generation device; wherein,

The mobile client is used to input the first voice information when logging into the data management system for the first time, and send it to the voiceprint key generation device to generate the first voiceprint key information; when it is not the first time to log into the data management system, input the second voice information , sent to the voiceprint key generating device to generate a second voiceprint key; receiving the login confirmation information from the authentication server to log in;

The authentication server is used to receive and store the first voiceprint key generated by the voiceprint key generating device when logging into the data management system for the first time; to receive the first voiceprint key generated by the voiceprint key generating device when logging into the data management system for the first time; Two voiceprint keys, the second voiceprint key is authenticated and matched with the first voiceprint key, and when the authentication matching is completed, an authentication confirmation message is sent to the mobile client;

The voiceprint key generation device includes an extraction unit, a calculation unit, a construction fusion unit and a fitting generation unit; wherein,

The extraction unit is used to receive the first voice information, and use two voiceprint feature extraction methods to perform voiceprint feature extraction on the first voice information respectively, to obtain two sets of voiceprint feature vectors; to receive the second voice information, use two voiceprint feature extraction methods The fingerprint feature extraction method performs voiceprint feature extraction on the second voice information respectively, and obtains two voiceprint feature vector sets.

The two voiceprint feature extraction methods include any two of the following methods: MFCC, MFSL, residual phase, and LPCC.

The calculation unit is used to calculate the set of voiceprint feature vectors of the first sound information, respectively obtain the set of matching score vectors of the first preset value number of the first sound information; calculate the set of voiceprint feature vectors of the second sound information , respectively obtaining a set of matching score vector sets of the first preset number of second sound information.

The computing unit is specifically configured to input the vectors of each voiceprint feature vector set of the first voice information into the GMM-UBM or GMM model, so as to respectively obtain a first preset number of matching score vector sets of the first voice information ; By inputting the vectors of each voiceprint feature vector set of the second sound information into the GMM-UBM or GMM model, the matching score vector sets of the first preset number of second sound information are respectively obtained.

Constructing a fusion unit, which is used to construct and fuse corresponding elements in each matching score vector set of the first sound information into a binary group; construct and fuse corresponding elements in each matching score vector set of the second sound information into a binary group;

The fitting generation unit is used to use the first sound information to construct the fused binary group as the coordinate point, perform the fitting of the curve analysis formula, sort and discretize the coefficients of the curve analysis formula as the first voiceprint density key; use the second sound information to construct and fuse the binary group as the coordinate point, carry out the fitting of the curve analysis formula, and sort and discretize the coefficients of the curve analysis formula as the second voiceprint key.

The authentication server also includes a preset unit and a content matching unit; wherein,

The preset unit is used to preset the content of the second voice information for each authentication match, and send it to the client for second voice input;

The content matching unit is used to log in when it is determined that the content of the second voice information is the preset content before the authentication matching of the voiceprint key is completed; otherwise, the log-in is refused.

The method of the present invention will be described clearly and in detail below in conjunction with specific examples. The examples are only to illustrate the present invention more clearly and are not intended to limit the protection scope of the present invention.

Example 1

In this embodiment, the login application of the cloud platform data management system is used as the implementation environment.

In order to ensure the secure login of the cloud platform data management system on the mobile client, this embodiment describes the user voiceprint key authentication scheme based on the cloud platform data management system, including:

First, when the mobile client logs in to the cloud platform to access the database, a voiceprint key needs to be generated. The specific process is as follows:

The staff inputs user information and a voice message of a certain duration (I think it should be set, it should not be too short, here it is temporarily set to 10 seconds), and the mobile client submits the registered voice message to the cloud platform for registration through the Secure Sockets Layer (SSL) channel server.

It should be noted that the language information here may be voices with the same content, or languages with different content. Here, in order to prevent the voice information of the staff from being stolen, the content of the second voice information logged in each time can be preset, and the voice input is performed according to the content of the second voice information. When the input voice content is the same as the preset content, the System login can be realized.

After the cloud platform registration server receives the registered voice information, it transmits the user information and voice information to the voiceprint key generation device to generate the first voiceprint key. And the user information and the first voiceprint key are stored in the cloud platform storage server and in the remoteuser table in the registered user sample database.

When the staff logs in again, input the voice information through the mobile client, generate the second voiceprint key according to the method of generating the first voiceprint key, and combine the second voiceprint key with the first voiceprint key stored in the cloud platform storage server. The fingerprint key is matched, and when the match is passed, the system is logged in.

The method for generating the voiceprint key is the same as the method for generating the voiceprint key in the above embodiment.

Although the embodiments disclosed in the present application are as above, the content described is only the embodiments adopted to facilitate understanding of the present application, and is not intended to limit the present application. Anyone skilled in the field of this application can make any modifications and changes in the form and details of implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application must still be The scope defined by the appended claims shall prevail.

Claims (12)

1. A voiceprint key generation method, comprising:

performing voiceprint feature extraction on the voice information by using two voiceprint feature extraction methods to obtain two voiceprint feature vector sets;

calculating each voiceprint characteristic vector set, and respectively obtaining a first preset number of matching score vector sets;

constructing and fusing corresponding elements in each matching score vector set into a binary group;

and taking the constructed and fused binary group as a coordinate point, fitting a curve analytic expression, and sequencing and discretizing coefficients of the curve analytic expression to be used as a voiceprint key.

