CN110265040A - Training method, device, storage medium and the electronic equipment of sound-groove model - Google Patents
Training method, device, storage medium and the electronic equipment of sound-groove model Download PDFInfo
- Publication number
- CN110265040A CN110265040A CN201910537129.9A CN201910537129A CN110265040A CN 110265040 A CN110265040 A CN 110265040A CN 201910537129 A CN201910537129 A CN 201910537129A CN 110265040 A CN110265040 A CN 110265040A
- Authority
- CN
- China
- Prior art keywords
- age
- vocal print
- model
- user
- voice data
- 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.)
- Granted
Links
- 238000012549 training Methods 0.000 title claims abstract description 135
- 238000000034 method Methods 0.000 title claims abstract description 69
- 230000001755 vocal effect Effects 0.000 claims abstract description 192
- 230000006870 function Effects 0.000 claims description 43
- 230000004913 activation Effects 0.000 claims description 20
- 230000002618 waking effect Effects 0.000 claims description 15
- 230000007613 environmental effect Effects 0.000 claims description 9
- 239000000284 extract Substances 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 6
- 235000013399 edible fruits Nutrition 0.000 claims description 6
- 238000010200 validation analysis Methods 0.000 claims description 3
- 230000005055 memory storage Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 4
- 230000008569 process Effects 0.000 description 20
- 238000010586 diagram Methods 0.000 description 15
- 239000013598 vector Substances 0.000 description 14
- 238000013507 mapping Methods 0.000 description 9
- 238000012545 processing Methods 0.000 description 9
- 238000004891 communication Methods 0.000 description 7
- 238000000605 extraction Methods 0.000 description 6
- 230000003044 adaptive effect Effects 0.000 description 5
- 238000012790 confirmation Methods 0.000 description 5
- 238000012706 support-vector machine Methods 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 4
- 238000001914 filtration Methods 0.000 description 4
- 238000003062 neural network model Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 239000000203 mixture Substances 0.000 description 3
- 230000005236 sound signal Effects 0.000 description 3
- 230000037007 arousal Effects 0.000 description 2
- 235000019606 astringent taste Nutrition 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 239000004973 liquid crystal related substance Substances 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- SFRVOKMRHPQYGE-UHFFFAOYSA-N 1-(9h-fluoren-9-yl)ethyl carbonochloridate Chemical compound C1=CC=C2C(C(OC(Cl)=O)C)C3=CC=CC=C3C2=C1 SFRVOKMRHPQYGE-UHFFFAOYSA-N 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001052 transient effect Effects 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
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- 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
- G10L17/04—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
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Signal Processing (AREA)
- Electrically Operated Instructional Devices (AREA)
- User Interface Of Digital Computer (AREA)
Abstract
The embodiment of the present application discloses training method, device, storage medium and the electronic equipment of a kind of sound-groove model, belongs to field of computer technology.The described method includes: estimating the voice data progress age of user to obtain age estimation result, result corresponding vocal print universal model is estimated according to the age, the vocal print individual model of the user is obtained to model training, realize and vocal print wake-up is carried out using different vocal print individual models to the user of different age group.The vocal print individual model that the application obtains is related with the age of user, and the accuracy rate waken up can be improved by carrying out vocal print wake-up according to vocal print individual model.
Description
Technical field
This application involves speech processes field more particularly to a kind of training method of sound-groove model, device, storage medium and
Electronic equipment.
Background technique
Application on Voiceprint Recognition is the biological characteristic for having user according to sound, identifies a kind of identification technology of the identity of user.With
Traditional identity recognizing technology is compared, and the advantage of Application on Voiceprint Recognition is that extraction process is simple and at low cost, is widely used in various
Need to carry out the occasion of safeguard protection, such as: the financial institutions such as bank, security, insurance.Since voice is a kind of very random mistake
Journey, various internal factors or external factor are very big on pronunciation influence, and the acoustic feature for the voice data that such user issues can be sent out
Changing, the problem that existing awakening method can not adapt to variation to cause the accuracy rate waken up not high.
Summary of the invention
Training method, device, storage medium and the terminal for the sound-groove model that the embodiment of the present application provides, can solve not
With the user of the age level problem not high to the accuracy rate of equipment progress vocal print wake-up.The technical solution is as follows:
In a first aspect, the embodiment of the present application provides a kind of training method of sound-groove model, which comprises
Obtain the voice data of user；
The voice data progress age is estimated to obtain age estimation result；
The corresponding target vocal print universal model of the age estimation result is inquired in multiple vocal print universal models；Wherein,
The multiple vocal print universal model respectively corresponds to different age brackets, and the vocal print universal model user wakes up word identification；
Training unit, for being trained to obtain the user's to the target vocal print universal model according to voice data
Vocal print individual model；Wherein, whether the user identity of vocal print individual model voice data to be identified for identification is described
User.
Second aspect, the embodiment of the present application provide a kind of vocal print Rouser, and described device includes:
Acquiring unit, for obtaining the voice data of user；
Assessment unit obtains age estimation result for estimating the voice data progress age；
Query unit, it is logical for inquiring the corresponding target vocal print of the age estimation result in multiple vocal print universal models
Use model；Wherein, multiple vocal print universal models respectively correspond to different age brackets, and the vocal print universal model is for waking up word knowledge
Not；
Training unit, for being trained to obtain the use to the target vocal print universal model according to the voice data
The vocal print individual model at family；Wherein, the vocal print individual model for identification voice data to be identified user identity whether be
The user.
The third aspect, the embodiment of the present application provide a kind of computer storage medium, and the computer storage medium is stored with
A plurality of instruction, described instruction are suitable for being loaded by processor and executing above-mentioned method and step.
Fourth aspect, the embodiment of the present application provide a kind of electronic equipment, it may include: processor and memory；Wherein, described
Memory is stored with computer program, and the computer program is suitable for being loaded by the processor and being executed above-mentioned method step
Suddenly.
The technical solution bring beneficial effect that some embodiments of the application provide includes at least:
The voice data progress age of user is estimated to obtain age estimation as a result, estimating the corresponding sound of result according to the age
Line universal model is trained to obtain the vocal print individual model of the user, realizes and uses different sound to the user of different age group
Line individual model carries out vocal print wake-up, solves existing vocal print individual model and identifies wake-up caused by the vocal print of non-designated age bracket
The not high problem of accuracy rate, what the application can be adaptive selects suitable vocal print individual model to carry out according to the user of all ages and classes
Vocal print wakes up, and improves the accuracy rate that vocal print wakes up.
Detailed description of the invention
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of application for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the schematic diagram of speech control process provided by the embodiments of the present application；
Fig. 2 is the flow diagram of vocal print awakening method provided by the embodiments of the present application；
Fig. 3 is another flow diagram of vocal print awakening method provided by the embodiments of the present application；
Fig. 4 is the schematic diagram of sport career age estimation models provided by the embodiments of the present application；
Fig. 5 is the schematic diagram of trained vocal print individual model provided by the embodiments of the present application；
Fig. 6 is the schematic diagram provided by the embodiments of the present application for extracting acoustic feature；
Fig. 7 is the schematic diagram that a kind of vocal print provided by the embodiments of the present application wakes up；
Fig. 8 is a kind of structural schematic diagram of vocal print Rouser provided by the present application；
Fig. 9 is the structural schematic diagram of a kind of electronic equipment provided by the present application.
