CN109192216A - A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device - Google Patents
A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device Download PDFInfo
- Publication number
- CN109192216A CN109192216A CN201810895193.XA CN201810895193A CN109192216A CN 109192216 A CN109192216 A CN 109192216A CN 201810895193 A CN201810895193 A CN 201810895193A CN 109192216 A CN109192216 A CN 109192216A
- Authority
- CN
- China
- Prior art keywords
- voice
- data
- emulation
- noise
- module
- 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.)
- Pending
Links
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
-
- 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
- G10L17/06—Decision making techniques; Pattern matching strategies
- G10L17/08—Use of distortion metrics or a particular distance between probe pattern and reference templates
-
- 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
- G10L17/20—Pattern transformations or operations aimed at increasing system robustness, e.g. against channel noise or different working conditions
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/32—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols including means for verifying the identity or authority of a user of the system or for message authentication, e.g. authorization, entity authentication, data integrity or data verification, non-repudiation, key authentication or verification of credentials
- H04L9/3226—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols including means for verifying the identity or authority of a user of the system or for message authentication, e.g. authorization, entity authentication, data integrity or data verification, non-repudiation, key authentication or verification of credentials using a predetermined code, e.g. password, passphrase or PIN
- H04L9/3231—Biological data, e.g. fingerprint, voice or retina
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/32—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols including means for verifying the identity or authority of a user of the system or for message authentication, e.g. authorization, entity authentication, data integrity or data verification, non-repudiation, key authentication or verification of credentials
- H04L9/3263—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols including means for verifying the identity or authority of a user of the system or for message authentication, e.g. authorization, entity authentication, data integrity or data verification, non-repudiation, key authentication or verification of credentials involving certificates, e.g. public key certificate [PKC] or attribute certificate [AC]; Public key infrastructure [PKI] arrangements
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Computer Security & Cryptography (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Telephonic Communication Services (AREA)
Abstract
The invention belongs to vocal print acquisition and vocal print information technology fields, disclose a kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device, the acquisition methods include the method for channel coding, the method for environmental noise emulation, method, the method for data decoded method and vocal print database establishment of communication pattern emulation, and the acquisition device includes the voice acquisition module being successively linked in sequence, voice coding module, environmental noise emulation module, communication pattern scrambling emulation module, voice codec module and vocal print database module；Its recognition effect that can have been obtained, it is able to carry out noise processed, application training can be directly carried out in data training process, can be improved Application on Voiceprint Recognition environmental noise robustness and cha nnel robustness, and the Application on Voiceprint Recognition that can be used under different communication modes is emulated with training dataset to be obtained.
Description
Technical field
The invention belongs to vocal print acquisition and vocal print information technology field more particularly to a kind of Application on Voiceprint Recognition training datasets
Emulate acquisition methods and its acquisition device.
Background technique
Application on Voiceprint Recognition, also referred to as Speaker Identification are that the life of speaker's identity is judged automatically according to voice using computer
Object feature identification technique；According to different application scenarios, there are many classification methods for sound groove recognition technology in e: according to voice content whether
It is known that can be divided into text relevant unrelated with text for Application on Voiceprint Recognition；According to the difference of identification mission, Application on Voiceprint Recognition can be divided into
Talk about people's identification and speaker verification；Sound groove recognition technology in e is mainly used in the fields such as public safety, the criminal investigation administration of justice and finance.
In recent years, the unrelated speaker of the text of mainstream recognizes (hereinafter referred to as Speaker Identification) technology and is based on
Gauss hybrid models-universal background model (Gaussian mixture that Douglas A. Reynolds was proposed in 2000
Model-universal background models, GMM-UBM) Speaker Recognition System.GMM-UBM system is from speaker
It identifies angle, proposes the theoretical frame and implementation method for measuring two sections of voice similarity degrees, there is landmark meaning.
Product under different scenes can all have different environmental noises, even identical product also has different background rings
Border, such as intelligent sound box are used in family using and company, and environmental noise also can be different, needs before using Application on Voiceprint Recognition
The environmental robustness of product is assessed, this index shows adaptability of this technology under varying environment noise, keeps away
Exempt from all to be carefully, once ineffective to user environment when company debugs.
Voice communication refers to by voice and by the communication way of transmission medium, there is base call, mobile phone communication, intercommunication
Machine is conversed, and the voice-enabled chat etc. above network is referred to as voice communication.Voice communication mode common at present has fixed line, mobile phone
GSM(Global System for Mobile Communication, global system for mobile communications), mobile TD-SDMA(Time
Division-Synchronous Code Division Multiple Access, TD SDMA), connection
WCDMA(Wideband Code Division Multiple Access, wideband code division multiple access), telecommunications CDMA(Code
Division Multiple Access, CDMA), LTE(Long Term Evolution, long term evolution) and sending out
The 5G etc. of exhibition.
