CN106448685B - A kind of voiceprint authentication system and method based on phoneme information - Google Patents
A kind of voiceprint authentication system and method based on phoneme information Download PDFInfo
- Publication number
- CN106448685B CN106448685B CN201610880776.6A CN201610880776A CN106448685B CN 106448685 B CN106448685 B CN 106448685B CN 201610880776 A CN201610880776 A CN 201610880776A CN 106448685 B CN106448685 B CN 106448685B
- Authority
- CN
- China
- Prior art keywords
- phoneme
- numeric string
- correlation model
- digital
- neural network
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 239000013598 vector Substances 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 230000001755 vocal effect Effects 0.000 claims abstract description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 3
- 230000000295 complement effect Effects 0.000 claims description 2
- 230000009286 beneficial effect Effects 0.000 abstract description 2
- 238000012795 verification Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 239000011159 matrix material Substances 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000018109 developmental process Effects 0.000 description 3
- 230000004069 differentiation Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000000556 factor analysis Methods 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000005267 amalgamation Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000035479 physiological effects, processes and functions Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/06—Decision making techniques; Pattern matching strategies
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/02—Preprocessing operations, e.g. segment selection; Pattern representation or modelling, e.g. based on linear discriminant analysis [LDA] or principal components; Feature selection or extraction
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/04—Training, enrolment or model building
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/18—Artificial neural networks; Connectionist approaches
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Game Theory and Decision Science (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Telephonic Communication Services (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of voiceprint authentication system and method based on phoneme information, system include that the phoneme based on Chinese putonghua speech identifier forces alignment module, the relevant model creation module of phoneme and the neural network classifier module based on dropout strategy；Method includes defining 16 phoneme class of standard Chinese numeric string vocal print, the explicit each pronunciation classification information for utilizing numeric string；Based on Chinese putonghua speech identifier, the phoneme boundary of each corresponding numeric string content of text is obtained using Viterbi pressure alignment algorithm；Phoneme correlation model is established using the unrelated algorithm of text；Phoneme correlation model is calculated, scores vector is obtained.Beneficial effects of the present invention: the present invention is while realizing that phoneme information cutting, phoneme modeling and phoneme correlation model separating capacity are analyzed, propose the neural network training method using dropout strategy, it solves the problems, such as numeric string phoneme missing, and improves the performance of numeric string voiceprint authentication system.
Description
Technical field
The present invention relates to voiceprint authentication system technical fields, it particularly relates to which a kind of vocal print based on phoneme information is recognized
Demonstrate,prove system and method.
Background technique
Living things feature recognition is a kind of physiological characteristic intrinsic according to human body itself and behavioural characteristic to identify identity
Technology, have many advantages, such as to be not easy to forget, anti-counterfeiting performance is good, be not easy to forge or it is stolen, have and can use whenever and wherever possible with oneself.With
Internet is fast-developing, and traditional identity authentication techniques means are increasingly unable to satisfy the need of user experience and security capabilities
It asks.Sound groove recognition technology in e easy to use has been caused due to its wealthy application prospect, huge Social benefit and economic benefit
The extensive concern of all trades and professions and great attention.
Application on Voiceprint Recognition, also known as Speaker Identification are one kind of biological identification technology.The technology in speech waveform by reflecting
The speech parameter for human physiology and the behavioural characteristic of speaking, and then tell speaker's identity.With high security, data acquisition is convenient
The features such as.
In recent years, the Speaker Identification of text related (Text-dependent) becomes the hot spot in user authentication field.By
Major progress in unrelated (Text-independent) the Speaker Identification field of text, many researchers attempt by text without
It closes speaker's recognizer and is applied to text related fields, such as numeric string Application on Voiceprint Recognition.
Under numeric string authentication condition, have researcher using simultaneous factor analysis (Joint Factor Analysis,
JFA), gauss hybrid models-interference properties map (Gaussian Mixture Model-Nuisance Attribute
Projection, GMM-NAP) and Hidden Markov Model-interference properties mapping (Hidden Markov Model-Nuisance
Attribute Projection, HMM-NAP) it is compared.For JFA, the algorithm based on NAP is performed better than, reason
It is that JFA is trained to need a large amount of tagged data, and exists between the training data of JFA matrix and numeric string test data and lose
Match.