Specific embodiment
To keep the purposes, technical schemes and advantages of the application clearer, below in conjunction with attached drawing to the embodiment of the present application
Mode is described in further detail.
Firstly, to the invention relates to some nouns explain:
Gauss hybrid models (gaussianmixture model, GMM) are used for Gaussian probability-density function (normal distribution
Curve) accurately quantify things, it is one and things is decomposed into several models formed based on Gaussian probability-density function.
Unit vector (identityvector): the overall compact vector extracted from GMM mean value super vector.
Support vector machines (supportvectormachine): one kind carries out binary classification to data by supervised learning mode
Generalized linear classifier, decision boundary be to learning sample solve maximum back gauge hyperplane.
Age estimation models: for estimating the age of the corresponding user of voice data, age estimation models can be basis
Sample set is trained initial model parameter, the type of age estimation models can be neural network model,
Hidden Markov model or gauss hybrid models.
Vocal print individual model: for identification whether including preset wake-up word and according in voice data in voice data
Voiceprint carry out user identification confirmation, user identification confirmation is to identify whether the user for issuing voice data is default use
Family, voiceprint are analogous to a kind of biological characteristic of finger print information, and voiceprint has uniqueness, and different users has not
Same voiceprint, therefore can effectively distinguish different users.Neural network model, hidden horse can be used in vocal print individual model
Er Kefu model or gauss hybrid models.
It wakes up word identification model: whether including for identification preset wake-up word in voice data, preset wake-up word can
Be in electronic equipment it is preconfigured, can also be also that user is customized.Wherein, word identification model is waken up in identification voice number
While according to whether including preset wake-up word, text data can also be converted voice data into, and in voice control circle
Display text data on face.
Vocal print universal model: it with then initial sound-groove model, is obtained using the voice data sample training of different users
The sound-groove model arrived avoids training vocal print individual model from the beginning, improves the convergence rate of model training.
Vocal print wakes up: whether identification voice data includes keyword and the corresponding user of voice data is pre-set user, if
It is yes, activation voice control function (such as: activation voice assistant), starting recording carries out subsequent according to the voice data of recording
Voice control interaction.
Acoustic feature: one of amplitude information, phase information and spectrum information of voice or the features such as a variety of are indicated, no
Same voice data has different acoustic features.Such as: sound spectrograph can be used to indicate in acoustic feature.
Posterior probability: expression event has occurred and that, determine the event due to the size for a possibility that some factor causes, after
The calculating for testing probability will be based on prior probability.Posterior probability can be according to Bayesian formula be passed through, with prior probability and seemingly
Right function calculates.
Prior probability: expression event not yet occurs, and according to previous experiences and analyzes the probability for determining that the event occurs.
It is the schematic diagram that user carries out voice control to electronic equipment with reference to Fig. 1, Fig. 1.Wherein, the display screen of electronic equipment
In when putting out screen state, electronic equipment acquires the voice that user 100 issues, and converts speech into voice data, extracts voice number
Acoustic feature is input to vocal print individual model by the acoustic feature in, and vocal print individual model carries out vocal print to acoustic feature and calls out
It wakes up, whether identification user 100 is pre-set user, if it is, continue to judge in voice data whether to include wake-up word, if it is,
Voice control function is switched to state of activation, state of activation is kept into preset duration, and display screen is switched to and lights shape
State.
Wherein, in active state, electronic equipment 101 can receive the control voice of the sending of user 100, will control voice
Control instruction is converted to, the corresponding operation of control instruction is then executed.Such as: it makes a phone call to XX contact person, inquires weather, plays
The operations such as music, starting application program.Wherein, electronic equipment 101 can also convert voice data into text data, then exist
Show screen display this article notebook data.
Such as: shown in Figure 1, preset wake-up word is " XX is smart " in electronic equipment 101, and user 100 issues one section
Voice, electronic equipment collect voice and obtain voice data, and converting voice data into text data is " XX is smart ", and electronics is set
Standby determined according to vocal print individual model includes preset wake-up word in text data, then extracts the acoustic feature of voice data,
Acoustic feature is input to vocal print individual model, identifies that the user 100 for issuing voice data is default uses according to acoustic feature
Voice control function is activated at family, then shows that voice control interface 102, voice control interface 102 include microphone icon 103,
Voice control function is active, and microphone icon 103 is switched to Dynamically Announce, the Mike of Dynamically Announce by static status display
Wind map mark 103 is for prompting the voice control function of consumer electronic devices to be active, and electronic equipment 103 is in preset duration
The interior state of activation for keeping voice control function, beyond after preset duration, electronic equipment 103 is by voice control function by activation shape
State is switched to dormant state, while display screen is switched to screen state of putting out, while microphone icon 103 is shown using static mode
Show.In the case where putting out screen state, if user 100 needs to need to make to reactivate language in manner just described using voice control function
Sound control function, if electronic equipment 101 is under bright screen state, user 100 can click microphone icon 103 for voice control
Function switch processed is state of activation.
Wherein, electronic equipment 101 also on various communication customer end applications can be installed, such as: interactive voice application, view
Frequency records application, interactive voice application, searching class application, timely means of communication, mailbox client, social platform software etc..
Wherein, electronic equipment 101 can be hardware, be also possible to software.When electronic equipment 101 is hardware, can be
Various electronic equipments with display screen, including but not limited to smart phone, tablet computer, laptop portable computer and platform
Formula computer etc..When electronic equipment 101 is software, it can be and install in above-mentioned cited electronic equipment.It can be with
It realizes in multiple softwares or software module (such as: for providing Distributed Services), single software or software also may be implemented into
Module is not specifically limited herein.
When electronic equipment 101 is hardware, it is also equipped with display equipment thereon, it is real that display equipment can be various energy
The equipment of existing display function, such as: cathode-ray tube display (Cathode ray tubedisplay, abbreviation CR), luminous two
Pole pipe display (Light-emitting diode display, abbreviation LED), electronic ink screen, liquid crystal display (Liquid
Crystal display, abbreviation LCD), Plasmia indicating panel (Plasma displaypanel, abbreviation PDP) etc..User can
With using the display equipment on electronic equipment 101, come information such as the texts, picture, video of checking display.
It should be noted that the training of vocal print individual model provided by the embodiments of the present application is generally executed by electronic equipment,
Correspondingly, the training device of vocal print individual model is generally positioned in electronic equipment.