2. The voiceprint key generation method according to claim 1, wherein the two voiceprint feature extraction methods include any two of: mel-frequency cepstrum coefficients MFCC, Mel-linear spectral frequencies MFSL, linear prediction cepstrum coefficients LPCC, residual phase angles residual phase.

3. The method for generating a voiceprint key according to claim 1, wherein the calculating each voiceprint feature vector set and obtaining a first preset number of matching score vector sets respectively comprises: and respectively inputting the vectors of each voiceprint characteristic vector set into a Gaussian mixture model-general background model GMM-UBM or GMM to respectively obtain a first preset number of matching score vector sets.

4. A login method based on a voiceprint key is characterized by comprising the following steps:

when a mobile client logs in a data management system for the first time, inputting first sound information to generate a first voiceprint key and store the first voiceprint key on an authentication server; inputting second sound information to generate a second voiceprint key when the data management system is not logged in for the first time, carrying out authentication matching on the second voiceprint key and the first voiceprint key by the authentication server, and sending authentication determining information to the mobile client to log in when the authentication matching is completed;

the method for generating the first voiceprint key/the second voiceprint key comprises the following steps:

respectively extracting a voiceprint feature vector set of the first sound information/the second sound information by adopting two voiceprint feature extraction methods;

calculating each voiceprint characteristic vector set of the first sound information/the second sound information, and respectively obtaining a first preset number of matching score vector sets of the first sound information/the second sound information;

constructing and fusing corresponding elements in each matching score vector set of the first sound information/the second sound information into a binary group;

and taking the constructed and fused binary group of the first sound information/the second sound information as a coordinate point, fitting a curve analytic expression of the first sound information/the second sound information, sequencing and discretizing coefficients of the curve analytic expression of the first sound information/the second sound information, and taking the coefficients as a voiceprint key.

5. The method according to claim 4, wherein the two voiceprint feature extraction methods include any two of the following methods: mel-frequency cepstrum coefficients MFCC, Mel-linear spectral frequencies MFSL, linear prediction cepstrum coefficients LPCC, residual phase angles residual phase.

6. The method of claim 4, wherein the step of calculating each voiceprint feature vector set of the first voice message/the second voice message and obtaining the first predetermined number of matching score vector sets of the first voice message/the second voice message respectively comprises: and respectively obtaining a first preset number of matching score vector sets of the first sound information/the second sound information by inputting the vectors of the voiceprint feature vector sets of the first sound information/the second sound information into the GMM-UBM or the GMM model.

7. The method of claim 4, wherein the authentication server presets the content of the second voice information at each authentication matching, and before logging in after the authentication matching of the voiceprint key is completed, the method further comprises: the authentication server logs in when determining that the content of the second sound information is the preset content; otherwise, the login is rejected.

8. A voiceprint key generation apparatus, comprising: the device comprises an extraction unit, a calculation unit, a construction fusion unit and a fitting generation unit; wherein,

the extraction unit is used for respectively extracting the voiceprint characteristics of the voice information by utilizing two voiceprint characteristic extraction methods to obtain two voiceprint characteristic vector sets;

the computing unit is used for computing each voiceprint characteristic vector set and respectively obtaining a first preset number of matching score vector sets;

the construction fusion unit is used for constructing and fusing corresponding elements in each matching score vector set into a binary group;

and the fitting generation unit is used for fitting a curve analytic expression by taking the constructed and fused binary group as a coordinate point, and sequencing and discretizing coefficients of the curve analytic expression to be used as a voiceprint key.

9. The voiceprint key generation apparatus according to claim 8, wherein said two voiceprint feature extraction methods include any two of: mel-frequency cepstrum coefficients MFCC, Mel-linear spectral frequencies MFSL, linear prediction cepstrum coefficients LPCC, residual phase angles residual phase.

10. The apparatus according to claim 8, wherein the computing unit is specifically configured to obtain the first preset number of matching score vector sets by inputting a vector of each voiceprint feature vector set to the GMM-UBM or the GMM model.

11. A voiceprint key based login system, comprising: a mobile client, an authentication server and a voiceprint key generation apparatus of claims 8 to 10; wherein,

the mobile client is used for inputting first voice information when logging in the data management system for the first time and sending the first voice information to the voiceprint key generation device so as to generate first voiceprint key information; inputting second sound information when the system is not managed for the first time, and sending the second sound information to a voiceprint key generation device to generate a second voiceprint key; receiving login determination information of an authentication server for login;

an authentication server for receiving and storing a first voiceprint key generated by the voiceprint key generation device when logging in the data management system for the first time; receiving a second voiceprint key generated by a voiceprint key generation device when the data management system is not logged in for the first time, carrying out authentication matching on the second voiceprint key and the first voiceprint key, and sending authentication confirmation information to the mobile client when the authentication matching is completed;

and the voiceprint key generating device is used for receiving the first voice information/the second voice information to generate a first voiceprint key/a second voiceprint key and sending the first voiceprint key/the second voiceprint key to the authentication server service for storage/authentication matching.

12. The system according to claim 11, wherein the authentication server further comprises a preset unit and a content matching unit; wherein,

the preset unit is used for presetting the content of the second sound information when the authentication is matched every time and sending the second sound information to the client for second sound input;

the content matching unit is used for logging in when the content of the second sound information is determined to be the preset content before the voiceprint key is authenticated, matched and logged in; otherwise, the login is rejected.