CN102238190B - Identity authentication method and system - Google Patents

Identity authentication method and system Download PDF

Info

Publication number
CN102238190B
CN102238190B CN2011102180452A CN201110218045A CN102238190B CN 102238190 B CN102238190 B CN 102238190B CN 2011102180452 A CN2011102180452 A CN 2011102180452A CN 201110218045 A CN201110218045 A CN 201110218045A CN 102238190 B CN102238190 B CN 102238190B
Authority
CN
China
Prior art keywords
vocal print
model
characteristic sequence
likelihood score
print characteristic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN2011102180452A
Other languages
Chinese (zh)
Other versions
CN102238190A (en
Inventor
潘逸倩
胡国平
何婷婷
魏思
胡郁
王智国
刘庆峰
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
iFlytek Co Ltd
Original Assignee
iFlytek Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by iFlytek Co Ltd filed Critical iFlytek Co Ltd
Priority to CN2011102180452A priority Critical patent/CN102238190B/en
Publication of CN102238190A publication Critical patent/CN102238190A/en
Application granted granted Critical
Publication of CN102238190B publication Critical patent/CN102238190B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention discloses an identity authentication method and an identity authentication system. The method comprises the following steps of: in the login of a user, receiving a continuous voice signal recorded by the current login user; extracting a voiceprint characteristic sequence from the continuous voice signal; computing likelihood between the voiceprint characteristic sequence and a background model; computing the likelihood between the voiceprint characteristic sequence and a speaker model of the current login user, wherein the speaker model is a polyhybrid Gaussian model constructed according to the repetition times and frame number of the registration voice signals recorded in the login of the current login user; computing a likelihood ratio according to the likelihood between the voiceprint characteristic sequence and the speaker model and the likelihood between the voiceprint characteristic sequence and the background model; and if the likelihood ratio is greater than a preset threshold value, determining the current login user is an effectively authenticated user, otherwise determining the current login user is an unauthenticated user. By the method and the system, the voiceprint-password-based identity authentication accuracy can be improved.