In the unrelated Speaker Identification of text, JFA and be based on probability linear discriminant analysis (Probabilistic Linear
Discriminant Analysis, PLDA) population variance modeling factors (iVector) algorithm all rely on a large amount of exploitation
Collect data.More and more work are dedicated to development set data in the limited field of processing and ask to outside field using the migration of data
Topic, such as the adaptive and backoff algorithm of lexical gap.
By the mobile phone of Android system (Android) and apple system (iOS), records and construct the number comprising 536 people
Word string voice set.It is divided into two kinds of scenes: global condition and rand-n condition.Global condition indicates that registration and verifying use
Identical numeric string content；Rand-n condition indicates that each digit string is the random number word string that length is n, this is at certain
It is more safer than global condition in the application system of a little anti-recording attacks.Involved in the present invention it is as shown in Table 1 three kinds registration/
Authentication condition: fixed whole numerical ciphers, 6 bit digital passwords of 8 bit digital passwords of dynamic and dynamic.Every kind of scene partitioning development set
Collect with evaluation and test.Development set is for training global context model (Universal Background Model, UBM), population variance
Modeling matrix (iVector T matrix) and linearly differentiation analysis matrix (Linear Discriminant Analysis,
LDA) etc..In three kinds of conditions for evaluating and testing collection, everyone includes three registration voices and a tested speech, every tested speech and
All speaker models are compared.
Table 1: several form examples of password figure
Table 2 be GMM-NAP and using iVector voiceprint authentication system etc. error rates (Equal Error Rate,
EER it) compares.The result shows that the performance of voiceprint authentication system has obtained significantly consistently mentioning with the increase of digital string length
It rises.But GMM-NAP and iVector system does not account for the utilization of phoneme (Phone/Phoneme) information, be based on text without
Close direct application of the Application on Voiceprint Recognition under text associated scenario.In the application of numeric string vocal print, ignore phoneme information or no sound
The effective use of prime information, it will the effect of the unrelated recognizer of limitation text in practical applications.
Table 2:GMM-NAP and iVector system under different test conditions etc. error rates comparison
| Fixed whole numerical ciphers | Dynamically 8 bit digital password | Dynamically 6 bit digital password | |
| GMM-NAP | 2.09% | 2.64% | 3.76% |
| iVector | 1.87% | 2.40% | 3.32% |
Summary of the invention
It is an object of the invention to propose a kind of voiceprint authentication system and method based on phoneme information, sound can be being realized
While prime information cutting, phoneme model (Phone-dependent) model separating capacity related to phoneme analysis, solves number
The problem of word string phoneme lacks, and improve the performance of numeric string voiceprint authentication system.
To realize the above-mentioned technical purpose, the technical scheme of the present invention is realized as follows:
A kind of voiceprint authentication system based on phoneme information is forced including the phoneme based on Chinese putonghua speech identifier
The relevant model creation module of alignment module, phoneme and the neural network classifier module based on dropout strategy；
The phoneme based on Chinese putonghua speech identifier forces alignment module to be used for 16 sounds to numeric string
Plain classification carries out cutting；
The relevant model creation module of the phoneme analyzes each phoneme correlation model for establishing phoneme correlation model
To the separating capacity of voiceprint, the differentiation feature of speaker is featured, rather than difference between vocabulary；
The neural network classifier module based on dropout strategy is used to merge the complementary letter of phoneme correlation model
Breath.
A kind of voiceprint authentication method based on phoneme information, includes the following steps:
S01: 16 phoneme class of standard Chinese numeric string vocal print, the explicit each pronunciation for utilizing numeric string are defined
Classification information；
S02: being based on Chinese putonghua speech identifier, forces alignment algorithm to obtain each corresponding numeric string using Viterbi
The phoneme boundary of content of text is completed the phone segmentation to voice content, the i.e. mapping relations of speech feature vector to phoneme, is obtained
To the feature vector subclass for belonging to phoneme, each character subset conjunction is considered as independent data flow and carries out subsequent processing；
S03: phoneme correlation model is established using the unrelated algorithm of text, the relevant model foundation process of phoneme reduces each
The parameter amount of phoneme correlation model, avoids model excessively trained；
S04: phoneme correlation model is calculated, scores vector is obtained.
Further, using the dropout Strategies Training rear end integrated classification device in neural network algorithm in step S04.