PSTN(Public Switched Telephone Network, Public Switched Telephony Network) i.e. in daily life
Common telephone network.PSTN is a kind of circuit-switched network based on analogue technique, and the speech coding algorithm used is
G.711 a rate coding mode or u rate coding mode.VOIP(Voice over Internet Protocol, the networking telephone) often
The coding mode used is the G.723 standard of International Telecommunication Union, specially algebraic code-excited linear predictive coding ACELP
(Algebraic Code Excited Linear Prediction, algebraic code-excited linear predictive coding) coding.Wechat
The communication mode that phone, QQ phone use is that narrowband self-adaption multi code Rate of Chinese character AMR-NB(Adaptive Multi-Rate(is adaptive
Multi-rate coding) narrowband-Narrow Band()) coding mode.2G communication, i.e. Generation Mobile Telecommunication System technology, packet Chinese juniper GSM are logical
Letter system and CDMA1x communication system, wherein the 2G of China Mobile and China Unicom uses GSM standard, and China Telecom 2G makes
It is CDMA1x standard.GSM voice coding is Regular-Pulse Excitation long-term linearity predictive coding RPE-LTP.3G is communicated
The WCDMA of TD-SCDMA and China Unicom that China Mobile independently formulates and the CDMA2000 of China Telecom.SCDMA and
WCDMA uses adaptive multi-beam forming AMR-NB or AMR-WB(Adaptive Multi-Rate(adaptive multi-rate to compile
Code) broadband-Wide Band()) coding.Telecommunications 2G, 3G use enhanced variable rate codec EVRC or
QCELP coding mode.4G communication, China Mobile use TD-LTE(Time Division Long Term
Evolution, time-division-long term evolution) standard, China Unicom and China Telecom use FDD-LTE(frequency division duplex-long-term
Evolution) standard.That 4G communication uses is high definition voice communication VoLTE, and voice coding modes are adaptive multi-beam forming AMR.
Due to the extensive use of digital voice communication system, sound passes through after microphone acquisition into Voiceprint Recognition System
Many links, include different microphone types, different audio CODEC(coders), different transmission channel etc.,
These can all have vocal print feature and influence, or be illustrated with intelligent sound box, if in registration be with mobile phone terminal app, and
Verifying is then directly to speak against speaker when using, mobile phone MIC(microphone) with the speaker MIC channel that be exactly two different, this
The accuracy rate that verifying may be reduced in the case of kind, is channel mismatch, therefore, in addition to advising in product level on technical term
It keeps away, it is also desirable to consider performance of the sound groove recognition technology in e in different channels；Voiceprint Recognition System can obtain in the actual environment
Training voice and tested speech coding it is often different, Application on Voiceprint Recognition at this moment is just faced with since trained and tested speech is compiled
The voice channel mismatch problem that code is different and generates, this will have a huge impact the performance of system；Channel is not solved not
It is to improve Speaker Identification performance with problem, enhances one of the key of system degree of being practical.
To solve the problems, such as channel mismatch in Application on Voiceprint Recognition, there are two types of technological approaches；One is vocal print of the research across channel to build
Modulo n arithmetic；Main vocal print modeling algorithm has NAP(Nuisance Attribute Proje ction, disturbance component under the technology
Projection model), JFA(Joint Factor Analysis, simultaneous factor analysis model), i-vector(identity-
Vector, identity-based authentication vector) Speaker Identification modeling method and combine speech recognition DNN (Deep Neural
Networks, deep neural network) acoustic model and the Speaker Identification modeling method of i-vector model etc..
NAP and JFA is the subspace model put forward for channel mismatch problem；Wherein NAP direct estimation goes out one
Channel subspace, then by the subspace from GMM(Gaussian Mixed Model, gauss hybrid models) mean value super vector is empty
Between it is middle removal to reduce interference of the channel information to Speaker Identification；JFA thinks empty in the higher-dimension of GMM mean value super vector
Between in, there are two sub-spaces to have separately included speaker information and channel information, by combining to the two subspaces
Modeling can more effectively separate speaker information in voice and channel information, thus the speaker under promoting Complex Channel
Identifying system performance；Due to containing speaker information more abundant in the channel subspace in JFA model, JFA is to saying
The method that words people and channel separately model can generate biggish damage to speaker information, and 2010, Dehak et al. was in JFA
Basis propose i-vector model；A sub-spaces are defined only in i-vector model, referred to as entire change is empty
Between, speaker information and channel information are contained in the subspace simultaneously.Further every section of voice has been expressed as the subspace
In a low dimension vector, i.e. i-vector；Weaken letter finally by the mode of channel compensation is carried out in i-vector level
Influence of the road to Speaker Recognition System performance；Compared with JFA model, the complexity of i-vector model is greatly reduced, together
When it is more flexible by way of carrying out channel compensation in subspace, and show better Speaker Identification performance,
And this is but also i-vector model becomes the Speaker Identification modeling method of most mainstream and forefront.
2014, Lei and Kenny et al. proposed a kind of combination speech recognition DNN acoustic model and i-vector model
Speaker Identification modeling method: during the valuation of i-vector model correlation sufficient statistic, using in speech recognition
Traditional UBM model is replaced to calculate frame posterior probability the DNN acoustic model that phoneme state is classified；This method reduce
System modelling complexity, recognition effect are promoted obvious.
Above-mentioned method for recognizing sound-groove may serve to solve the problems, such as channel mismatch, but there are corresponding problems；By taking JFA as an example,
It is required that amount of training data is very big, operand is also very big when test, is often difficult to obtain fine knowledge in practical applications
Other effect；And the problem of for environmental noise, in the prior art in addition to noise processed, there has been no direct in data training process
Carry out the relevant technologies of application training.
Summary of the invention
In view of the problems of the existing technology, the present invention provides one kind to be able to solve channel mismatch problem, can obtain
Good recognition effect, is able to carry out noise processed, application training can be directly carried out in data training process, can be improved sound
Line environment-identification noise robust and cha nnel robustness can be used in the Application on Voiceprint Recognition training dataset under different communication modes
Emulate acquisition methods and its acquisition device.