Description

Identity identifying method and system
Technical field
The present invention relates to identity identification technical field, particularly a kind of identity identifying method and system.
Background technology
Application on Voiceprint Recognition (Voiceprint Recognition, VPR), also referred to as Speaker Identification, has two classes, i.e. speaker's identification and speaker verification.The former,, in order to judge that certain section voice are which in some people is said, is " multiselect one " problem; And the latter is " differentiating one to one " problem in order to confirm that whether certain section voice are that the someone of appointment is said.Different tasks can be used different sound groove recognition technology in es with application.
Voiceprint refers to according to the voice signal collected confirms speaker ' s identity, belongs to the discrimination of " one to one ".The voiceprint authentication system of main flow has adopted the framework based on hypothesis testing now, by calculate respectively the vocal print signal with respect to the likelihood score of speaker model and background model and relatively they likelihood ratio and in advance rule of thumb the threshold size of setting confirm.Obviously the accuracy of background model and speaker model will directly have influence on the voiceprint effect, and under the statistical model setting driven at based on data, more modelling effect is better for amount of training data.
The vocal print cipher authentication is a kind of speaker ' s identity authentication method of text-dependent.The method requires the user speech input to determine cryptogram, and confirms accordingly speaker ' s identity.In this application, user's registration and authentication all adopt the phonetic entry of determining cryptogram, thereby its vocal print is often more consistent, can obtain accordingly than the speaker verification of text-independent and better authenticate effect.
Now the vocal print cipher authentication system the most the technology path of main flow be the GMM-UBM algorithm, adopted respectively mixed Gaussian (Gaussian Mixture Model, GMM) modeling background model (Universal Background Model, UBM) and speaker model.The UBM model is for describing the general character of speaker's vocal print.Because each speaker's vocal print always has specificity separately, the corresponding UBM model based on many speakers training data needs complicated model structure to meet the matching requirement of distribution separate data.The UBM model is selected the GMM model of 1024 even larger Gaussages usually at present.
By system, when the user registers, according to the registration voice, online training obtains speaker model.Because registration is often limited with speech samples, directly train accordingly complex model because Sparse easily causes the problems such as model is accurate not.For this reason, in the prior art, normally take background model as initial model by various adaptive approachs according to a small amount of speaker's data adjustment model partial parameters, as at present commonly used based on maximum a posteriori probability (Maximum A Posterior, MAP) adaptive algorithms etc., should be current speaker's individual character by user's vocal print general character is adaptive.
Form man-to-man corresponding relation between each Gauss of speaker's mixed Gauss model and common background Gauss model under the adaptive updates algorithm, therefore, make the speaker model parameter too much, easily cause following problem in the log-on data amount in less vocal print cipher authentication system:
1. model redundancy: in the vocal print cipher authentication system, speaker model is to be obtained by several sample datas training that repeat all over the registration speech cipher.Very few sample data causes adaptive algorithm can only upgrade part Gauss in initial back-ground model, and has much all retained and the similar Gaussian component of background model.The existence of redundancy model parameter easily causes the increase of storage and computing pressure, and then the efficiency of impact decoding.
2. the model training amount is larger: in adaptive algorithm, and each Gauss's of 1024 even larger Gaussages of needs calculating initial back-ground model sample statistic, and its parameter is upgraded.
3. in adaptive algorithm, because the variance revaluation of speaker model is comparatively difficult, thereby often directly adopt the variance of background model.Because background model is based on the model of the simulation vocal print general character that many speakers training data obtains, its model probability distribution variance is often larger.And the characteristics of the specific vocal print of speaker of the variance of speaker model simulation have specificity.Directly by the background model variance, can not embody well the speaker model characteristics, reduce the differentiation between different speaker models, thereby affected recognition accuracy.
Summary of the invention
The embodiment of the present invention provides a kind of identity identifying method and system, to improve the accuracy rate of based on the vocal print password, carrying out authentication.
The embodiment of the present invention provides a kind of identity identifying method on the one hand, comprising:
When the user logins, receive the continuous speech signal of current login user typing;
Extract the vocal print characteristic sequence in described continuous speech signal, described vocal print characteristic sequence comprises one group of vocal print feature;
Calculate the likelihood score of described vocal print characteristic sequence and background model;
Calculate the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model that frame number builds;
According to the likelihood score of described vocal print characteristic sequence and speaker model and the likelihood score of described vocal print characteristic sequence and background model, calculate likelihood ratio;
If described likelihood ratio is greater than the threshold value of setting, determines that described current login user is effective authenticated user, otherwise determine that described current login user is non-authenticated user.
The embodiment of the present invention provides a kind of identity authorization system on the other hand, comprising:
The voice signal receiving element, for when the user logins, receive the continuous speech signal of current login user typing;
Extraction unit, for extracting the vocal print characteristic sequence of described continuous speech signal, described vocal print characteristic sequence comprises one group of vocal print feature;
The first computing unit, for calculating the likelihood score of described vocal print characteristic sequence and background model;
The second computing unit, for calculating the likelihood score of speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model of frame number structure;
The 3rd computing unit, for the likelihood score of the likelihood score according to described vocal print characteristic sequence and speaker model and described vocal print characteristic sequence and background model, calculate likelihood ratio;
Judging unit, while for the likelihood ratio calculated at described the 3rd computing unit, being greater than the threshold value of setting, determine that described current login user is effective authenticated user, otherwise determine that described current login user is non-authenticated user.