Beneficial effects of the present invention:
(1) present invention forces alignment algorithm to obtain using typical Chinese putonghua speech identifier is based on using Viterbi
The phoneme boundary of each corresponding numeric string content of text is taken, completes to be based on the phone segmentation of voice content compared to common
The cutting effect of dynamic time warping (Dynamic Time Warping, DTW) scheduling algorithm is advantageously；
(2) present invention defines 16 pronunciation classifications to the numeric string pronunciation of standard Chinese, avoids affiliated phoneme class
Model caused by feature vector is very few crosses training problem, establishes phoneme correlation model, and analyze each phoneme correlation model pair
The separating capacity of voiceprint, phoneme correlation model feature the differentiation feature of speaker, rather than the difference between vocabulary；
(3) it in order to further increase the use of information effect of phoneme correlation model, and considers to authenticate language in practical application
Sound only includes the partial content of set of phonemes, it is understood that there may be the problem of vector dimension lacks, using dropout Strategies Training nerve
Network backend classifier, realizes the amalgamation judging of phoneme associated score vector, and has been obviously improved the system performance of voiceprint.
Detailed description of the invention
Fig. 1 is the rear end classifier process flow diagram in the present invention based on the relevant scores vector of phoneme；
Fig. 2 be in the present invention for different phoneme correlation models etc. error rates experimental result picture.
Specific embodiment
With reference to the attached drawing in the embodiment of the present invention, technical solution in the embodiment of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, those of ordinary skill in the art's every other embodiment obtained belong to the model that the present invention protects
It encloses.
Phoneme information is explicitly utilized the numeric string voiceprint authentication method combined with neural network classification by present invention proposition,
For every digit string, alignment algorithm is forced to be completed to voice content using the Viterbi of Chinese putonghua speech identifier
Phone segmentation；The training parameter amount for reducing phoneme correlation model, avoids the training phonetic feature of each phoneme model is less can
Caused training problem can be crossed, analyze each phoneme model to the separating capacity of Application on Voiceprint Recognition；To the score of phoneme correlation model
The problem of vector is lacked there may be dimension, using the dropout Strategies Training rear end integrated classification device in neural network algorithm,
The utilizing status of phoneme relevant information is improved, the system performance of numeric string voiceprint is further improved.
Table 3 gives the phonemic representation of ten standard Chinese numeric utterances.It is noted that digital " 1 " has " y i " and " y
Two kinds of ao " pronunciations, therefore corresponding ten standard Chinese numeric utterances share 16 phonemes.
3: ten, table digital standard Chinese pronunciation phonemes
In " fixed whole numerical ciphers " condition, phoneme content immobilizes." 8 bit digital passwords of dynamic " and " dynamic 6
The phoneme content of numerical ciphers " is also known, because digital text is typically based on random algorithm push or the base of background system
It is generated in OTP dynamic password (One-time Password) according to special algorithm.
Based on Chinese putonghua speech identifying system, alignment algorithm is forced to obtain each corresponding content of text using Viterbi
Phoneme boundary, complete to the phone segmentation of voice content, the i.e. mapping of speech feature vector to phoneme.
Therefore, acoustic feature sequence vector χ=x of one section of digit string is given1,...,xT, discrete son can be cut into
Set χ1,...,χ16.Wherein x ∈ χiIndicate the feature vector subclass for belonging to i-th of phoneme.Each subclass is considered as solely
Vertical data flow carries out subsequent processing.Voiceprint registration stage, the relevant model of 16 phonemes(i-th of sound of speaker s
Sub-prime set) it is obtained by the unrelated algorithm training of text.It should be noted that registration voice needs to cover ten numbers.This hair
In bright, registration phase registers voice using three numeric strings, guarantees that each number at least occurs one in everyone registration voice
Time.
During voiceprint, for " fixed whole numerical ciphers " condition, ten sextuple scores vector ξ are obtained,Can by scores vector ξ is averaged or the methods of logistic regression training rear end classifier into
Row judgement.However it may for rand-n condition, the scores vector ξ such as " 8 bit digital passwords of dynamic " and " 6 bit digital passwords of dynamic "
There are missings, because tested speech only includes the partial content of set of phonemes.In order to solve this problem, using neural network algorithm
In dropout strategy, this is a kind of implementation method for effectively promoting generalization ability.
The dropout training algorithm of neural network is stochastic gradient descent (the Standard Stochastic of standard
Gradient Descent), certain input units and hidden layer are only ignored at random with certain probability γ during forward calculation
Unit.Only activation unit participates in backpropagation (Back-propagation) and gradient calculates.Because dropout is not used to
Identification, in the training process, the output to every layer is readjusted:
Wherein δ (), WlAnd blIt is activation primitive, l layers of weight and l layers of biasing respectively.Bm is binary cover
(Binary mask) indicates which dimension is removed, and * indicates vector multiplication.