The embodiment of the present application provides a kind of training method of sound-groove model, and the training method of the sound-groove model can be applied
In electronic equipment.The electronic equipment can be smart phone, tablet computer, game station, AR
(AugmentedReality, augmented reality) equipment, automobile, data storage device, audio playing apparatus, video play device,
Notebook, Desktop computing device, wearable device such as electronic watch, electronic glasses, electronic helmet, electronic bracelet, electron term
The equipment such as chain, electronic clothes.
Below in conjunction with attached drawing 2- attached drawing 7, the training method of sound-groove model provided by the embodiments of the present application is carried out detailed
It introduces.Wherein, the training device of the vocal print individual model in the embodiment of the present application can be Fig. 2-electronic equipment shown in Fig. 7.
Fig. 2 is referred to, a kind of flow diagram of the training method of sound-groove model is provided for the embodiment of the present application.Such as figure
Shown in 2, the embodiment of the present application the method may include following steps:
S201, the voice data for obtaining user.
Wherein, for training vocal print individual model, voice data can be user and reads some spy the voice data of user
The voice data generated after fixed word, such as: the voice data that user reads aloud some specific character string or is digitally generated.With
After family issues voice, electronic equipment converts speech into the voice signal of analog form, audio collection by audio collecting device
Device can be single microphone, be also possible to the microphone array of multiple microphone compositions.Then, electronic equipment will simulate shape
The voice signal of formula obtains the voice data of digital form after pretreatment, and preprocessing process includes but is not limited to filter, put
Greatly, sampling, analog-to-digital conversion and format conversion.Voice data can with the voice data of lossless format, such as: the format of voice data
Are as follows: CD, WAV (wave file), FLAC (Free Lossless Audio Codec, Lossless Audio Compression coding) format etc..
Wherein, electronic equipment can acquire the multistage voice data of user, and it is identical that multistage voice data is all that user reads aloud
Content generate, guarantee the stability of the vocal print feature of user, improve the efficiency of model training.
Such as: electronic equipment can only handle the voice data of the WAV format for the monophonic that sample rate is 32kHz, and electronics is set
The voice data of the standby FLAC format that 16kHz is collected by single microphone, electronic equipment carry out interpolation according to voice data
Processing obtains the voice data that sample rate is 32kHz, and then FLAC format is converted to WAV format by electronic equipment, so as to electronics
Equipment carries out candidate vocal print and wakes up.
S202, the voice data progress age is estimated to obtain age estimation result.
Specifically, age estimation result can be a specific age value, it is also possible to an age bracket, i.e. most off year
Range between age and max age.Electronic equipment can be used the trained age estimation models of pre-selection and carry out to voice data
Age is estimated to obtain age estimation result.
In one embodiment, before the vocal print individual model of training user, electronic equipment can show that the age inputs
Frame, to prompt the user to input the age of oneself in age input frame, electronic equipment receives input in age input frame
Age is to know the age of the user.
In one embodiment, the process of age estimation includes: to carry out acoustic feature extraction, dimensionality reduction to voice data and gather
It closes, then loads the age estimation models of pre-stored SVM (SupportVectorMachine, support vector machines) type,
And according to the characteristic parameter of age estimation models and extraction identify voice data belonging to age bracket.
In one embodiment, the process of age estimation includes: to convert voice data into sound spectrograph, and each language is composed
Figure is input to age estimation models, to obtain the corresponding age bracket of sound spectrograph, which is age estimation result.
In one embodiment, the process of age estimation includes: that i-vector feature is extracted from voice data, will be mentioned
The i-vector feature taken out is matched with the target i-vector feature in age estimation models, and matching degree is maximum
The target i-vector feature corresponding age is determined as the corresponding age estimation result of voice data.
In one embodiment, the process of age estimation includes: that electronic equipment is provided with multiple age estimation models in advance,
Multiple age estimation models are to carry out model training using the training sample set of different age group to obtain respectively.Electronic equipment
It extracts voice data and obtains acoustic feature, be the voice signal in time domain since audio collecting device is collected, for the ease of
Voice signal is analyzed, voice signal in time domain can be converted into the voice signal on frequency domain, according to the language on frequency domain
Sound signal is pre-processed to obtain voice data, it will be understood that current voice data is the digital signal on frequency domain.Then,
Electronic equipment carries out feature extraction to voice data and obtains acoustic feature, and acoustic feature is input to multiple age estimation models point
An age estimation is not obtained as a result, since multiple age estimation models are carried out according to the training sample set of different age group
What model training obtained, thus have in multiple age estimation models and only one age estimation models output age estimation knot
Fruit is matched, electronic equipment needs determining matched age estimation result from estimation of multiple ages result.
In one embodiment, electronic equipment determines a most accurate age estimation knot from estimation of multiple ages result
The method of fruit may include:
Electronic equipment calculates the posterior probability of multiple age estimation results, and prior probability and pattra leaves can be used in posterior probability
This formula obtains.The posterior probability expression of age estimation result has carried out age estimation, is estimated using the obtained age
The probability of modified result age estimation result.The prior probability that age estimates result is indicated not yet to carry out age estimation, be calculated
Age estimates the probability that result occurs.Electronic equipment determines the maximum age estimation knot of posterior probability in multiple age estimation results
The maximum age estimation result of the posterior probability age the most final is estimated result by fruit.
Wherein, the method for determining one most accurate age estimation result may include: from estimation of multiple ages result
Acoustic feature is input to each age estimation models by electronic equipment, respectively obtains a feature vector, and electronics is set
It is pre-stored in standby or is pre-configured with label vector, then calculate separately the similarity between each feature vector and label vector,
The maximum feature vector of similarity is determined from multiple feature vectors, and the maximum feature vector of the similarity corresponding age is estimated
It surveys the age estimation result that model obtains and estimates result as the final age.
S203, selected target vocal print universal model corresponding with age estimation result.
Wherein, electronic equipment is pre-stored or is pre-configured with multiple vocal print universal models, and n is the integer greater than 1, multiple vocal prints
Universal model respectively corresponds to different age brackets, multiple vocal print universal models be according to different training sample set, it is different
Training sample respectively corresponds to different age brackets.The age bracket division of each different training sample set can be according to practical need
Depending on asking, the application is with no restriction.
Such as: electronic equipment is pre-configured or is pre-stored with 3 vocal print universal models, and 3 vocal print universal models are respectively as follows: sound
Line universal model 1, vocal print universal model 2 and vocal print universal model 3, vocal print universal model 1 are the instructions using 14 years old or less crowd
Practice what sample set training obtained, vocal print universal model 2 is obtained using the training sample set training of 14 years old~60 years old crowd
, vocal print individual model 3 is obtained using the training sample set training of 60 years old or more crowd.
Wherein, the mapping relations that electronic equipment is pre-stored or is pre-configured between vocal print universal model and age bracket, electronics
Equipment determines that the age that S202 is obtained estimates age bracket belonging to result according to the mapping relations.Such as: vocal print universal model and
Mapping relations between age bracket are as shown in table 1.
| Age bracket | Vocal print universal model |
| 14 years old or less | Vocal print universal model 1 |
| 14 years old~60 years old | Vocal print universal model 2 |
| 60 years old or more | Vocal print universal model 3 |
Table 1
S204, target vocal print universal model is trained according to voice data to obtain the vocal print individual model of user.