The identity identifying method that the embodiment of the present invention provides and system, according to the vocal print characteristic sequence in the continuous speech signal of current login user typing, calculate respectively the speaker model of vocal print characteristic sequence and current login user and the likelihood score of background model, then calculate likelihood ratio, according to the likelihood ratio obtained, determine whether current login user is effective authenticated user.Due in this scheme, the speaker model used is the polyhybird Gauss model that the voice signal of typing builds during according to the registration of current login user, thereby can simulate the characteristics that described user says the difference pronunciation variation of same voice signal (being password) existence, improve the accuracy rate of carrying out authentication based on the vocal print password.
The accompanying drawing explanation
In order to be illustrated more clearly in technical scheme of the invention process, below will the accompanying drawing of required use in embodiment be briefly described, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skills, under the prerequisite of not paying creative work, can also obtain according to these accompanying drawings other accompanying drawing.
Fig. 1 is the flow chart of embodiment of the present invention identity identifying method;
Fig. 2 is a kind of flow chart of background model parameters training process in the embodiment of the present invention;
Fig. 3 is that traditional adaptive algorithm of utilizing builds the flow chart of speaker model;
Fig. 4 builds the flow chart of speaker model in the embodiment of the present invention;
Fig. 5 is a kind of structural representation of embodiment of the present invention identity authorization system;
Fig. 6 is the another kind of structural representation of embodiment of the present invention identity authorization system.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is only the present invention's part embodiment, rather than whole embodiment.Embodiment based in the present invention, those of ordinary skills, not making under the creative work prerequisite the every other embodiment obtained, belong to the scope of protection of the invention.
As shown in Figure 1, be the flow chart of embodiment of the present invention identity identifying method, comprise the following steps:
Step 101, when the user logins, receive the continuous speech signal of current login user typing.
Step 102, extract the vocal print characteristic sequence in described continuous speech signal.
This vocal print characteristic sequence comprises one group of vocal print feature, can effectively distinguish different speakers, and same speaker's variation is kept relative stability.
Described vocal print feature mainly contains: spectrum envelop parameter phonetic feature, pitch contour, formant frequency bandwidth feature, linear predictor coefficient, cepstrum coefficient etc.Consider the quantification property of above-mentioned vocal print feature, the quantity of training sample and the problems such as evaluation of systematic function, can select MFCC (Mel Frequency Cepstrum Coefficient, the Mel frequency cepstral coefficient) feature, every frame speech data that the long 25ms frame of window is moved to 10ms is done short-time analysis and is obtained MFCC parameter and single order second differnce thereof, amounts to 39 dimensions.Like this, every voice signal can be quantified as one 39 dimension vocal print feature vector sequence X.
Step 103, calculate the likelihood score of described vocal print characteristic sequence and background model.
The vocal print feature vector sequence X that frame number is T corresponding to the likelihood score of background model (UBM) is:
p ( X | UBM ) = 1 T Σ t = 1 T Σ m = 1 M c m N ( X t ; μ m , Σ m ) - - - ( 1 )
Wherein, c mbe m Gauss's weight coefficient, meet μ mand ∑ mrespectively m Gauss's average and variance.Wherein N (.) meets normal distribution, for calculating t vocal print characteristic vector X constantly tlikelihood score on single Gaussian component:
N ( X t ; μ m , Σ m ) = 1 ( 2 π ) n | Σ m | e - 1 2 ( X t - μ m ) ′ Σ m - 1 ( X t - μ m ) - - - ( 2 )
Step 104, calculate the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model that frame number builds.
Because speaker model is the polyhybird Gauss model that the voice signal of typing builds during according to described current login user registration, therefore, in this step, while calculating the likelihood score of speaker model of described vocal print characteristic sequence and described current login user, need to calculate respectively the likelihood score of each vocal print feature and each mixed Gauss model in described vocal print characteristic sequence; Then determine the likelihood score of the speaker model of described vocal print feature and described current login user according to all likelihood scores that calculate.Specifically multiple implementation can be arranged, such as:
1. first calculate respectively the likelihood score of described vocal print characteristic sequence and each mixed Gauss model, and then determine the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user according to result of calculation.
In this mode, can calculate respectively in described vocal print characteristic sequence the likelihood score of each mixed Gauss model in each vocal print feature and described polyhybird Gauss model; Select the likelihood score of the time average of the likelihood score summation that in described vocal print characteristic sequence, mixed Gauss model of one group of vocal print feature correspondence calculates as described vocal print characteristic sequence and this mixed Gauss model.
And, after the likelihood score that obtains described vocal print characteristic sequence and each mixed Gauss model, can select one of them maximum or the average likelihood score as the speaker model of described vocal print characteristic sequence and described current login user.
2. first calculate respectively in described vocal print characteristic sequence each vocal print feature with respect to the likelihood score of described polyhybird Gauss model, and then determine the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user according to result of calculation.
In this mode, can calculate respectively in described vocal print characteristic sequence the likelihood score of each mixed Gauss model in each vocal print feature and described polyhybird Gauss model; Select in described vocal print characteristic sequence maximum in likelihood score that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model the calculates likelihood score as this vocal print feature and described polyhybird Gauss model; Perhaps, select in described vocal print characteristic sequence the likelihood score of the mean value of all likelihood scores that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model calculates as this vocal print feature and described polyhybird Gauss model.
And in obtaining described vocal print feature after the likelihood score of each vocal print feature and polyhybird Gauss model, select the summation time average of all vocal print feature likelihood scores of vocal print characteristic sequence as the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user.
Certainly, other selection modes can also be arranged, such as all likelihood scores to calculating are weighted on average etc., this embodiment of the present invention not done to restriction.
Step 105, according to the likelihood score of described vocal print characteristic sequence and speaker model and the likelihood score of described vocal print characteristic sequence and background model, calculate likelihood ratio.
Likelihood ratio is: p = p ( X | U ) p ( X | UBM ) - - - ( 3 )
Wherein, the likelihood score that p (X|U) is described vocal print feature and speaker model, the likelihood score that p (X|UBM) is described vocal print feature and background model.
Step 106, judge whether described likelihood ratio is greater than the threshold value of setting, if so, performs step 107; Otherwise, perform step 108.