The invention is realized in this way one aspect of the present invention provides a kind of Application on Voiceprint Recognition training dataset emulation acquisition side
Method, the acquisition methods include the method for channel coding, the method for environmental noise emulation, method, the data of communication pattern emulation
The method of decoded method and vocal print database establishment,
The method of channel coding the following steps are included:
Read primary voice data；
Header file is removed according to format standard according to the voice print database of the primary voice data, obtains pure speech data block；
Select the voice communication mode to be emulated；
Data volume is carried out to the obtained pure speech data block according to the voice communication mode corresponding speech coding standard
Code, obtains compressed voice data；
Environmental noise emulation method the following steps are included:
Select the environmental noise mode to be emulated；
The compressed voice number obtained according to the method for the varying environment noise and sound pressure levels of selection and channel coding
According to same channel mixing is carried out, the voice data comprising environmental noise is obtained；
Communication pattern emulation method the following steps are included:
Obtain the scrambling parameter that voice channel transmits under different relative amplitudes and state of signal-to-noise；
Scrambling parameter is selected, includes environmental noise to passing through described in coding, noise mixing in the method for environmental noise emulation
Voice data carries out Channel scrambling operation, obtains the Hybrid communication model simulated voice data comprising environmental noise；
The decoded method of data the following steps are included:
Corresponding tone decoding algorithm is selected according to voice communication mode；
The Hybrid communication model simulated voice data comprising environmental noise are added with corresponding tone decoding algorithm corresponding
Audio file head obtains trained voice identical with tested speech channel condition；
Voice print database construction method the following steps are included:
Data sample library is established according to voice and characteristic model data item；
Data sample library is established according to voice messaging data item；
Database interface is established according to voice print database interface specification.
Application on Voiceprint Recognition of the invention emulates acquisition methods with training dataset, first acquisition primary voice data, then right
The voice print database of primary voice data carries out environmental noise emulation, then in communication pattern emulation, i.e. communication pattern soft simulation,
First make an uproar according to being mixed in the method that the communication pattern to be emulated emulates environmental noise by coding, noise comprising environment
The voice data of sound carries out voice coding；Then, scrambling parameter is emulated to the language after coding according to respective communication mode lower channel
Sound data carry out communication pattern and scramble simulation operations；Finally, carrying out voice solution to the voice data after communication pattern scrambling emulation
Code operation, obtains the voice under respective communication mode；It is by carrying out communication pattern emulation to acquired original voice, so that training
Voice is identical with the channel condition of tested speech, solves channel mismatch phenomenon；It effectively can carry out channel to voice communication
Soft simulation, the voice communication courses such as simulation fixed line, 2G, 3G, 4G, VOIP virtual speech and Internet chat, to obtain and test
The identical trained voice of voice channel condition, efficiently solves the problems, such as channel mismatch, is suitable for actual application demand.
One aspect of the present invention provides a kind of Application on Voiceprint Recognition training dataset emulation as described in one aspect of the invention and obtains
Take the acquisition device of method, the acquisition device include voice acquisition module, voice coding module, environmental noise emulation module,
Communication pattern scrambles emulation module, voice codec module and vocal print database module；And voice acquisition module, voice coding mould
Block, environmental noise emulation module, communication pattern scramble emulation module, voice codec module and vocal print database module successively sequence
Connection.
Application on Voiceprint Recognition of the invention training dataset emulates acquisition device, and voice acquisition module is for acquiring speaker's
The voice print database of primary voice data；Voice coding module is used to carry out voice coding to the voice print database of primary voice data,
To obtain the compressed voice data under respective communication mode, i.e. vocoded data；Environmental noise emulation module is used for
To the environmental noise data of vocoded data mixing selection, the voice data comprising environmental noise is obtained, that is, realizes different works
Sound coding data under condition environment；Communication pattern scrambles emulation module and selects respective channel according to communication pattern, realizes to sound
The communication pattern of sound coded data emulates；Voice codec module is according to corresponding decoding algorithm, to by environmental noise emulation and letter
The compressed data of road emulation carries out voice codec, to obtain the output voice of needs；Voice print database library module is used to acquisition
Simulated voice data according to voice and characteristic model data item and voice messaging data item, establish data training sample storehouse respectively
Library, and calling is provided according to authority data interface.
Beneficial effects of the present invention:
Application on Voiceprint Recognition of the invention emulates acquisition methods and its acquisition device with training dataset, is able to solve institute in the prior art
Existing to require amount of training data very big, operand is also very big when test, is often difficult to obtain in practical applications
Fine recognition effect, the problem of application training can not be directly carried out in data training process, the identification effect that can have been obtained
Fruit is able to carry out noise processed, and application training can be directly carried out in data training process, can be improved Application on Voiceprint Recognition environment
Noise robust and cha nnel robustness, the Application on Voiceprint Recognition that can be used under different communication modes is emulated with training dataset to be obtained.
Specifically, Application on Voiceprint Recognition of the invention training dataset, which emulates acquisition methods and acquisition device, has following two o'clock advantage:
Application on Voiceprint Recognition of the invention emulates acquisition methods and its acquisition device with training dataset, with traditional method for recognizing sound-groove phase
Than environmental noise emulation mode and information channel simulation method are applied in Application on Voiceprint Recognition, only need that the original of training will be used for
Beginning voice data carries out environmental noise mixing and communication pattern emulation, can obtain training identical with tested speech channel condition
Voice, to solve the problems, such as that environmental noise robustness and cha nnel robustness existing for traditional method for recognizing sound-groove are bad.
Application on Voiceprint Recognition of the invention emulates acquisition methods and its acquisition device with training dataset, knows with the vocal print across channel
Other modeling algorithm is compared, and only needs to carry out environmental noise emulation and communication pattern emulation to original trained speech samples, without
Need to change Application on Voiceprint Recognition modeling algorithm, to reduce the complexity of recognizer, while recognition effect is also than across channel
The more preferable environmental suitability of Application on Voiceprint Recognition modeling algorithm is more preferably；Therefore, more suitable for the foundation of Application on Voiceprint Recognition training sample database, symbol
Close the demand of engineering application.