Wherein, whether the user identity of vocal print individual model voice data to be identified for identification is the user.This Shen
The voice data that multiple users please can be used is trained to obtain the user to the target vocal print universal model determined in S203
Vocal print individual model, such as: electronic equipment shows voice assistant interface, voice assistant interface prompt user A input 3 sections it is specified
The voice data of content, 3 sections of voice data that electronic equipment obtains input, which are trained target vocal print universal model, to be used
The vocal print individual model of family A.Electronic equipment carries out voice data to be identified using the vocal print individual model of trained user A
User identification confirmation.
In one embodiment, in the application when using the age of multiple age estimation models estimating subscriber's, electronics is set
The quantity of pre-set multiple age estimation models and the quantity of multiple vocal print universal models are equal in standby, and multiple years
Age estimation models compare the two with multiple vocal print individual models and the division of age bracket are also consistent.
Such as: the application is provided with 2 age estimation models: age estimation models 1 and age estimation models 2, the age is estimated
Model 1 is surveyed corresponding 14 years old hereinafter, age estimation models 2 are 14 years old corresponding or more；Meanwhile the application is provided with 2 vocal print Universal Dies
Type: vocal print universal model 1 and vocal print universal model 2, the correspondence of vocal print universal model 1 14 years old or more, vocal print universal model corresponding 14
Year old or more, it can be seen that, the division that the 2 vocal print universal models and 2 age estimation models of the application correspond to age bracket has been
It is complete consistent.
The scheme of the embodiment of the present application when being executed, estimates the voice data progress age of user to obtain age estimation knot
Fruit estimates the corresponding vocal print universal model of result according to the age and is trained to obtain the Application on Voiceprint Recognition model of the user, realization pair
The user of different age group carries out vocal print wake-up using different vocal print individual models, solves existing vocal print individual model identification
The not high problem of accuracy rate caused by the vocal print of non-designated age bracket, what the application can be adaptive selects according to the user of all ages and classes
It selects suitable vocal print individual model and carries out vocal print wake-up, improve the accuracy rate that vocal print wakes up.
Fig. 3 is referred to, a kind of flow diagram of the training method of sound-groove model is provided for the embodiment of the present application.This reality
It applies example and is applied to illustrate in electronic equipment with the training method of sound-groove model.The training method of the sound-groove model can wrap
Include following steps:
S301, it model training is respectively carried out to multiple age training sample set obtains multiple age estimation models.
Wherein, age estimation models use the age of the corresponding user of estimation voice data.Sport career age estimation models it
Before, an initial age estimation models can be preset, the parameters in initial age estimation models are initialized.Example
Such as: in neural network model, the biasing of neural network model and weights initialisation are 0.Electronic equipment is to pre-stored or prewired
The multiple age training sample set set respectively carry out model training and obtain multiple age estimation models, multiple age training samples
Set respectively corresponds to different age brackets.
It should be understood that multiple age estimation models are also possible to other equipment and train in addition to electronic equipment trains
Come, then trained multiple age estimation models are transplanted on the electronic equipment of the application.
Wherein, multiple age training sample set respectively correspond to different age brackets, can be not between all age group
Be overlapped, each age training sample set may include multiple voice data, the corresponding age estimation of voice data the result is that
Known, i.e., each voice data sample in age training sample set carries age label, each age training sample set
Depending on closing corresponding age bracket according to actual needs, division the application of specific age bracket is with no restriction.
For example: the training process schematic diagram of age estimation models shown in Figure 4, m=3,3 age training samples
This set is respectively age training sample set 40, age training sample set 43 and age training sample set 46, age instruction
Practicing the corresponding age bracket of voice data sample in sample set 40 is 14 years old or less (adult), language in age training sample set 43
The corresponding age bracket of sound data sample is 14 years old~60 years old (children), and voice data sample is corresponding in age training sample set 46
Age bracket be 60 years old or more (old man).Age training sample set 40 is obtained into age estimation models by age training 41
42, age training sample set 43 is obtained into age estimation models 45 by age training 44, by age training sample set 46
Age estimation models 48 are obtained by age training 44.
Wherein, the quantity of the voice data sample in each age training sample set can according to demand depending on.According to
The example of Fig. 4, includes 5000 voice data samples in age training sample set 40, and age training sample set 43 includes
5000 voice data samples, age training sample set 46 include 5000 voice data samples, each age training sample
The quantity of each age corresponding voice data sample is equal or roughly equal in set, age training sample set each in this way
In each age corresponding voice data sample quantity be it is balanced, avoid the language at some age in age training sample set
The problem of sound data sample is excessive or how much causes model training poor astringency.
S302, multiple wake-up word identification models are obtained to multiple wake-up word training sample set progress model training.
Wherein, wake up whether word model includes for identification preset wake-up word in voice data.Multiple wake-up word training
Sample set corresponds to different age brackets, can not be overlapped between different age brackets, in each wake-up word training sample set
It may include multiple voice data, each quantity for waking up the voice data for including in word training sample set can be identical,
It can not be identical.Each voice data waken up in word training sample set has recognition result label, recognition result label list
Show to include preset wake-up word and do not include preset two results of wake-up word in voice data.Each wake-up word training sample set
Each age, corresponding sample number was equal or roughly equal in conjunction, it is ensured that the equilibrium of the sample number at each age avoids some year
The problem of voice data sample size in age is excessive or the very few poor astringency for causing model training.
For example: it is shown in Figure 5, for it is provided by the present application wake up word identification model training process, n=3,3
Waking up word training sample set is respectively to wake up word training sample set 500, wake up word training sample set 503 and wake up word instruction
Practice sample set 506, waking up the corresponding age bracket of voice data sample in word training sample set 500 is 14 years old hereinafter, waking up
The corresponding age bracket of voice data sample is 14 years old~60 years old in word training sample set 503, wakes up word training sample set 506
The corresponding age bracket of middle voice data sample is 60 years old or more.The progress model training 501 of word training sample set 500 will be waken up to obtain
To word identification model 502 is waken up, the progress model training 504 of word training sample set 503 will be waken up and obtain waking up word identification model
505, the progress model training 507 of word training sample set 506 will be waken up and obtain waking up word identification model 508.
Wherein, the age estimation models in S301 and S302 and wake up word identification model type can be support vector machines,
Convolutional neural networks model and depth residue network model (such as: resnet etc.).
Such as: it is illustrated by taking convolutional neural networks model as an example, the specification of sound spectrograph is 112 × 112, and sound spectrograph is defeated
After entering into neural network, finally output is that the value of 2 nodes judges that age categories are if the value of first node is larger
Children；It if the value of second node is larger, can determine that age categories as adult, that is, execute a softmax operation.Convolutional Neural
There are many more function is finely tuned in network model, can be adjusted according to actual needs, such as: learning rate can be set to
0.001, activation primitive selects amendment linear unit.