Above-mentioned threshold value can be preset by system, in general, this threshold value is larger, the sensitivity of system is higher, require user's pronunciation of the voice signal (password) of typing during as far as possible according to registration when login, otherwise the sensitivity of system is lower, there is certain variation in pronunciation when while allowing the user to login, the pronunciation of the voice signal of typing is with registration.
Step 107, determine that described current login user is effective authenticated user.
Step 108, determine that described current login user is non-authenticated user.
It should be noted that, in order to improve the robustness of system, before above-mentioned steps 101 and step 102, can also carry out noise reduction process to described continuous speech signal, such as, at first by short-time energy and short-time zero-crossing rate analysis to voice signal, continuous voice signal is divided into to independently voice snippet and non-voice segment.Then reduce the interference of channel noise and background noise by the front end noise reduction process, improve the voice signal to noise ratio, for follow-up system, process clean signal is provided.
The existing relative stability of user's vocal print feature, have again variability.Easily be subject on the one hand the impact of health, age, mood etc., easily be subject on the other hand the interference of external environment noise and voice collecting channel, so speaker model needs the different vocal prints that can distinguish preferably same speaker to change.In embodiments of the present invention, speaker model is the polyhybird Gauss model that the voice signal of typing builds during according to described current login user registration, when the Gaussage of mixed Gauss model number and each mixed Gauss model is registered to the user, the frame number of the number of repetition of the voice signal of typing and this voice signal is relevant, thereby the characteristics that can utilize a plurality of mixed Gauss model analog subscribers to change the difference pronunciation of saying same password (being above-mentioned voice signal) existence, improved the accuracy rate of carrying out authentication based on the vocal print password.
In embodiments of the present invention, background model is for describing the general character of speaker's vocal print, this background model need to build in advance, specifically can adopt modes more of the prior art, such as, the mixed Gauss model simulation background model of employing 1024 or larger Gaussage, its model parameter training process as shown in Figure 2.
Step 201 is extracted respectively the vocal print feature from many speakers training utterance signal, and each vocal print feature is as a characteristic vector.
Step 202, utilize clustering algorithm to carry out cluster to above-mentioned characteristic vector, obtains K Gauss's initialization average, and K is the mixed Gauss model number set in advance.
Such as, can adopt traditional LBG (Linde, Buzo, Gray) clustering algorithm, approach optimum regeneration code book by trained vector collection and certain iterative algorithm.
Step 203, utilize EM (Expectation Maximization) algorithm iteration to upgrade above-mentioned average, variance and weight coefficient corresponding to each Gauss, obtains background model.
Concrete iteration renewal process is same as the prior art, at this, is not described in detail.
Certainly, can also adopt other modes to build background model, this embodiment of the present invention is not done to restriction.
In embodiments of the present invention, needing to distinguish the user is in login mode or registration mode, if login mode, need to carry out the authentication based on the vocal print password to this user according to the flow process shown in Fig. 1, if registration mode, need to receive the registration voice signal of described user's typing, and build described user's speaker model according to described registration voice signal.
In the embodiment of the present invention, the building process of the building process of speaker model and traditional speaker model is diverse, for this point is described better, below at first the building process of traditional speaker model is done to simple declaration.
The building process of traditional speaker model is to take background model as initial model, by adaptive approach adjustment model partial parameters, as the adaptive algorithm based on maximum a posteriori probability at present commonly used etc.Adaptive algorithm should be current speaker's individual character according to a small amount of speaker's data by user's vocal print general character is adaptive, and it specifically trains flow process as shown in Figure 3, comprises the following steps:
Step 301 is extracted the vocal print feature from the registration voice signal of user's typing.
Step 302, utilize the average μ of described vocal print feature adaptive updates background model mixed Gaussian m.
Particularly, new Gaussian mean μ mbe calculated as the weighted average of sample statistic and original Gaussian mean, that is:
μ m ^ = Σ t = 1 T γ m ( x t ) x t + τ μ m Σ t = 1 T γ m ( x t ) + τ - - - ( 4 )
Wherein, x tmean t frame vocal print feature, γ m(x t) meaning that t frame vocal print feature falls within m Gauss's probability, τ is forgetting factor, the update intensity for the historical average of balance and sample to new average.In general, the τ value is larger, and new average is restricted by original average mainly.And if the τ value is less, new average is mainly determined by sample statistic, has more embodied the characteristics that new samples distributes.
Step 303, copy the speaker model variance of background model variance as described user.
Step 304, generate described user's speaker model.
In embodiments of the present invention, need to when the user registers, receive the registration voice signal of described user's typing, and build described user's speaker model according to described registration voice signal.This speaker model consists of a plurality of mixed Gauss models, the characteristics that the difference pronunciation of saying same password and existing changed with the simulation speaker, and, illustrate that in human model, each mixed Gauss model is trained separately variance, in the solution conventional method, directly to copy the background model variance, cause variance excessive, do not meet the problem of practical application.
As shown in Figure 4, be to build the flow chart of speaker model in the embodiment of the present invention, comprise the following steps:
Step 401, the registration voice signal that user recording is entered saves as a conventional sequence.
Suppose that the user registers the same password content N of input (such as N=2,3 etc.) inferior, obtain N independently conventional sequence.
Step 402 is extracted the vocal print feature from the conventional sequence obtained.
The step 102 of detailed process and front is similar, at this, is not described in detail.
Step 403, all mixed Gauss models of the speaker model of determining described user according to number of repetition and the frame number of described registration voice signal.
In the vocal print cipher application, the user inputs unified content of text and uses as password.Such as, the mixed Gauss model number that can set described user's speaker model equals the number of repetition of described registration voice signal, and set the frame number that equals the registration voice signal that described mixed Gauss model is corresponding corresponding to the Gaussage of each mixed Gauss model, specifically can be expressed as:
p ( O | M k ) = Σ m = 1 T ( k ) c m k N ( O ; μ m k , Σ m k ) - - - ( 5 )
Wherein, T (k) is mixed Gauss model M kgaussage, be equal to the frame number of k the speech samples that model is corresponding.And
Figure BDA0000080250750000092
respectively mixed Gauss model M kweight coefficient, average and the variance of m Gaussian component.