Detailed description of the invention
Fig. 1 is the flow diagram that Application on Voiceprint Recognition training dataset of the invention emulates acquisition methods.
Fig. 2 is the structural block diagram that Application on Voiceprint Recognition training dataset of the invention emulates acquisition device.
Specific embodiment
The specific embodiment of the invention is described with reference to the accompanying drawings and embodiments:
Embodiment 1:
A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods, referring to Fig. 1, the acquisition methods include the side of channel coding
The side of method, the method for environmental noise emulation, the method for communication pattern emulation, data decoded method and vocal print database establishment
Method,
The method of channel coding the following steps are included:
Read primary voice data；
Header file is removed according to format standard according to the voice print database of the primary voice data, obtains pure speech data block；
Select the voice communication mode to be emulated；
Data volume is carried out to the obtained pure speech data block according to the voice communication mode corresponding speech coding standard
Code, obtains compressed voice data；
Environmental noise emulation method the following steps are included:
Select the environmental noise mode to be emulated；
The compressed voice number obtained according to the method for the varying environment noise and sound pressure levels of selection and channel coding
According to same channel mixing is carried out, the voice data comprising environmental noise is obtained；
Communication pattern emulation method the following steps are included:
Obtain the scrambling parameter that voice channel transmits under different relative amplitudes and state of signal-to-noise；
Scrambling parameter is selected, includes environmental noise to passing through described in coding, noise mixing in the method for environmental noise emulation
Voice data carries out Channel scrambling operation, obtains the Hybrid communication model simulated voice data comprising environmental noise；
The decoded method of data the following steps are included:
Corresponding tone decoding algorithm is selected according to voice communication mode；
The Hybrid communication model simulated voice data comprising environmental noise are added with corresponding tone decoding algorithm corresponding
Audio file head obtains trained voice identical with tested speech channel condition；
Voice print database construction method the following steps are included:
Data sample library is established according to voice and characteristic model data item；
Data sample library is established according to voice messaging data item；
Database interface is established according to voice print database interface specification.
The Application on Voiceprint Recognition of the present embodiment emulates acquisition methods with training dataset, first acquisition primary voice data, then
Environmental noise emulation is carried out to the voice print database of primary voice data, is then emulated in communication pattern, i.e. communication pattern soft simulation
When, it include first ring according to being mixed in the method that the communication pattern to be emulated emulates environmental noise by coding, noise
The voice data of border noise carries out voice coding；Then, after emulating scrambling parameter to coding according to respective communication mode lower channel
Voice data carry out communication pattern scramble simulation operations；Finally, carrying out language to the voice data after communication pattern scrambling emulation
Sound decoding operate obtains the voice under respective communication mode；Its by acquired original voice carry out communication pattern emulation so that
Training voice is identical with the channel condition of tested speech, solves channel mismatch phenomenon；It can effectively carry out voice communication
Channel soft simulation, the voice communication courses such as simulation fixed line, 2G, 3G, 4G, VOIP virtual speech and Internet chat, thus obtain with
The identical trained voice of tested speech channel condition, efficiently solves the problems, such as channel mismatch, is suitable for actual application demand.
It may also be preferred that the speech samples of the primary voice data are WAV, MP3 or ACC format.
It may also be preferred that the voice communication mode to be emulated includes fixed line, mobile phone in the method for channel coding
GSM, mobile TD-SDMA, connection WCDMA, telecommunications CDMA, LTE, 5G, recording pen or network virtual phone.
It may also be preferred that the environmental noise mode to be emulated includes interior in the method for environmental noise emulation
Environmental noise.
It may also be preferred that the indoor environment noise is machine operation noise in the method for environmental noise emulation.
It may also be preferred that the environmental noise mode to be emulated includes outdoor in the method for environmental noise emulation
Environmental noise.
It may also be preferred that the outdoor environment noise includes sound of the wind noise, the patter of rain in the method for environmental noise emulation
Noise, vehicle noise or machine operation noise.Here machine operation noise refers to what the mechanical equipment other than vehicle generated
Running noise.
Embodiment 2:
A kind of Application on Voiceprint Recognition as described in Example 1 emulates the acquisition device of acquisition methods, the acquisition dress with training dataset
It sets including voice acquisition module, voice coding module, environmental noise emulation module, communication pattern scrambling emulation module, sound solution
Code module and vocal print database module；The voice acquisition module, the voice coding module, the environmental noise emulate mould
Block, communication pattern scrambling emulation module, the voice codec module and the voice print database library module are successively linked in sequence.
The Application on Voiceprint Recognition of the present embodiment training dataset emulates acquisition device, and voice acquisition module is for acquiring speaker
Primary voice data voice print database；Voice coding module is used to carry out voice coder to the voice print database of primary voice data
Code, to obtain the compressed voice data under respective communication mode, i.e. vocoded data；Environmental noise emulation module is used
In the environmental noise data to vocoded data mixing selection, the voice data comprising environmental noise is obtained, that is, is realized different
Sound coding data under work condition environment；Communication pattern scrambles emulation module and selects respective channel, realization pair according to communication pattern
The communication pattern of sound coding data emulates；Voice codec module according to corresponding decoding algorithm, to by environmental noise emulation and
The compressed data of channel simulator carries out voice codec, to obtain the output voice of needs；Voice print database library module is used to obtaining
The simulated voice data obtained establish data training sample according to voice and characteristic model data item and voice messaging data item respectively
Warehouse, and calling is provided according to authority data interface.
It should be noted that the voice acquisition module, the voice coding module, the environmental noise emulation module,
The communication pattern scrambling emulation module, the voice codec module and the voice print database library module are successively suitable by data line
Sequence connection.
It may also be preferred that the voice acquisition module is set as recording pen.
It may also be preferred that the environmental noise emulation module is loudspeaker.