Wherein, according to above-mentioned web results, model of the invention can be built with any a deep learning frame, such as may be used
Paddle including TensorFlow, caffe and Baidu is trained using above-mentioned frame using the sample to label,
Age estimation models are obtained after convergence or wake up word identification model.
In one embodiment, multiple age estimation models and multiple quantity for waking up word identification model are equal, i.e. m=n,
And the corresponding age bracket of multiple age estimation models age bracket corresponding with multiple wake-up word identification models is identical, for example, see figure
The division of the age bracket of 4 and Fig. 5 corresponds to 14 years old or less, 14 years old~60 years old and 60 years old three above age bracket, year is achieved
The quantity and age bracket of age estimation models and wake-up word identification model are consistent, and be can be further improved and are waken up the accurate of word identification
Rate.
S303, the voice data for obtaining user.
Wherein, sample of the voice data of user as the vocal print individual model of training user, the voice data of user can
It is that user reads aloud the voice data generated after specified content, such as: user A reads aloud the voice that " 0123456789 " generates afterwards, electricity
Sub- equipment collects the voice data that the user is generated after the voice.Electronic equipment can acquire the multistage voice data of user,
The content of multiple sections of voice data is identical, with the stability of the vocal print feature of the user of guarantee, improves the efficiency of model training.
In one embodiment, electronic equipment obtains the voice data of pre-stored user, which is number
Form, voice data can be lossless format, such as: the format of voice data can be FLEC, CD or WAV.
In one embodiment, electronic equipment obtains the voice data of user, audio collection dress by audio collecting device
The microphone array that can be a microphone or multiple microphones composition is set, each microphone pair in microphone array is passed through
An acquisition channel is answered, obtains the higher language of clarity by merging to collected voice signal on multiple acquisition channels
Sound signal, the collected voice signal of audio collecting device are analog forms, and electronic equipment is needed the language of the analog form
Sound signal is pre-processed to obtain the voice data of digital form.
Wherein, the process for the voice signal that electronic equipment acquires user's sending by audio collecting device further includes, to root
According to the duration of voice signal, voice signal is segmented, is divided into multiple speech frames.Such as: collected voice letter
Number duration be 6 seconds, then then can cutting obtain 61 second long speech frames.
S304, voice data progress speech feature extraction is obtained into acoustic feature.
Specifically, audio collecting device acquisition is voice signal in time domain, for the ease of dividing voice signal
Analysis, needs to be converted to the voice signal in time domain the voice signal on frequency domain, the acoustic feature of the application can be on frequency domain
Acoustic feature, sound spectrograph can be used to indicate in acoustic feature.
In one embodiment, shown in Figure 5, the method for extracting acoustic feature can be pretreatment, adding window, Fourier
Variation and MFCC are extracted, using the MFCC feature finally obtained as acoustic feature.Preprocessing process includes high-pass filtering, and electronics is set
Standby to carry out high-pass filtering to voice data using high-pass filter, the filtering performance expression formula of high-pass filter may is that H (z)
=1-a × z-1, a is correction factor, generally takes the numerical value between 0.95~0.97.Adding window is used for the edge of smooth signal, such as:
Using Hamming window to being to carry out windowing process after pretreatment, Hamming window is expressed asIts
In, n is integer, n=0,1,2 ..., M, M is the points of Fourier transformation.MFCC is extracted from the signal extraction after Fourier transformation
MFCC feature.Such as: use formulaWherein f is the frequency point after Fourier's variation.
In one embodiment, the MFCC feature extracted includes multiple MFCC characteristic components, different MFCC feature point
The different priority that measurer has can be by the 2nd component C in order to reduce the calculation amount and calculation delay of electronic equipment2With
16 component C16Between 15 components as final acoustic feature, reduce using component all in MFCC feature as acoustics
Data volume caused by feature is big and postpones high problem.
In one embodiment, it can will extract obtained acoustic feature to store, so as to subsequent use vocal print
People's model carries out vocal print wake-up, does not need to recalculate acoustic feature, reduces calculation amount.
S305, it acoustic feature is input to multiple age estimation models obtains multiple ages estimation results.
Wherein, multiple age estimation models are trained in S301, and multiple age estimation models respectively correspond different
Acoustic feature is input to multiple age estimation models respectively and obtains multiple age estimations as a result, multiple by age bracket, electronic equipment
Age estimates result may be identical, it is also possible to not identical.
S306, the posterior probability for calculating multiple age estimation results.
Wherein, electronic equipment calculates the posterior probability of multiple age estimation results, and the calculating of posterior probability can be according to existing
There is any one method of technology, such as: posterior probability can be calculated according to Bayesian formula and prior probability.
Such as: Bayesian formula can be usedTo calculate posterior probability, wherein middle λi
It is GMM (gaussian mixture model, gauss hybrid models) model parameter of i-th of age estimation models, X is input
Voice data.
S307, result is estimated using the maximum age estimation result of posterior probability as the final age.
S308, according to preset mapping relations inquire the age estimation result where age bracket.
Wherein, mapping relations indicate the mapping relations between age bracket and vocal print universal model.Such as: mapping relations such as table
Shown in 2:
| Age bracket | Vocal print universal model |
| 14 years old or less | Vocal print universal model 77 |
| 14 years old~60 years old | Vocal print universal model 78 |
| 60 years old or more | Vocal print universal model 79 |
Table 2
S309, the inquiry target vocal print universal model corresponding with age estimation result in multiple vocal print universal models.
For example, shown in Figure 7, determine the schematic illustration of target vocal print universal model, vocal print wakeup process
Include: the voice data 70 for obtaining user, acoustic feature is extracted in voice data, the acoustic feature extracted is input to year
Age identification model 72, age identification model 73 and age identification model 74, age estimation models 72 carry out year according to acoustic feature
Age estimates to obtain age estimation result 1, age estimation models 73 carry out the age according to acoustic feature and estimate to obtain age estimation knot
Fruit 2, the progress age of age estimation models 74 are estimated to obtain age estimation result 3.Result is estimated to three above-mentioned ages respectively
Posterior probability 75 is calculated, after the posterior probability for calculating age estimation result 1 obtains posterior probability 1, the calculating age estimates result 2
It tests probability and obtains posterior probability 2, the posterior probability for calculating age estimation result 3 obtains posterior probability 3, more above-mentioned 3 posteriority
The size of probability, it is assumed that the value of posterior probability 2 is maximum, and age estimation result 2 age the most final is estimated result.Electronics is set
The standby mapping relations (as shown in table 2) being stored between age bracket and vocal print universal model, electronic equipment is according to the mapping relations
Inquire corresponding target vocal print universal model, it is assumed that estimating result queries to target vocal print universal model is corresponded to according to the age is sound
Line universal model 78.