The above process can regard a kind of effective model averaging method as, i.e., by the missing of a large amount of shared weight of training to
The average expression of the heterogeneous networks measured.As shown in Figure 1, training includes the neural network classifier of a hidden layer.It is wherein defeated
Entering is scores vector, and output includes two units, respectively represents target authentication classification and emits imitative certification classification.For " dynamic 8
Vector dimension under the conditions of the rand-n such as numerical ciphers " and " 6 bit digital passwords of dynamic " lacks problem, to input layer with probability γ
Network training is carried out using dropout strategy.In Qualify Phase, calculates following log-likelihood ratio and is exported as system:
Wherein p (ξ | target verification class) and p (ξ | emit imitative verifying class) are the likelihood scores of scores vector ξ.Pass through Bayes's public affairs
Formula, likelihood score can be exchanged into posteriority expression,
P (ξ | target verification class)=p (target verification class | ξ) p (ξ)/p (target verification class)
P (ξ | emit imitative verifying class)=p (emit imitative verifying class | ξ) p (ξ)/p (emitting imitative verifying class)
Wherein p (target verification class | ξ) and p (emit imitative verifying class | ξ) are that scores vector ξ is obtained by network forward calculation
Posteriority.P (target verification class) and p (emitting imitative verifying class) is the priori and Mao Fang of the target verification class obtained from training set estimation
Verify the priori of class.P (ξ) is unrelated with any model, can ignore during calculating LLR.
Each phoneme model is analyzed first to the separating capacity of Application on Voiceprint Recognition.In view of the training voice of each phoneme model
Feature is less, in order to avoid crossing training problem, reduces the training parameter amount of each phoneme correlation model.Fig. 2 gives each
Phoneme correlation model etc. error rates comparison.
From figure 2 it can be seen that firstly, iVector to be better than GMM- more by a small margin in all phoneme correlation models
NAP model.Secondly, the EER numerical value for the consonant " w " that performance is worst is five times or so of the EER of the best vowel of performance " an ".This
A experimental result has directive function to practical application, and on-line system can limit the bad number of push performance, such as " 5
[wu]”。
By training dropout neural network rear end classifier, fusion output is carried out to the relevant scores vector of phoneme.Table
4 give phoneme correlation model using different rear end classifiers etc. error rates compare.Compare for convenience, also gives here
The authentication performance that the phoneme associated score of GMM-NAP and iVector system is averaged.Score average formula is as follows:
Table 4: phoneme correlation model using different rear end classifiers etc. error rates compare
From table 4, it can be seen that of the present invention explicitly utilize the calculation with the fusion of neural network rear end based on phoneme information
Method can effectively promote the system performance of numeric string voiceprint.It is average compared to score as a result, neural network rear end point
Class device etc. error rates it is lower, performance is more excellent.With GMM-NAP the and iVector Comparative result of table 2, registers/recognize in three kinds of differences
Under the conditions of card, the algorithm of phoneme correlation model and neural network rear end classifier is achieved under about 20% or so opposite EER
Drop.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (2)
1. a kind of voiceprint authentication system based on phoneme information, which is characterized in that including being based on Chinese putonghua speech identifier
Phoneme force alignment module, the relevant model creation module of phoneme and the neural network classifier mould based on dropout strategy
Block；
The phoneme based on Chinese putonghua speech identifier forces alignment module to be used for 16 phoneme classes to numeric string
Not carry out cutting, digital 0 corresponding pronunciation phonemes are li, ig after cutting, and digital 1 corresponding pronunciation phonemes are yi, i, ao, number
2 corresponding pronunciation phonemes are er, and digital 3 corresponding pronunciation phonemes are s, an, and digital 4 corresponding pronunciation phonemes are s, ih, number 5
Corresponding pronunciation phonemes are w, u, and digital 6 corresponding pronunciation phonemes are li, ou, and digital 7 corresponding pronunciation phonemes are qi, i, number
8 corresponding pronunciation phonemes are b, a, and digital 9 corresponding pronunciation phonemes are ji, ou；；
The relevant model creation module of the phoneme analyzes each phoneme correlation model to sound for establishing phoneme correlation model
The separating capacity of line certification；
The neural network classifier module based on dropout strategy is used to merge the complementary information of phoneme correlation model, leads to
It crosses and trains dropout neural network rear end classifier, fusion output is carried out to the relevant scores vector of phoneme, and provide to phoneme
The authentication performance that associated score is averaged, score average formula are as follows:
WhereinFor the average value of ten sextuple scores vectors, ξiIt is 16
Tie up a certain vector in scores vector, χtestFor tested speech acoustic feature sequence vector.