Certainly, the embodiment of the present invention does not limit the topological mode of above-mentioned speaker model, the Gaussage of its mixed Gauss model number and each mixed Gauss model can be not and number of repetition and the complete correspondent equal of frame number of described voice signal yet, also can choose the number of repetition that the mixed Gauss model number is less than described registration voice signal by adopting clustering algorithm, equally, the Gaussage of each mixed Gauss model also can be less than the frame number of described registration voice signal.
Step 404, estimate the Gaussian mean parameter of all mixed Gauss models according to the vocal print feature of extracting.
In embodiments of the present invention, determine the Gaussian mean parameter of its corresponding mixed Gauss model according to single training sample.Particularly, each Gaussian mean vector that can mixed Gauss model is set to the Characteristic Vectors value of sample,
Figure BDA0000080250750000101
wherein the average that means m Gauss of k mixed model, and
Figure BDA0000080250750000103
the vocal print characteristic vector that means the m frame voice of k voice signal.
Step 405, estimate Gauss's variance parameter of all mixed Gauss models according to the vocal print feature of extracting.
Can suppose that in speaker model, a plurality of Gausses of each mixed Gaussian have for the unified matrix of the overall situation, to realize the variance revaluation problem on less data.Under this hypothesis,
Figure BDA0000080250750000104
(the covariance matrix of all Gaussian component of k mixed Gauss model has identical matrix numerical value).Particularly, to given sample vocal print characteristic sequence O k, according to
Figure BDA0000080250750000105
the i.e. statistical information of all residue sample vocal print characteristic sequences, revaluation mixed Gauss model M kvariance, be calculated as follows:
Σ k = Σ n ≠ k Σ i = 1 T ( n ) Σ m = 1 T ( k ) ( γ m k ( O i n ) ( O i n - μ m k ) ( O i n - μ m k ) T ) Σ n ≠ k Σ i = 1 T ( n ) Σ m = 1 T ( k ) γ m k ( O i n ) - - - ( 6 )
Wherein,
Figure BDA0000080250750000107
i the speech frame (being sample) that means n sentence log-in password (registering voice signal),
Figure BDA0000080250750000108
m the Gaussian mean that means k mixed Gauss model,
Figure BDA0000080250750000109
mean sample
Figure BDA00000802507500001010
dropping on average is
Figure BDA00000802507500001011
gauss on probability.
Like this, to each independent mixed Gaussian M of speaker model k, can utilize non-O ksample data obtain corresponding variance parameter.If register voice signal as the N sentence, obtain N different variance matrix.
Especially, can suppose that this variance matrix is that diagonal matrix is further to reduce Sparse Problem,
Figure BDA00000802507500001012
a plurality of Gausses' the variance that can also further consider in addition a plurality of mixed Gauss models of speaker model has the unified diagonal matrix of the overall situation, to solve better the model variance revaluation problem in the Sparse situation.Under this hypothesis,
Step 406, estimate Gauss's weight coefficient parameter of all mixed Gauss models.
The Gaussian mean of considering mixed Gauss model in the present embodiment directly determined by the sample vector, thereby each Gauss exists with 1 probability on sample, and probability of occurrence is identical.The weight coefficient equalization of each Gauss in mixed model can be set for this reason, that is: in the present embodiment
c m k = c k = 1 T ( k ) - - - ( 7 )
Utilize flow process shown in above-mentioned Fig. 4, can be according to sentence number and the long topological structure that the number of the mixed Gauss model in speaker model is set and determines model of sentence of registration voice, reasonable setting by the Gaussian mean to all mixed Gauss models, variance and weight coefficient, effectively solved the training problem of the Sparse existed in the system of tradition based on the vocal print cipher authentication, improve the differentiation between mixed Gauss model, and then can improve the accuracy rate of authentication.And the mixed Gauss model of use is less more effective, with respect to prior art, greatly improved the required memory pressure of arithmetic speed and storage data.
Correspondingly, the embodiment of the present invention also provides a kind of identity authorization system, as shown in Figure 5, is a kind of structural representation of embodiment of the present invention identity authorization system.
In this embodiment, described system comprises:
Voice signal receiving element 501, for when the user logins, receive the continuous speech signal of current login user typing;
Extraction unit 502, for extracting the vocal print characteristic sequence of described continuous speech signal;
The first computing unit 503, for calculating the likelihood score of described vocal print characteristic sequence and background model;
The second computing unit 504, for calculating the likelihood score of speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model of frame number structure;
The 3rd computing unit 505, for the likelihood score of the likelihood score according to described vocal print characteristic sequence and speaker model and described vocal print characteristic sequence and background model, calculate likelihood ratio;
Judging unit 506, while for the likelihood ratio calculated at described the 3rd computing unit 505, being greater than the threshold value of setting, determine that described current login user is effective authenticated user, otherwise determine that described current login user is non-authenticated user.
Above-mentioned this vocal print characteristic sequence comprises one group of vocal print feature, can effectively distinguish different speakers, and same speaker's variation is kept relative stability.
Such as, the vocal print feature that extraction unit 502 can extract mainly contains: spectrum envelop parameter phonetic feature, pitch contour, formant frequency bandwidth feature, linear predictor coefficient, cepstrum coefficient etc.Consider the quantification property of above-mentioned vocal print feature, the quantity of training sample and the problems such as evaluation of systematic function, can select MFCC (Mel Frequency Cepstrum Coefficient, the Mel frequency cepstral coefficient) feature, every frame speech data that the long 25ms frame of window is moved to 10ms is done short-time analysis and is obtained MFCC parameter and single order second differnce thereof, amounts to 39 dimensions.Like this, every voice signal can be quantified as one 39 dimension vocal print characteristic sequence X.
The above-mentioned background model can be that system builds in advance and is written into when initialization, and the concrete building process embodiment of the present invention of background model is not done restriction.
Above-mentioned speaker model is the polyhybird Gauss model that the voice signal of typing builds during according to described current login user registration, and correspondingly, in embodiments of the present invention, above-mentioned the second computing unit 504 can have multiple implementation, such as:
In one implementation, described the second computing unit 504 comprises: the first computation subunit and first is determined subelement.Wherein:
Described the first computation subunit, for calculating respectively the likelihood score of described vocal print characteristic sequence and each mixed Gauss model;
Described first determines subelement, determines the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user for the result of calculation according to described the first computation subunit.
Above-mentioned the first computation subunit can comprise: the first computing module and first is selected module, wherein:
Described the first computing module, for calculating respectively the likelihood score of each mixed Gauss model in each vocal print feature of described vocal print characteristic sequence and described polyhybird Gauss model;