S310, target vocal print universal model is trained to obtain the vocal print individual of the user using the voice data of user
Model.
Wherein, the process of the vocal print individual model of electronic equipment training user can also include: that electronic equipment shows voice
Assistant interface, and when voice control function is active, in voice assistant interface display dialog box, prompt user's registration more
A voice data, the mesh that electronic equipment respectively obtains S310 as vocal print training sample set using multiple voice data of registration
Mark vocal print universal model carries out vocal print training and obtains the vocal print individual model of user.
For example, shown in Figure 5, the vocal print training sample set 509 of user A is inputted to vocal print universal model 502
It carries out vocal print training and obtains vocal print individual model 513；Or the vocal print training sample set 509 of input user A is to vocal print Universal Die
Type 505 carries out vocal print training and obtains vocal print wake-up module 513；Or the vocal print training sample set 509 of input user A is logical to vocal print
Vocal print training, which is carried out, with model 508 obtains vocal print individual model 515.
S311, corresponding target is selected to wake up word identification mould in multiple wake-up word identification models according to age estimation result
Type.
Wherein, multiple wake-up word identification models are to train to come in S302, and different wake-up word identification models is corresponding not
With age bracket, age estimation estimates result according to the age and corresponding target selected to wake up the result is that be calculated in S307
The process of word identification model, which can refer in S308, estimates the process that result selects the corresponding general sound-groove model of target according to the age,
Details are not described herein again.
S312, voice data to be identified is obtained.
Wherein, electronic equipment complete user vocal print individual model after training, can use vocal print individual model
Carry out vocal print wake-up.Electronic equipment can refer to S304's by the acoustic feature that audio collecting device obtains voice data to be identified
Description, details are not described herein again.
S312, identify to include preset wake-up word in voice data to be identified using target wake-up word identification model.
Wherein, if in voice data to be identified including preset wake-up word, S312 is executed, if in voice data to be identified not
Including preset wake-up word, electronic equipment continues to keep dormant state.
S313, voice data to be identified progress user is confirmed to obtain score value according to the vocal print individual model of user.
Wherein, Application on Voiceprint Recognition result is quantified using score value, and the size of score value is being preset in value range, such as: it is default
Value range is [0,1].
S314, judge whether score value is greater than threshold value.
Specifically, electronic equipment is pre-stored or is pre-configured with a threshold value, electronic equipment compare the score value that S313 is obtained and
The size of the threshold value executes S315, otherwise executes 316 if score value is greater than threshold value.The size of threshold value is the default value model of score value
Within enclosing, the size of threshold value is depending on actual needs.
S315, it wakes up successfully.
Wherein, waking up successfully indicates that the corresponding user of voice data to be identified is in pre-set user and voice data including pre-
If wake-up word.After waking up successfully, in the case where electronic equipment is in and puts out screen state, electronic equipment is switched to bright screen state, simultaneously will
Voice control function is switched to state of activation and keeps chance state preset duration；Electronic equipment is under bright screen state, electronics
Equipment keeps bright screen state, while voice control function is switched to state of activation and keeps state of activation preset duration.
Therein, it can be seen that electronic equipment realizes that second level arousal function, second level arousal function include waking up word to identify harmony
Line identification, wake up word identification be using target wake up word identification model to voice data to be identified carry out wake up word identify, if
Voice data to be identified includes preset wake-up word, then enters Application on Voiceprint Recognition, vocal print；If voice data to be identified does not include pre-
If wake-up word, then continue keep put out screen state.During Application on Voiceprint Recognition, electronic equipment treats knowledge according to personal sound-groove model
Other voice data carries out user identification confirmation, and the user identity of voice data to be identified is matched with the user, can be just switched to
Bright screen state, is switched to state of activation for voice control function and keeps state of activation preset duration.In some embodiments, electric
Sub- equipment includes digital signal processor (digital singnal processor, DSP) and application processor, wakes up word and knows
It can not executed by DSP, Application on Voiceprint Recognition can be executed by application processor.
In one embodiment, the method also includes:
When voice control function is active, voice control data are obtained；
The voice control data and preset reference environment noise data are compared, from the voice control data
In isolate environmental noise data；
It is compared, obtains matched with preset order word list from the voice control data of removal environmental noise data
Order word；
Execute the corresponding operation of the matched order word.
Wherein, it is pre-set user, voice that electronic equipment, which carries out the corresponding user of identification confirmation to voice control data,
Control data be for carrying out voice control to electronic equipment, such as: inquiry weather, the control such as navigate, make a phone call.Electronics is set
Standby to be pre-stored or be pre-configured with reference environment noise data, the relevant parameter of reference environment noise data can be electronics and set
It is standby gathered in advance.Electronic equipment, which can be used, carries out difference fortune for voice control data and preset reference environment noise data
The mode of calculation isolates environmental noise from voice control data.Electronic equipment is pre-stored or is pre-configured with order word list, orders
Each order word in word list is enabled to respectively correspond an operation, voice control data can be carried out text conversion by electronic equipment
Voice control text is obtained, and shows voice control text on a display screen, then compares voice control text and order word
Similarity in list between each order word takes the maximum order word of similarity as matched order word, executes the matching
The corresponding operation of order word.To sum up, by filtering out to wake-up noise in voice control data, effective voice can be obtained
Control instruction improves the success rate of voice control.
S316, failure is waken up.
Wherein, waking up unsuccessfully indicates that the corresponding user of voice data to be identified is not in pre-set user and/or voice data
It does not include preset wake-up word.
The scheme of the embodiment of the present application when being executed, to the voice of user carry out the age estimate to obtain age estimation as a result,
The corresponding vocal print universal model of result is estimated according to the age to be trained to obtain the vocal print individual model of the user, is realized to difference
The user of age bracket carries out vocal print wake-up using different vocal print individual models, solves the prior art and identifies non-designated age bracket
The not high problem of the accuracy rate of wake-up caused by vocal print, what the application can be adaptive selects suitable according to the user of all ages and classes
Vocal print individual model carries out vocal print wake-up, improves the accuracy rate that vocal print wakes up.
Following is the application Installation practice, can be used for executing the application embodiment of the method.It is real for the application device
Undisclosed details in example is applied, the application embodiment of the method is please referred to.
Fig. 8 is referred to, it illustrates the training devices for the vocal print individual model that one exemplary embodiment of the application provides
Structural schematic diagram, hereinafter referred to as training device 8.The training device 8 being implemented in combination with by software, hardware or both
As all or part of of terminal.Training device 8 includes acquiring unit 801, assessment unit 802, query unit 803 and training
Unit 804.
Acquiring unit 801, for obtaining the voice data of user.Acquiring unit 801 can be one or more microphones.
Assessment unit 802 obtains age estimation result for estimating the voice data progress age.
Query unit 803, for selected target vocal print universal model corresponding with age estimation result；Wherein, institute
Vocal print universal model is stated for waking up word identification.