2. a kind of voiceprint authentication method of voiceprint authentication system based on phoneme information described in accordance with the claim 1, feature
It is, includes the following steps:
S01: 16 phoneme class of standard Chinese numeric string vocal print, the explicit each pronunciation classification for utilizing numeric string are defined
Information；
S02: being based on Chinese putonghua speech identifier, forces alignment algorithm to obtain each corresponding numeric string text using Viterbi
The phoneme boundary of content is completed to obtain the feature vector subclass for belonging to phoneme to the phone segmentation of voice content；
S03: phoneme correlation model is established using the unrelated algorithm of text；
S04: calculating phoneme correlation model, obtain scores vector, wherein using the dropout Strategies Training in neural network algorithm
Rear end integrated classification device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610880776.6A CN106448685B (en) | 2016-10-09 | 2016-10-09 | A kind of voiceprint authentication system and method based on phoneme information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610880776.6A CN106448685B (en) | 2016-10-09 | 2016-10-09 | A kind of voiceprint authentication system and method based on phoneme information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106448685A CN106448685A (en) | 2017-02-22 |
| CN106448685B true CN106448685B (en) | 2019-11-22 |
Family
ID=58172115
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610880776.6A Active CN106448685B (en) | 2016-10-09 | 2016-10-09 | A kind of voiceprint authentication system and method based on phoneme information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106448685B (en) |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198574B (en) * | 2017-12-29 | 2020-12-08 | 科大讯飞股份有限公司 | Sound change detection method and device |
| CN108648760B (en) * | 2018-04-17 | 2020-04-28 | 四川长虹电器股份有限公司 | Real-time voiceprint identification system and method |
| CN109065023A (en) * | 2018-08-23 | 2018-12-21 | 广州势必可赢网络科技有限公司 | A kind of voice identification method, device, equipment and computer readable storage medium |
| CN110875044B (en) * | 2018-08-30 | 2022-05-03 | 中国科学院声学研究所 | A speaker recognition method based on word correlation score calculation |
| CN110111798B (en) * | 2019-04-29 | 2023-05-05 | 平安科技（深圳）有限公司 | Method, terminal and computer readable storage medium for identifying speaker |
| CN110689895B (en) * | 2019-09-06 | 2021-04-02 | 北京捷通华声科技股份有限公司 | Voice verification method and device, electronic equipment and readable storage medium |
| CN111243603B (en) * | 2020-01-09 | 2022-12-06 | 厦门快商通科技股份有限公司 | Voiceprint recognition method, system, mobile terminal and storage medium |
| CN111341320B (en) * | 2020-02-28 | 2023-04-14 | 中国工商银行股份有限公司 | Phrase voice voiceprint recognition method and device |
| CN111785284B (en) * | 2020-08-19 | 2024-04-30 | 科大讯飞股份有限公司 | Text-independent voiceprint recognition method, device and equipment based on phoneme assistance |
| CN114093371B (en) * | 2021-10-11 | 2024-12-03 | 浙江大学 | Phoneme-level voiceprint recognition adversarial sample construction system and method based on neural network generation model |
| CN114299921B (en) * | 2021-12-07 | 2022-11-18 | 浙江大学 | Voiceprint security scoring method and system for voice command |
| CN115966210A (en) * | 2022-12-06 | 2023-04-14 | 四川启睿克科技有限公司 | Phoneme selection method and device for voiceprint recognition |
| CN115831120B (en) * | 2023-02-03 | 2023-06-16 | 北京探境科技有限公司 | Corpus data acquisition method and device, electronic equipment and readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101467204A (en) * | 2005-05-27 | 2009-06-24 | 普提克斯科技股份有限公司 | Method and system for bio-metric voice print authentication |
| CN204465555U (en) * | 2015-04-14 | 2015-07-08 | 时代亿宝(北京)科技有限公司 | Based on the voiceprint authentication apparatus of time type dynamic password |
| CN104834849A (en) * | 2015-04-14 | 2015-08-12 | 时代亿宝(北京)科技有限公司 | Dual-factor identity authentication method and system based on voiceprint recognition and face recognition |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070033041A1 (en) * | 2004-07-12 | 2007-02-08 | Norton Jeffrey W | Method of identifying a person based upon voice analysis |
-
2016