Described first selects module, the likelihood score for the time average of the likelihood score summation of selecting the corresponding mixed Gauss model of one group of vocal print feature of described vocal print characteristic sequence to calculate as described vocal print characteristic sequence and this mixed Gauss model.
Correspondingly, above-mentioned first determines that subelement also can have multiple implementation, such as, obtain the likelihood score of described vocal print characteristic sequence and each mixed Gauss model in the first computation subunit after, first determines that subelement can select one of them maximum or the average likelihood score as the speaker model of described vocal print characteristic sequence and described current login user.
In another kind of implementation, described the second computing unit 504 comprises: the second computation subunit and second is determined subelement.Wherein:
Described the second computation subunit, for calculating respectively the likelihood score of each vocal print feature of described vocal print characteristic sequence with respect to described polyhybird Gauss model;
Described the second chooser unit, determine the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user for the result of calculation according to described the second computation subunit.
Above-mentioned the second computation subunit can comprise: the second computing module and second is selected module, wherein:
Described the second computing module, for calculating respectively the likelihood score of each mixed Gauss model in each vocal print feature of described vocal print characteristic sequence and described polyhybird Gauss model;
Described second selects module, for selecting maximum in likelihood score that in the corresponding described polyhybird Gauss model of vocal print feature of described vocal print characteristic sequence, each mixed Gauss model the calculates likelihood score as this vocal print feature and described polyhybird Gauss model; Perhaps select in described vocal print characteristic sequence the likelihood score of the mean value of all likelihood scores that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model calculates as this vocal print feature and described polyhybird Gauss model.
Correspondingly, above-mentioned second determines that subelement also can have multiple implementation, such as, in the second computation subunit, obtain in described vocal print characteristic sequence after the likelihood score of each vocal print feature with respect to described polyhybird Gauss model, second determines that subelement can select the likelihood score of the speaker model that each vocal print feature in described vocal print characteristic sequence is described vocal print characteristic sequence and described current login user with respect to the time average of the likelihood score of described polyhybird Gauss model.
Certainly, the second computing unit 504 can also adopt other modes to realize, this embodiment of the present invention is not done to restriction.
The concrete computational process of above-mentioned the first computing unit 503, the second computing unit 504 and the 3rd computing unit 505 can, with reference to the description in the embodiment of the present invention identity identifying method of front, not repeat them here.
In embodiments of the present invention, speaker model is the polyhybird Gauss model that the voice signal of typing builds during according to described current login user registration, when the Gaussage of mixed Gauss model number and each mixed Gauss model is registered to the user, the frame number of the number of repetition of the voice signal of typing and this voice signal is relevant, thereby the characteristics that can utilize a plurality of mixed Gauss model analog subscribers to change the difference pronunciation of saying same password (being above-mentioned voice signal) existence, improved the accuracy rate of carrying out authentication based on the vocal print password.
As shown in Figure 6, be the another kind of structural representation of embodiment of the present invention identity authorization system.
From embodiment illustrated in fig. 5 different, in this embodiment, described voice signal receiving element 501, also for when the user registers, receives the registration voice signal of described user's typing.
In addition, in this system, also further comprise: model construction unit 601, for build described user's speaker model according to described registration voice signal, this model construction unit 601 comprises:
Feature extraction subelement 611, for extracting the vocal print feature from described registration voice signal;
Topological structure is determined subelement 612, determines all mixed Gauss models of described user's speaker model for the number of repetition according to described registration voice signal and frame number;
Such as, the mixed Gauss model number that can set described user's speaker model is less than or equal to the number of repetition of described registration voice signal; Setting is less than or equal to the frame number of described registration voice signal corresponding to the Gaussage of each mixed Gauss model;
The first estimator unit 613, estimate that for the vocal print feature of utilizing feature extraction subelement 611 to extract described topological structure determines the Gaussian mean parameter of all mixed Gauss models that subelement 612 is determined;
The second estimator unit 614, estimate that for the vocal print feature of utilizing feature extraction subelement 611 to extract described topological structure determines Gauss's variance parameter of all mixed Gauss models that subelement 612 is determined.
Above-mentioned each estimator unit can, with reference to the description of front, not repeat them here the method for estimation of the relevant parameter in mixed Gauss model.
The identity authorization system of the embodiment of the present invention, can be according to sentence number and the long topological structure that the number of the mixed Gauss model in speaker model is set and determines model of sentence of registration voice, reasonable setting by the Gaussian mean to all mixed Gauss models, variance and weight coefficient, effectively solved the training problem of the Sparse existed in the system of tradition based on the vocal print cipher authentication, improve the differentiation between mixed Gauss model, and then can improve the accuracy rate of authentication.And the mixed Gauss model of use is less more effective, with respect to prior art, greatly improved the required memory pressure of arithmetic speed and storage data.
Each embodiment in this specification all adopts the mode of going forward one by one to describe, and between each embodiment, identical similar part is mutually referring to getting final product, and each embodiment stresses is the difference with other embodiment.Especially, for system embodiment, due to it, substantially similar in appearance to embodiment of the method, so describe fairly simplely, relevant part gets final product referring to the part explanation of embodiment of the method.System embodiment described above is only schematically, and wherein said unit and module as the separating component explanation can or can not be also physically to separate.In addition, the purpose that can also select according to the actual needs some or all of unit wherein and module to realize the present embodiment scheme.Those of ordinary skills in the situation that do not pay creative work, can understand and implement.
Above disclosed be only the preferred embodiment of the present invention; but the present invention is not limited thereto; any those skilled in the art can think there is no a creationary variation, and some improvements and modifications of doing without departing from the principles of the present invention, all should drop in protection scope of the present invention.