It should be noted that the voice coding module can be using pulse code modulation coding, i.e. pcm encoder.PCM is logical
Continuously varying analog signal is converted to digital coding by three oversampling, quantization, coding steps.
The communication pattern scrambling emulation module may refer to publishing house of BJ University of Aeronautics & Astronautics, 1 sunrise of September in 2007
" Communication System Simulation based on MATLAB " of version.
The voice codec module can be acquired and be handled to voice signal using audio chip, and audio coding decoding is calculated
Method is integrated in inside hardware, such as MP3 codec chip, speech synthesis analysis chip.Also it can use A/D capture card plus meter
Calculation machine forms hardware platform, and audio coding decoding algorithm is realized by the software on computer.Also A/D acquisition chip can be used
The acquisition for completing voice signal, the algorithm of Speech processing is realized using the strong chip of programmable data processing capacity,
Then it is controlled with ARM(Advanced RISC Machine) processor.
The voice print database library module may refer to " voice print database construction and application ", " first national audio-visual data
Test sensitivity technical conferences selected theses ", page number 609-611.
It should be noted that voiceprint, oice feature and model, refer to contained in voice, energy table
It seeks peace and identifies the phonetic feature of speaker, and the general name for the speech model established based on these features (parameter).It is collected
People, recording target object refer to the single natural person for being recorded voice.Vocal print acquisition, voice recording,
Refer to and acquire equipment using the vocal print of profession, according to certain operating process, acquisition meets the voice data of certain technical requirements
Process.Voice data, speech data refer to its for the voice and generation that people is collected obtained in vocal print collection process
His related data.Efficient voice, valid recorded speech refer to and belong to collected people in voice data and meet technology
The voice of parameter.Background noise, background noise refer in addition to mute, the part of non-effective voice.
For every collected people, the efficient voice duration acquired using different expression ways should meet the following conditions: chat
Predicate sound effective time is no less than 60 seconds；Reading voice effective time is read to be no less than 30 seconds.Reverberation time: when acquiring the reverberation in place
Between≤0.4 second.Noise: the ambient noise≤35dB in place is acquired.
Vocal print type can be following three kinds of situations: (1) basic data: basic data is the voice and feature that system prestores
Model data covers multilingual, more regions, multi channel feature.(2) sample data: sample data are unknown speakers
Voice and characteristic model data submit storage in such a way that inquiry compares by public security organs at different levels.(3) sample data: sample
Data are the voice and characteristic model data of known speaker, and storage is submitted by way of acquisition by public security organs at different levels.
User submits voice document to be detected, is converted to voiceprint after system is handled, touches in voice print database
Comparison is hit, collision comparison result is needed comprising four seed types: (1) sample and sample comparison result: sample audio to be checked and sample
Library comparison result.Such result can match clear identity, compare the part vocal print of highest scoring with audio to be checked, and then really
Personnel belonging to this fixed part vocal print.(2) sample and sample comparison result: sample audio to be checked and sample library comparison result.This
Class the result is that personal part belonging to unidentified sample audio to be checked, compared with the vocal print set for not yet identifying identity in sample library
Divide highest part vocal print, for matching not clear identity, but was registered in the past, has case information, the portion of highest scoring
Divide vocal print, and then determine whether the affiliated people in audio to be checked had case-involving history, carry out string and tracks down range to reduce.(3)
Sample and sample comparison result: sample to be examined audio and sample library comparison result.It is such the result is that personal part belonging to having identified
Not yet identify that the vocal print of identity compares the part vocal print of highest scoring in sample to be examined audio, with sample library, and then determination is to be checked
Whether affiliated people has case-involving history in audio.(4) sample to be examined audio and task sample database comparison result.It is such the result is that having known
The sample to be examined audio of personal part belonging to not is compared with the vocal print set for having identified affiliated personal part in sample database, thus
Whether people belonging to determining has multiple identities and case-involving history.
(1) the construction meaning of voice print database
With the management method of advanced technological means and science, processing, pipe are acquired to the voice messaging of speaker dependent
It ought to use, for solving criminal cases, fighting crime provides evidence, provides Informational support for work such as social security management, technological prevention.
(2) logic structure of data of voice print database
Establish volume of data resource management architecture and data back system, the basis as the system operation of entire vocal print library.
It should include following word bank in voice print database, the data structure in each library should meet " vocal print library data structure specification ", and
In the constraint of other database design specifications of national public safety field publication.
(3) personnel's sample database of voice print database
The comparison sample data submitted comprising acquisition system personal information collected and sample vocal print and user.Personnel's sample
Data in this library are corresponding with specific personal information, and carry out tissue according to multiple dimensions such as personnel's classification, personnel's attributes
And storage.
(4) the live sample library of voice print database
Live sample library is used to store the sample data relevant to case of user's submission, including the case voice being related to and case
The relevant background information of part.Live sample data should carry out category division according to case feature.
(5) the thematic special project library of voice print database
The needs for setting up special project, special topic in handling all kinds of cases in conjunction with business department, can establish more in voice print database
The other library of case, and will save in case-involving sample and sample set into the other library of case, play the work that word bank is divided in conjunction with business characteristic
With the precise alignment of realization small range data.Such as anti-terrorism special topic library, telephone fraud special topic library, ban taking addictive drugs thematic library, 75 special projects
Library etc..
(6) the basic vocal print library of voice print database
Basic vocal print is the data prestored in systems, covers multilingual, more regions, multi channel vocal print feature.Facilitate
Voiceprint analysis expert carries out technical research and study, simultaneously can be used for the self-optimization of vocal print comparison engine.