- 2016-10-09 CN CN201610880776.6A patent/CN106448685B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101467204A (en) * | 2005-05-27 | 2009-06-24 | 普提克斯科技股份有限公司 | Method and system for bio-metric voice print authentication |
| CN204465555U (en) * | 2015-04-14 | 2015-07-08 | 时代亿宝(北京)科技有限公司 | Based on the voiceprint authentication apparatus of time type dynamic password |
| CN104834849A (en) * | 2015-04-14 | 2015-08-12 | 时代亿宝(北京)科技有限公司 | Dual-factor identity authentication method and system based on voiceprint recognition and face recognition |
Non-Patent Citations (1)
| Title |
|---|
| 语音声纹密码验证技术研究;张涛涛;《中国优秀硕士学位论文全文数据库 信息科技辑》;20160915(第9期);I136-43第9-17页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106448685A (en) | 2017-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106448685B (en) | A kind of voiceprint authentication system and method based on phoneme information | |
| Hansen et al. | Speaker recognition by machines and humans: A tutorial review | |
| CN103456304B (en) | For the dual methods of marking and system with the relevant speaker verification of text | |
| Rúa et al. | Online signature verification based on generative models | |
| CN108766445A (en) | Method for recognizing sound-groove and system | |
| KR20060070603A (en) | Method and device for verifying two-stage speech in speech recognition system | |
| Wang et al. | A network model of speaker identification with new feature extraction methods and asymmetric BLSTM | |
| CN108091326A (en) | A kind of method for recognizing sound-groove and system based on linear regression | |
| Safavi et al. | Fraud detection in voice-based identity authentication applications and services | |
| CN101540170A (en) | Voiceprint recognition method based on biomimetic pattern recognition | |
| Maghsoodi et al. | Speaker recognition with random digit strings using uncertainty normalized HMM-based i-vectors | |
| CN114360553A (en) | Method for improving voiceprint safety | |
| CN110390948A (en) | A kind of method and system of Rapid Speech identification | |
| Li et al. | Cost‐Sensitive Learning for Emotion Robust Speaker Recognition | |
| KR20240132372A (en) | Speaker Verification Using Multi-Task Speech Models | |
| Büyük | Sentence‐HMM state‐based i‐vector/PLDA modelling for improved performance in text dependent single utterance speaker verification | |
| Zhonghua et al. | An overview of modeling technology of speaker recognition | |
| Georgescu et al. | GMM-UBM modeling for speaker recognition on a Romanian large speech corpora | |
| Saleema et al. | Voice biometrics: the promising future of authentication in the internet of things | |
| Laskar et al. | HiLAM-aligned kernel discriminant analysis for text-dependent speaker verification | |
| CN111081261B (en) | Text-independent voiceprint recognition method based on LDA | |
| Siu et al. | Discriminatively trained GMMs for language classification using boosting methods | |
| Al-Tekreeti et al. | Speaker voice recognition using a hybrid PSO/fuzzy logic system | |
| Mohamed et al. | An Overview of the Development of Speaker Recognition Techniques for Various Applications. | |
| WO2009110613A1 (en) | Personal collation device and speaker registration device, and method and program |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: East Zone 9A, 9th Floor, Building 1, No. 158 West Fourth Ring North Road, Haidian District, Beijing, 100142 Patentee after: Beijing Yuan Jian Polytron Technologies Inc. Address before: East Zone 9A, 9th Floor, Building 1, No. 158 West Fourth Ring North Road, Haidian District, Beijing, 100142 Patentee before: Beijing Yuanjian Technologies Co.,Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20231010 Address after: No. 016, Xiaocuigezhuang Village, Gaolou Town, Sanhe City, Langfang City, Hebei Province, 065200 Patentee after: Liu Xuefeng Address before: East Zone 9A, 9th Floor, Building 1, No. 158 West Fourth Ring North Road, Haidian District, Beijing, 100142 Patentee before: Beijing Yuan Jian Polytron Technologies Inc. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240218 Address after: Room 320, 3rd Floor, Building A, No. 119 West Fourth Ring North Road, Haidian District, Beijing, 100000 Patentee after: Beijing Yuanjian Information Technology Co.,Ltd. Country or region after: China Address before: No. 016, Xiaocuigezhuang Village, Gaolou Town, Sanhe City, Langfang City, Hebei Province, 065200 Patentee before: Liu Xuefeng Country or region before: China |