Claims (18)

1. an identity identifying method, is characterized in that, comprising:
When the user logins, receive the continuous speech signal of current login user typing;
Extract the vocal print characteristic sequence in described continuous speech signal, described vocal print characteristic sequence comprises one group of vocal print feature;
Calculate the likelihood score of described vocal print characteristic sequence and background model;
Calculate the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model that frame number builds;
According to the likelihood score p (X|U) of described vocal print characteristic sequence and speaker model and the likelihood score p (X|UBM) of described vocal print characteristic sequence and background model, calculate likelihood ratio
Figure FDA0000384388700000011
If described likelihood ratio is greater than the threshold value of setting, determines that described current login user is effective authenticated user, otherwise determine that described current login user is non-authenticated user.
2. the method for claim 1, is characterized in that, the likelihood score of the speaker model of the described vocal print characteristic sequence of described calculating and described current login user comprises:
Calculate respectively the likelihood score of described vocal print characteristic sequence and each mixed Gauss model;
Determine the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user according to result of calculation.
3. method as claimed in claim 2, is characterized in that, the described likelihood score that calculates respectively described vocal print characteristic sequence and each mixed Gauss model comprises:
Calculate respectively in described vocal print characteristic sequence the likelihood score of each mixed Gauss model in each vocal print feature and described polyhybird Gauss model;
Select the likelihood score of the time average of the likelihood score summation that in described vocal print characteristic sequence, mixed Gauss model of one group of vocal print feature correspondence calculates as described vocal print characteristic sequence and this mixed Gauss model.
4. method as claimed in claim 2, is characterized in that, describedly according to result of calculation, determines that the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user comprises:
The mean value of the likelihood score of selecting the corresponding all mixed Gauss models of described vocal print characteristic sequence to calculate is as the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user; Perhaps
The maximum of the likelihood score of selecting the corresponding all mixed Gauss models of described vocal print characteristic sequence to calculate is as the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user.
5. the method for claim 1, is characterized in that, the likelihood score of the speaker model of the described vocal print characteristic sequence of described calculating and described current login user comprises:
Calculate respectively in described vocal print characteristic sequence each vocal print feature with respect to the likelihood score of described polyhybird Gauss model;
Determine the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user according to result of calculation.
6. method as claimed in claim 5, is characterized in that, describedly calculates respectively each vocal print feature in described vocal print characteristic sequence and comprise with respect to the likelihood score of described polyhybird Gauss model:
Calculate respectively in described vocal print characteristic sequence the likelihood score of each mixed Gauss model in each vocal print feature and described polyhybird Gauss model;
Select in described vocal print characteristic sequence maximum in likelihood score that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model the calculates likelihood score as this vocal print feature and described polyhybird Gauss model; Perhaps, select in described vocal print characteristic sequence the likelihood score of the mean value of all likelihood scores that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model calculates as this vocal print feature and described polyhybird Gauss model.
7. method as claimed in claim 5, is characterized in that, describedly according to result of calculation, determines that the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user comprises:
Select the likelihood score of the time average of the likelihood score that in described vocal print characteristic sequence, the corresponding polyhybird Gauss model of all vocal print features calculates as the speaker model of described vocal print characteristic sequence and described current login user.
8. method as described as claim 1 to 7 any one, is characterized in that, described method also comprises:
When the user registers, receive the registration voice signal of described user's typing;
Build described user's speaker model according to described registration voice signal;
The described process that builds described user's speaker model according to described registration voice signal comprises:
Extract the vocal print feature from described registration voice signal;
All mixed Gauss models of the speaker model of determining described user according to number of repetition and the frame number of described registration voice signal;
Estimate the Gaussian mean parameter of all mixed Gauss models of described user's speaker model according to the vocal print feature of extracting from described registration voice signal;
Estimate Gauss's variance parameter of all mixed Gauss models of described user's speaker model according to the vocal print feature of extracting from described registration voice signal.
9. method as claimed in claim 8, is characterized in that, the described number of repetition according to described registration voice signal and frame number determine that all mixed Gauss models of described user's speaker model comprise:
The mixed Gauss model number of setting described user's speaker model is less than or equal to the number of repetition of described registration voice signal;
Setting is less than or equal to the frame number of the registration voice signal that described mixed Gauss model is corresponding corresponding to the Gaussage of each mixed Gauss model.
10. an identity authorization system, is characterized in that, comprising:
The voice signal receiving element, for when the user logins, receive the continuous speech signal of current login user typing;
Extraction unit, for extracting the vocal print characteristic sequence of described continuous speech signal, described vocal print characteristic sequence comprises one group of vocal print feature;
The first computing unit, for calculating the likelihood score of described vocal print characteristic sequence and background model;
The second computing unit, for calculating the likelihood score of speaker model of described vocal print characteristic sequence and described current login user, described speaker model is the number of repetition of the registration voice signal of typing during according to described current login user registration and the polyhybird Gauss model of frame number structure;
The 3rd computing unit, for the likelihood score p (X|UBM) of the likelihood score p (X|U) according to described vocal print characteristic sequence and speaker model and described vocal print characteristic sequence and background model, calculate likelihood ratio p = p ( X | U ) p ( X | UBM ) ;
Judging unit, while for the likelihood ratio calculated at described the 3rd computing unit, being greater than the threshold value of setting, determine that described current login user is effective authenticated user, otherwise determine that described current login user is non-authenticated user.
11. system as claimed in claim 10, is characterized in that, described the second computing unit comprises:
The first computation subunit, for calculating respectively the likelihood score of described vocal print characteristic sequence and each mixed Gauss model;
First determines subelement, determines the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user for the result of calculation according to described the first computation subunit.
12. system as claimed in claim 11, is characterized in that, described the first computation subunit comprises:
The first computing module, for calculating respectively the likelihood score of each mixed Gauss model in each vocal print feature of described vocal print characteristic sequence and described polyhybird Gauss model;
First selects module, the likelihood score for the time average of the likelihood score summation of selecting the corresponding mixed Gauss model of one group of vocal print feature of described vocal print characteristic sequence to calculate as described vocal print characteristic sequence and this mixed Gauss model.
13. system as claimed in claim 11, is characterized in that,
Described first determines subelement, the likelihood score specifically for the mean value of the likelihood score of selecting the corresponding all mixed Gauss models of described vocal print characteristic sequence to calculate as the speaker model of described vocal print characteristic sequence and described current login user; The maximum of the likelihood score of perhaps, selecting the corresponding all mixed Gauss models of described vocal print characteristic sequence to calculate is as the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user.
14. system as claimed in claim 10, is characterized in that, described the second computing unit comprises:
The second computation subunit, for calculating respectively the likelihood score of each vocal print feature of described vocal print characteristic sequence with respect to described polyhybird Gauss model;
Second determines subelement, determines the likelihood score of the speaker model of described vocal print characteristic sequence and described current login user for the result of calculation according to described the second computation subunit.
15. system as claimed in claim 14, is characterized in that, described the second computation subunit comprises:
The second computing module, for calculating respectively the likelihood score of each mixed Gauss model in each vocal print feature of described vocal print characteristic sequence and described polyhybird Gauss model;
Second selects module, for selecting maximum in likelihood score that in the corresponding described polyhybird Gauss model of vocal print feature of described vocal print characteristic sequence, each mixed Gauss model the calculates likelihood score as this vocal print feature and described polyhybird Gauss model; Perhaps select in described vocal print characteristic sequence the likelihood score of the mean value of all likelihood scores that in the corresponding described polyhybird Gauss model of a vocal print feature, each mixed Gauss model calculates as this vocal print feature and described polyhybird Gauss model.
16. system as claimed in claim 14, is characterized in that,
Described second determines subelement, specifically for selecting the likelihood score of the speaker model that each vocal print feature in described vocal print characteristic sequence is described vocal print characteristic sequence and described current login user with respect to the time average of the likelihood score of described polyhybird Gauss model.
17. system as described as claim 10 to 16 any one, is characterized in that,
Described voice signal receiving element, also for when the user registers, receive the registration voice signal of described user's typing;
Described system also comprises: the model construction unit, and for build described user's speaker model according to described registration voice signal, described model construction unit comprises:
The feature extraction subelement, for extracting the vocal print feature from described registration voice signal;
Topological structure is determined subelement, determines all mixed Gauss models of described user's speaker model for the number of repetition according to described registration voice signal and frame number;
The first estimator unit, estimate that for the vocal print feature of utilizing described feature extraction subelement to extract described topological structure determines the Gaussian mean parameter of all mixed Gauss models that subelement is determined;
The second estimator unit, estimate that for the vocal print feature of utilizing described feature extraction subelement to extract described topological structure determines Gauss's variance parameter of all mixed Gauss models that subelement is determined.
18. system as claimed in claim 17, is characterized in that,
Described topological structure is determined subelement, is less than or equal to the number of repetition of described registration voice signal specifically for the mixed Gauss model number of the speaker model of setting described user; Setting is less than or equal to the frame number of the registration voice signal that described mixed Gauss model is corresponding corresponding to the Gaussage of each mixed Gauss model.
CN2011102180452A 2011-08-01 2011-08-01 Identity authentication method and system Active CN102238190B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2011102180452A CN102238190B (en) 2011-08-01 2011-08-01 Identity authentication method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2011102180452A CN102238190B (en) 2011-08-01 2011-08-01 Identity authentication method and system