(7) systematic functional structrue of voice print database
Based on the abundant data of core data layer, need to establish corresponding service system, to the data in voice print database into
Row analysis, use, management.The kernel service of voice print database system includes kernel service layer, integrated application layer and data exchange
Interface.System architecture diagram refers to appendix A.
(8) the kernel service layer of voice print database
(1) voiceprint registration service
The registration engine that sample or sample voice document are registered as to vocal print is provided, passes through the side of service for upper-layer service system
Formula is called；
(2) vocal print compares service
It provides vocal print comparison service to call for upper-layer service system, by automating vocal print comparison engine, realize in voice print database
Data are checked in library, and return to the comparison result that can reduce data area, and wherein comparison result includes sample and sample, sample
With sample, sample and sample, sample and sample；
(3) vocal print management service
Management service, the service such as modification, deletion, inquiry including data are provided for the data stored in library.
Integrated application layer
(9) user management of voice print database
The login mode that voice print database system should be combined using user name encrypted code or PKI verifying, each login system
With there is a corresponding user account per family, and stringent permission control is carried out to it by role.
(10) rights management of voice print database
Voice print database user is related to various rolls and application terminal, and power should be strictly distinguished in the function and data processing of system
Limit only can use the function and data of system when Authority Verification passes through, and the operation log recording of user exists automatically
In system.
(1) operation monitoring
Operation monitoring includes the data volume of real-time display current system and the operating condition of task, and can pass through monitoring system
The operating status of the server nodes such as storage, calculating in real-time understanding system.
(2) data exchange interface
The voice print database of national public security organ should be built according to the requirement in portion, province's (city) two-level configuration, the province of various regions
(city) grade voice print database should meet the requirement of " voice print database access criteria ", be realized and national library by data exchange interface
Linkage, wherein the data information exchanged should include voiceprint report, compare mission dispatching.
The vocal print acquisition terminal of (11) voice print database
Sample acquisition system in voice print database should meet the requirement of " voiceprint acquisition technique specification ", pass through police network
It accesses vocal print and acquires equipment, realize the acquisition of voice print database and report.
The vocal print library server-side of (12) voice print database
Voice print database server-side is divided into registration subsystem according to functional characteristics, storage subsystem, compares subsystem.It considers
The data characteristics of vocal print and voice, each subsystem should use distributed storage and Distributed Computing Platform, can be according to data processing
The needs of amount, it is flexible to realize horizontal extension and vertical extension, and can support Single Point of Faliure transfer and data thermal backup, in list
When platform server breaks down, safeguards system can be operated normally.
The construction principle of (13) voice print database
(1) nurturing of network environment principle
National public security organ's voice print database system will be supported in Functional Design planning centered on the Ministry of Public Security, province's (city) grade is built
The two-stage framework in library, the Ministry of Public Security and province's (city) grade platform can access the terminal of service application units at different levels.The level knot of system
Composition refers to Appendix B.
(2) for national public security organ voice print database system installation and deployment in public security internal network, all application terminals are equal
Intranet access application system is crossed in public security Netcom.
The principles of planning design of (14) voice print database
The construction in vocal print library should meet following generic principles:
(1) reliability: the design of system should be using mature technology and equipment, to reduce technical risk；
(2) scalability: the design of system will have certain scalability, to meet the needs of business development from now on and constantly introduce newly
Technology, new equipment avoid disposable excess investment to improve the overall performance of system；
(3) safety: the master-plan of system will fully consider the security performance of system, prevention and dissolve technology risk；
(4) advanced and rational combination: information system correlative technology field must be taken into consideration in the master-plan of system
Development and status realize advanced and rational combination, should eliminate using bottleneck, limited fund is used again
Key aspect avoids unnecessary waste；
(5) open: system Construction should follow related international standard, and when network equipment type selecting should confirm both there is extensive manufacturer
With the support of standard, and meet the main trend of network technical development, and good technical support can be obtained.
The preferred embodiment for the present invention is explained in detail above in conjunction with attached drawing, but the present invention is not limited to above-mentioned implementations
Mode within the knowledge of a person skilled in the art can also be without departing from the purpose of the present invention
It makes a variety of changes.
Many other changes and remodeling can be made by not departing from the spirit and scope of the present invention.It should be appreciated that the present invention is not
It is limited to specific embodiment, the scope of the present invention is defined by the following claims.
Claims (10)
1. a kind of Application on Voiceprint Recognition emulates acquisition methods with training dataset, the acquisition methods include the method for channel coding, ring
Method, method, the method for data decoded method and vocal print database establishment of communication pattern emulation of border noise emulation, it is special
Sign is,
The method of channel coding the following steps are included:
Read primary voice data；
Header file is removed according to format standard according to the voice print database of the primary voice data, obtains pure speech data block；
Select the voice communication mode to be emulated；
Data volume is carried out to the obtained pure speech data block according to the voice communication mode corresponding speech coding standard
Code, obtains compressed voice data；
Environmental noise emulation method the following steps are included:
Select the environmental noise mode to be emulated；
The compressed voice number obtained according to the method for the varying environment noise and sound pressure levels of selection and channel coding
According to same channel mixing is carried out, the voice data comprising environmental noise is obtained；
Communication pattern emulation method the following steps are included:
Obtain the scrambling parameter that voice channel transmits under different relative amplitudes and state of signal-to-noise；
Scrambling parameter is selected, includes environmental noise to passing through described in coding, noise mixing in the method for environmental noise emulation
Voice data carries out Channel scrambling operation, obtains the Hybrid communication model simulated voice data comprising environmental noise；
The decoded method of data the following steps are included:
Corresponding tone decoding algorithm is selected according to voice communication mode；
The Hybrid communication model simulated voice data comprising environmental noise are added with corresponding tone decoding algorithm corresponding
Audio file head obtains trained voice identical with tested speech channel condition；
Voice print database construction method the following steps are included:
Data sample library is established according to voice and characteristic model data item；
Data sample library is established according to voice messaging data item；
Database interface is established according to voice print database interface specification.