Training unit 804, for being trained to obtain institute to the target vocal print universal model according to the voice data
State the vocal print individual model of user；Wherein, user body of the vocal print individual model for voice data to be identified for identification
Whether part is the user.
In one embodiment, further includes:
Selecting unit wakes up word identification model for selected target corresponding with age estimation result；Wherein, described
Target wakes up word identification model for waking up word identification.
In one embodiment, assessment unit 802 is used for:
The voice data progress age is estimated according to multiple age estimation models to obtain age estimation result；Wherein, described
Multiple age estimation models respectively correspond to different age brackets, and the age estimation models are used for according to voice data estimating subscriber's
Age.
In one embodiment, assessment unit 802 carries out age estimation to voice data according to multiple age estimation models
Obtain age estimation result, comprising:
Extract the acoustic feature in the voice data, and the storage acoustic feature；
The acoustic feature is input to multiple age estimation models and obtains multiple age estimation results；Wherein, Duo Genian
Age estimation models respectively correspond to different age brackets；
Calculate the multiple age estimation corresponding posterior probability of result；
Result is estimated using the maximum age estimation result of posterior probability as the final age.
In one embodiment, training device 8 further include:
Model training unit obtains multiple ages for respectively carrying out model training to multiple age training sample set and estimates
Survey model；Wherein, the multiple age training sample set respectively corresponds to different age brackets；And/or
Model training, which is respectively carried out, according to multiple wake-up word training sample set obtains multiple vocal print universal models；Wherein,
The multiple wake-up word training sample set respectively corresponds to different age brackets.
In one embodiment, training unit 804 is used for:
Obtain the pre-stored acoustic feature；
The acoustic feature is input to the target vocal print universal model to be trained to obtain the vocal print of the user
People's model.
In one embodiment, training device 8 further include:
Recognition unit, for obtaining voice data to be identified；
Word identification model, which is waken up, according to the target identifies to include preset wake-up word in the voice data to be identified
When, identity validation is carried out to the voice data to be identified according to the vocal print individual model；
If the user identity of the voice data to be identified is matched with the user, by voice control function by suspend mode shape
State is switched to state of activation；Or
If the user identity of the voice data to be identified and the user are unmatched, the voice control function is kept to be
Dormant state.
In one embodiment, there is display screen in training device 8, the recognition unit 8 is used for:
If the user identity of the voice data to be identified is matched with the user, when display screen is to put out screen state,
Bright screen state will be switched to by the screen state of putting out, and voice control function is switched to state of activation by dormant state, and
State of activation is remained into preset duration；Or
It is to put out screen state in display screen if the user identity of the voice data to be identified and the user are unmatched
When, continue to remain display screen and puts out screen state and keep voice control function for dormant state.
In one embodiment, training device 8 further include:
Execution unit, for obtaining voice control data when voice control function is active；
The voice control data and preset reference environment noise data are compared, from the voice control data
In isolate environmental noise data；
It is compared, obtains matched with preset order word list from the voice control data of removal environmental noise data
Order word；
Execute the corresponding operation of the matched order word.
It should be noted that training device 8 provided by the above embodiment is when executing vocal print awakening method, only with above-mentioned each
The division progress of functional module can according to need and for example, in practical application by above-mentioned function distribution by different function
Energy module is completed, i.e., the internal structure of equipment is divided into different functional modules, to complete whole described above or portion
Divide function.In addition, touch operation responding device provided by the above embodiment belong to touch operation response method embodiment it is same
Design embodies realization process and is detailed in embodiment of the method, and which is not described herein again.
Above-mentioned the embodiment of the present application serial number is for illustration only, does not represent the advantages or disadvantages of the embodiments.
The training device 8 of the application is estimated to obtain age estimation as a result, estimating result according to the age to the voice progress age
Corresponding vocal print individual model carries out vocal print wake-up to voice, realizes that the user to different age group is personal using different vocal prints
Model carries out vocal print wake-up, solves the prior art and identifies that the vocal print of non-designated age bracket causes the accuracy rate waken up is not high to ask
Topic, what the application can be adaptive selects suitable vocal print individual model to carry out vocal print wake-up according to the user of all ages and classes, improves
The accuracy rate that vocal print wakes up.
The embodiment of the present application also provides a kind of computer storage medium, the computer storage medium can store more
Item instruction, described instruction are suitable for being loaded by processor and being executed the method and step such as above-mentioned Fig. 2-embodiment illustrated in fig. 7, specifically hold
Row process may refer to Fig. 2-embodiment illustrated in fig. 7 and illustrate, herein without repeating.
Present invention also provides a kind of computer program product, which is stored at least one instruction,
At least one instruction is loaded as the processor and is executed the instruction to realize sound-groove model described in as above each embodiment
Practice method.
Fig. 9 is referred to, provides the structural schematic diagram of a kind of electronic equipment for the embodiment of the present application.As shown in figure 9, described
Electronic equipment 9 may include: at least one processor 901, at least one network interface 904, user interface 903, memory
905, at least one communication bus 902.
Wherein, communication bus 902 is for realizing the connection communication between these components.
Wherein, user interface 903 may include display screen (Display), camera (Camera), optional user interface
903 can also include standard wireline interface and wireless interface.
Wherein, network interface 904 optionally may include standard wireline interface and wireless interface (such as WI-FI interface).
Wherein, processor 901 may include one or more processing core.Processor 901 utilizes various excuses and line
Road connects the various pieces in entire terminal 900, by running or executing the instruction being stored in memory 905, program, code
Collection or instruction set, and the data being stored in memory 905 are called, execute the various functions and processing data of terminal 900.It can
Choosing, processor 901 can use Digital Signal Processing (Digital Signal Processing, DSP), field-programmable
Gate array (Field-Programmable Gate Array, FPGA), programmable logic array (Programmable
LogicArray, PLA) at least one of example, in hardware realize.Processor 901 can integrating central processor (Central
Processing Unit, CPU), in image processor (Graphics Processing Unit, GPU) and modem etc.
One or more of combinations.Wherein, the main processing operation system of CPU, user interface and application program etc.；GPU is for being responsible for
The rendering and drafting of content to be shown needed for display screen；Modem is for handling wireless communication.On it is understood that
Stating modem can not also be integrated into processor 901, be realized separately through chip piece.
Wherein, memory 905 may include random access memory (RandomAccess Memory, RAM), also may include
Read-only memory (Read-Only Memory).Optionally, which includes non-transient computer-readable medium
(non-transitory computer-readable storage medium).Memory 905 can be used for store instruction, journey
Sequence, code, code set or instruction set.Memory 905 may include storing program area and storage data area, wherein storing program area
Can store the instruction for realizing operating system, the instruction at least one function (such as touch function, sound play function
Energy, image player function etc.), for realizing instruction of above-mentioned each embodiment of the method etc.；Storage data area can store each above
The data etc. being related in a embodiment of the method.Memory 905 optionally can also be that at least one is located remotely from aforementioned processing
The storage device of device 901.As shown in figure 9, as may include operation system in a kind of memory 905 of computer storage medium
System, network communication module, Subscriber Interface Module SIM and vocal print wake up application program.