Publications (2)

Publication Number Publication Date
CN102238190A CN102238190A (en) 2011-11-09
CN102238190B true CN102238190B (en) 2013-12-11

Family

ID=44888395

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2011102180452A Active CN102238190B (en) 2011-08-01 2011-08-01 Identity authentication method and system

Country Status (1)

Country Link
CN (1) CN102238190B (en)

Families Citing this family (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102510426A (en) * 2011-11-29 2012-06-20 安徽科大讯飞信息科技股份有限公司 Personal assistant application access method and system
CN102496365A (en) * 2011-11-30 2012-06-13 上海博泰悦臻电子设备制造有限公司 User verification method and device
US8818810B2 (en) * 2011-12-29 2014-08-26 Robert Bosch Gmbh Speaker verification in a health monitoring system
CN102710602B (en) * 2012-04-28 2016-04-13 深圳创维－Rgb电子有限公司 A kind of electronic equipment voice login method, system and television set
CN102968990B (en) * 2012-11-15 2015-04-15 朱东来 Speaker identifying method and system
CN103226951B (en) * 2013-04-19 2015-05-06 清华大学 Speaker verification system creation method based on model sequence adaptive technique
CN105096954A (en) * 2014-05-06 2015-11-25 中兴通讯股份有限公司 Identity identifying method and device
CN104239471B (en) * 2014-09-03 2017-12-19 陈飞 The devices and methods therefor of data query/exchange is carried out in a manner of Behavior modeling
CN104361891A (en) * 2014-11-17 2015-02-18 科大讯飞股份有限公司 Method and system for automatically checking customized polyphonic ringtones of specific population
CN104766607A (en) * 2015-03-05 2015-07-08 广州视源电子科技股份有限公司 Television program recommendation method and system
CN106057206B (en) * 2016-06-01 2019-05-03 腾讯科技（深圳）有限公司 Sound-groove model training method, method for recognizing sound-groove and device
CN106157135A (en) * 2016-07-14 2016-11-23 微额速达（上海）金融信息服务有限公司 Antifraud system and method based on Application on Voiceprint Recognition Sex, Age
CN106228990A (en) * 2016-07-15 2016-12-14 北京光年无限科技有限公司 Login method and operating system towards intelligent robot
CN107705791B (en) * 2016-08-08 2021-06-04 中国电信股份有限公司 Incoming call identity confirmation method and device based on voiceprint recognition and voiceprint recognition system
CN107767863B (en) * 2016-08-22 2021-05-04 科大讯飞股份有限公司 Voice awakening method and system and intelligent terminal
CN107068154A (en) * 2017-03-13 2017-08-18 平安科技（深圳）有限公司 The method and system of authentication based on Application on Voiceprint Recognition
CN109102810B (en) * 2017-06-21 2021-10-15 北京搜狗科技发展有限公司 Voiceprint recognition method and device
CN110223078A (en) * 2019-06-17 2019-09-10 国网电子商务有限公司 Identity authentication method, device, electronic equipment and storage medium
CN111023470A (en) * 2019-12-06 2020-04-17 厦门快商通科技股份有限公司 Air conditioner temperature adjusting method, medium, equipment and device
CN115171727A (en) * 2022-09-08 2022-10-11 北京亮亮视野科技有限公司 Method and device for quantifying communication efficiency

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101833951A (en) * 2010-03-04 2010-09-15 清华大学 Multi-background modeling method for speaker recognition
CN102024455A (en) * 2009-09-10 2011-04-20 索尼株式会社 Speaker recognition system and method

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7447633B2 (en) * 2004-11-22 2008-11-04 International Business Machines Corporation Method and apparatus for training a text independent speaker recognition system using speech data with text labels
CN101136199B (en) * 2006-08-30 2011-09-07 纽昂斯通讯公司 Voice data processing method and equipment

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102024455A (en) * 2009-09-10 2011-04-20 索尼株式会社 Speaker recognition system and method
CN101833951A (en) * 2010-03-04 2010-09-15 清华大学 Multi-background modeling method for speaker recognition

Also Published As

Publication number Publication date
CN102238190A (en) 2011-11-09

Similar Documents

Publication Publication Date Title
CN102238190B (en) Identity authentication method and system
CN102238189B (en) Voiceprint password authentication method and system
CN107610707B (en) A kind of method for recognizing sound-groove and device
CN102509547B (en) Method and system for voiceprint recognition based on vector quantization based
JP6303971B2 (en) Speaker change detection device, speaker change detection method, and computer program for speaker change detection
Chavan et al. An overview of speech recognition using HMM
CN101241699B (en) A speaker identification method for remote Chinese teaching
EP1989701B1 (en) Speaker authentication
CN104900235B (en) Method for recognizing sound-groove based on pitch period composite character parameter
CN108417201B (en) Single-channel multi-speaker identity recognition method and system
CN107221320A (en) Train method, device, equipment and the computer-readable storage medium of acoustic feature extraction model
CN1963917A (en) Method for estimating distinguish of voice, registering and validating authentication of speaker and apparatus thereof
CN105261246B (en) A kind of Oral English Practice error correction system based on big data digging technology
US20080065380A1 (en) On-line speaker recognition method and apparatus thereof
CN102324232A (en) Method for recognizing sound-groove and system based on gauss hybrid models
CN104575490A (en) Spoken language pronunciation detecting and evaluating method based on deep neural network posterior probability algorithm
CN104835498A (en) Voiceprint identification method based on multi-type combination characteristic parameters
CN102270451A (en) Method and system for identifying speaker
CN101923855A (en) Test-irrelevant voice print identifying system
CN104765996A (en) Voiceprint authentication method and system
CN102184654A (en) Reading supervision method and device
CN106297769B (en) A kind of distinctive feature extracting method applied to languages identification
Shivakumar et al. Simplified and supervised i-vector modeling for speaker age regression
CN104901807A (en) Vocal print password method available for low-end chip
Kockmann et al. Recent progress in prosodic speaker verification

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
C56 Change in the name or address of the patentee

Owner name: IFLYTEK CO., LTD.

Free format text: FORMER NAME: ANHUI USTC IFLYTEK CO., LTD.

CP03 Change of name, title or address

Address after: 666, No. 230088 Wangjiang West Road, Hefei hi tech Development Zone, Anhui, China (230088)

Patentee after: Iflytek Co., Ltd.

Address before: 230088 No. 616, Mount Huangshan Road, hi tech Development Zone, Anhui, Hefei

Patentee before: Anhui USTC iFLYTEK Co., Ltd.