2. Application on Voiceprint Recognition as described in claim 1 emulates acquisition methods with training dataset, which is characterized in that channel coding
In method, the speech samples of the primary voice data are WAV, MP3 or ACC format.
3. Application on Voiceprint Recognition as described in claim 1 emulates acquisition methods with training dataset, which is characterized in that channel coding
In method, the voice communication mode to be emulated includes fixed line, mobile phone GSM, mobile TD-SDMA, connection WCDMA, telecommunications
CDMA, LTE, 5G, recording pen or network virtual phone.
4. Application on Voiceprint Recognition as described in claim 1 emulates acquisition methods with training dataset, which is characterized in that environmental noise is imitative
In genuine method, the environmental noise mode to be emulated includes indoor environment noise.
5. Application on Voiceprint Recognition as claimed in claim 4 emulates acquisition methods with training dataset, which is characterized in that environmental noise is imitative
In genuine method, the indoor environment noise is machine operation noise.
6. Application on Voiceprint Recognition as described in claim 1 emulates acquisition methods with training dataset, which is characterized in that environmental noise is imitative
In genuine method, the environmental noise mode to be emulated includes outdoor environment noise.
7. Application on Voiceprint Recognition as claimed in claim 6 emulates acquisition methods with training dataset, which is characterized in that environmental noise is imitative
In genuine method, the outdoor environment noise includes sound of the wind noise, patter of rain noise, vehicle noise or machine operation noise.
8. such as the Application on Voiceprint Recognition of any of claims 1-7 acquisition device of training dataset emulation acquisition methods,
The acquisition device includes voice acquisition module, voice coding module, environmental noise emulation module, communication pattern scrambling emulation mould
Block, voice codec module and vocal print database module；It is characterized in that, the voice acquisition module, the voice coding module,
The environmental noise emulation module, communication pattern scrambling emulation module, the voice codec module and the voice print database
Library module is successively linked in sequence.
9. the acquisition device that Application on Voiceprint Recognition as claimed in claim 8 emulates acquisition methods with training dataset, which is characterized in that
The voice acquisition module is set as recording pen.
10. Application on Voiceprint Recognition as claimed in claim 8 emulates the acquisition device of acquisition methods with training dataset, feature exists
In the environmental noise emulation module is loudspeaker.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810895193.XA CN109192216A (en) | 2018-08-08 | 2018-08-08 | A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810895193.XA CN109192216A (en) | 2018-08-08 | 2018-08-08 | A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109192216A true CN109192216A (en) | 2019-01-11 |
Family
ID=64920502
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810895193.XA Pending CN109192216A (en) | 2018-08-08 | 2018-08-08 | A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109192216A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109920435A (en) * | 2019-04-09 | 2019-06-21 | 厦门快商通信息咨询有限公司 | A kind of method for recognizing sound-groove and voice print identification device |
| CN110390937A (en) * | 2019-06-10 | 2019-10-29 | 南京硅基智能科技有限公司 | A kind of across channel method for recognizing sound-groove based on ArcFace loss algorithm |
| CN110544469A (en) * | 2019-09-04 | 2019-12-06 | 秒针信息技术有限公司 | Training method and device of voice recognition model, storage medium and electronic device |
| CN110970035A (en) * | 2019-12-06 | 2020-04-07 | 广州国音智能科技有限公司 | Stand-alone speech recognition method, device and computer-readable storage medium |
| CN111341323A (en) * | 2020-02-10 | 2020-06-26 | 厦门快商通科技股份有限公司 | Voiceprint recognition training data amplification method and system, mobile terminal and storage medium |
| CN112802482A (en) * | 2021-04-15 | 2021-05-14 | 北京远鉴信息技术有限公司 | Voiceprint serial-parallel identification method, individual soldier system and storage medium |
| CN113160834A (en) * | 2021-04-27 | 2021-07-23 | 河南能创电子科技有限公司 | Low-voltage centralized reading, operation and maintenance implementation method based on AI intelligent voice recognition technology |
| CN113611328A (en) * | 2021-06-30 | 2021-11-05 | 公安部第一研究所 | Voiceprint recognition voice evaluation method and device |
| CN114070441A (en) * | 2021-12-27 | 2022-02-18 | 北京中安智能信息科技有限公司 | Underwater PCM signal receiving simulation system based on m-sequence coding |
| CN114783447A (en) * | 2022-04-21 | 2022-07-22 | 浙江大学 | Physical domain identity camouflage system and method based on adversarial samples for voiceprint recognition |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321387A (en) * | 2008-07-10 | 2008-12-10 | 中国移动通信集团广东有限公司 | Voiceprint recognition method and system based on communication system |
| CN105580071A (en) * | 2013-05-06 | 2016-05-11 | 谷歌技术控股有限责任公司 | Method and apparatus for training a voice recognition model database |
| CN106384588A (en) * | 2016-09-08 | 2017-02-08 | 河海大学 | Additive noise and short time reverberation combined compensation method based on vector Taylor series |
| CN106531155A (en) * | 2015-09-10 | 2017-03-22 | 三星电子株式会社 | Apparatus and method for generating acoustic model, and apparatus and method for speech recognition |
| CN106782565A (en) * | 2016-11-29 | 2017-05-31 | 重庆重智机器人研究院有限公司 | A kind of vocal print feature recognition methods and system |
| CN107481723A (en) * | 2017-08-28 | 2017-12-15 | 清华大学 | A channel matching method and device for voiceprint recognition |
-
2018
- 2018-08-08 CN CN201810895193.XA patent/CN109192216A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101321387A (en) * | 2008-07-10 | 2008-12-10 | 中国移动通信集团广东有限公司 | Voiceprint recognition method and system based on communication system |
| CN105580071A (en) * | 2013-05-06 | 2016-05-11 | 谷歌技术控股有限责任公司 | Method and apparatus for training a voice recognition model database |
| CN106531155A (en) * | 2015-09-10 | 2017-03-22 | 三星电子株式会社 | Apparatus and method for generating acoustic model, and apparatus and method for speech recognition |
| CN106384588A (en) * | 2016-09-08 | 2017-02-08 | 河海大学 | Additive noise and short time reverberation combined compensation method based on vector Taylor series |
| CN106782565A (en) * | 2016-11-29 | 2017-05-31 | 重庆重智机器人研究院有限公司 | A kind of vocal print feature recognition methods and system |
| CN107481723A (en) * | 2017-08-28 | 2017-12-15 | 清华大学 | A channel matching method and device for voiceprint recognition |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109920435A (en) * | 2019-04-09 | 2019-06-21 | 厦门快商通信息咨询有限公司 | A kind of method for recognizing sound-groove and voice print identification device |
| CN110390937B (en) * | 2019-06-10 | 2021-12-24 | 南京硅基智能科技有限公司 | Cross-channel voiceprint recognition method based on ArcFace loss algorithm |
| CN110390937A (en) * | 2019-06-10 | 2019-10-29 | 南京硅基智能科技有限公司 | A kind of across channel method for recognizing sound-groove based on ArcFace loss algorithm |
| CN110544469A (en) * | 2019-09-04 | 2019-12-06 | 秒针信息技术有限公司 | Training method and device of voice recognition model, storage medium and electronic device |
| CN110544469B (en) * | 2019-09-04 | 2022-04-19 | 秒针信息技术有限公司 | Training method and device of voice recognition model, storage medium and electronic device |
| CN110970035A (en) * | 2019-12-06 | 2020-04-07 | 广州国音智能科技有限公司 | Stand-alone speech recognition method, device and computer-readable storage medium |
| CN111341323A (en) * | 2020-02-10 | 2020-06-26 | 厦门快商通科技股份有限公司 | Voiceprint recognition training data amplification method and system, mobile terminal and storage medium |
| CN112802482A (en) * | 2021-04-15 | 2021-05-14 | 北京远鉴信息技术有限公司 | Voiceprint serial-parallel identification method, individual soldier system and storage medium |
| CN113160834A (en) * | 2021-04-27 | 2021-07-23 | 河南能创电子科技有限公司 | Low-voltage centralized reading, operation and maintenance implementation method based on AI intelligent voice recognition technology |
| CN113611328A (en) * | 2021-06-30 | 2021-11-05 | 公安部第一研究所 | Voiceprint recognition voice evaluation method and device |
| CN114070441A (en) * | 2021-12-27 | 2022-02-18 | 北京中安智能信息科技有限公司 | Underwater PCM signal receiving simulation system based on m-sequence coding |
| CN114070441B (en) * | 2021-12-27 | 2024-07-30 | 北京中安智能信息科技有限公司 | Underwater PCM signal receiving simulation system based on m-sequence coding |
| CN114783447A (en) * | 2022-04-21 | 2022-07-22 | 浙江大学 | Physical domain identity camouflage system and method based on adversarial samples for voiceprint recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109192216A (en) | A kind of Application on Voiceprint Recognition training dataset emulation acquisition methods and its acquisition device | |
| CN107222865B (en) | Communication swindle real-time detection method and system based on suspicious actions identification | |
| CN111951823B (en) | Audio processing method, device, equipment and medium | |
| US7716048B2 (en) | Method and apparatus for segmentation of audio interactions | |
| US20120303369A1 (en) | Energy-Efficient Unobtrusive Identification of a Speaker | |
| CN109473108A (en) | Auth method, device, equipment and storage medium based on Application on Voiceprint Recognition | |
| CN103377651B (en) | The automatic synthesizer of voice and method | |
| CN108833722A (en) | Audio recognition method, device, computer equipment and storage medium | |
| CN110232932A (en) | Method for identifying speaker, device, equipment and medium based on residual error time-delay network | |
| CN103078995A (en) | Customizable individualized response method and system used in mobile terminal | |
| CN109873907A (en) | Call processing method, device, computer equipment and storage medium | |
| CN107481723A (en) | A channel matching method and device for voiceprint recognition | |
| CN113539232B (en) | Voice synthesis method based on lesson-admiring voice data set | |
| CN112037772A (en) | Multi-mode-based response obligation detection method, system and device | |
| Yi et al. | Scenefake: An initial dataset and benchmarks for scene fake audio detection | |
| CN117037772A (en) | Voice audio segmentation method, device, computer equipment and storage medium | |
| CN106710591A (en) | Voice customer service system for power terminal | |
| KR102389995B1 (en) | Method for generating spontaneous speech, and computer program recorded on record-medium for executing method therefor | |
| CN103474062A (en) | Voice identification method | |
| Yi et al. | ADD 2023: Towards Audio Deepfake Detection and Analysis in the Wild | |
| CN117351948A (en) | Training method of voice recognition model, voice recognition method, device and equipment | |
| CN110298150A (en) | A kind of auth method and system based on speech recognition | |
| KR102395399B1 (en) | Voice data disassemble method for speech recognition learning, and computer program recorded on record-medium for executing method therefor | |
| KR20130073643A (en) | Group mapping data building server, sound recognition server and method thereof by using personalized phoneme | |
| CN113990288A (en) | Method and system for automatically generating and deploying speech synthesis model by speech customer service |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190111 |