In electronic equipment 900 shown in Fig. 9, user interface 903 is mainly used for providing the interface of input for user, obtains
The data of user's input；And processor 901 can be used for calling the touch operation response application program stored in memory 905,
And specifically execute following operation:
Obtain the voice data of user；
The voice data progress age is estimated to obtain age estimation result；
Selected target vocal print universal model corresponding with age estimation result；
The vocal print for being trained to obtain the user to the target vocal print universal model according to the voice data is personal
Model；Wherein, whether the vocal print individual model is the user for the user identity of voice data to be identified for identification.
In one embodiment, processor 901, which executes, described carry out the age to the voice data and estimates to obtain the age and estimate
Survey result, comprising:
The voice data progress age is estimated according to multiple age estimation models to obtain age estimation result；Wherein, described
Multiple age estimation models respectively correspond to different age brackets, and the age estimation models are used for according to voice data estimating subscriber's
Age.
In one embodiment, processor 901 execute it is described according to multiple age estimation models to voice data carry out year
Age is estimated to obtain age estimation result, comprising:
Extract the acoustic feature in the voice data, and the storage acoustic feature；
The acoustic feature is input to multiple age estimation models and obtains multiple age estimation results；Wherein, Duo Genian
Age estimation models respectively correspond to different age brackets；
Calculate the multiple age estimation corresponding posterior probability of result；
Result is estimated using the maximum age estimation result of posterior probability as the final age.
In one embodiment, processor 901 is also used to execute:
Model training is respectively carried out to multiple age training sample set and obtains multiple age estimation models；Wherein, described
Multiple age training sample set respectively correspond to different age brackets；And/or
Model training, which is respectively carried out, according to multiple wake-up word training sample set obtains multiple vocal print universal models；Wherein,
The multiple wake-up word training sample set respectively corresponds to different age brackets.
In one embodiment, processor 901 execute it is described according to the voice data to the target vocal print Universal Die
Type is trained to obtain the vocal print individual model of the user, comprising:
Obtain the pre-stored acoustic feature；
The acoustic feature is input to the target vocal print universal model to be trained to obtain the vocal print of the user
People's model.
In one embodiment, processor 901 is also used to execute:
Obtain voice data to be identified；
Word identification model, which is waken up, according to the target identifies to include preset wake-up word in the voice data to be identified
When, identity validation is carried out to the voice data to be identified according to the vocal print individual model；
If the user identity of the voice data to be identified is matched with the user, by voice control function by suspend mode shape
State is switched to state of activation；Or
If the user identity of the voice data to be identified and the user are unmatched, the voice control function is kept to be
Dormant state.
In one embodiment, electronic equipment 9 further includes display screen, and processor 901 is used for:
It, will be by institute when to put out screen state if the user identity of the voice data to be identified is matched with the user
It states and puts out screen state and be switched to bright screen state, and voice control function is switched to state of activation by dormant state, and will activation
State remains preset duration；Or
It is to put out screen state in display screen if the user identity of the voice data to be identified and the user are unmatched
When, continue to remain display screen and puts out screen state and keep voice control function for dormant state.
In one embodiment, processor 901 is also used to execute:
When voice control function is active, voice control data are obtained；
The voice control data and preset reference environment noise data are compared, from the voice control data
In isolate environmental noise data；
It is compared, obtains matched with preset order word list from the voice control data of removal environmental noise data
Order word；
Execute the corresponding operation of the matched order word.
In one embodiment, processor 901 is also used to execute:
The file data is subjected to text conversion and generates text data；
Instruction display screen shows the text data on voice control interface；Wherein, the text data is in and can compile
The state of collecting.
In the embodiment of the present application, electronic equipment estimates to obtain age estimation as a result, root to the voice progress age of user
It is trained to obtain the vocal print individual model of the user according to the corresponding vocal print universal model of age estimation result, realize to not the same year
The user of age section carries out vocal print wake-up using different vocal print individual models, solves the sound that the prior art identifies non-designated age bracket
The problem that line causes the accuracy rate waken up not high, what the application can be adaptive selects suitable vocal print according to the user of all ages and classes
Individual model carries out vocal print wake-up, improves the accuracy rate that vocal print wakes up.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory or random access memory etc..
Above disclosed is only the application preferred embodiment, cannot limit the right model of the application with this certainly
It encloses, therefore according to equivalent variations made by the claim of this application, still belongs to the range that the application is covered.
Claims (11)
1. a kind of training method of sound-groove model, which is characterized in that the described method includes:
Obtain the voice data of user；
Estimate to obtain age estimation result based on the voice data progress age；
Selected target vocal print universal model corresponding with age estimation result；
The target vocal print universal model is trained to obtain the vocal print individual model of the user；Wherein, the vocal print
Whether the user identity of people's model voice data to be identified for identification is the user.
2. the method according to claim 1, wherein further include:
Selected target corresponding with age estimation result wakes up word identification model；Wherein, the target wakes up word and identifies mould
Type is for waking up word identification.
3. according to the method described in claim 2, it is characterized by further comprising:
Obtain voice data to be identified；
When being identified in the voice data to be identified according to target wake-up word identification model including preset wake-up word, root
Identity validation is carried out to the voice data to be identified according to the vocal print individual model；
If the user identity of the voice data to be identified is matched with the user, voice control function is cut by dormant state
It is changed to state of activation；Or
If the user identity of the voice data to be identified and the user mismatch, keeping voice control function is suspend mode shape
State.
4. according to the method described in claim 3, it is characterized in that, the method also includes:
When voice control function is active, voice control data are obtained；
The voice control data and preset reference environment noise data are compared, are divided from the voice control data
Separate out environmental noise data；
It is compared from the voice control data of removal environmental noise data with preset order word list, obtains matched order
Word；
Execute the corresponding operation of the matched order word.
5. the method according to claim 1, which is characterized in that described to carry out year to the voice data
Age is estimated to obtain age estimation result, comprising:
Extract the acoustic feature of the voice data, and the storage acoustic feature；
The voice data progress age is estimated according to multiple age estimation models using the acoustic feature to obtain age estimation knot
Fruit；Wherein, the multiple age estimation models respectively correspond to different age brackets, and the age estimation models are used for according to voice
Data estimation age of user.
6. according to the method described in claim 5, it is characterized in that, described logical to the target vocal print according to the voice data
It is trained to obtain the vocal print individual model of the user with model, comprising:
Obtain the pre-stored acoustic feature；
The acoustic feature is input to the target vocal print universal model to be trained to obtain vocal print individual's mould of the user
Type.
7. method according to claim 5 or 6, which is characterized in that it is described according to multiple age estimation models to voice number
Estimate to obtain age estimation result according to the age is carried out, comprising:
The acoustic feature is input to multiple age estimation models and obtains multiple age estimation results；Wherein, multiple ages are estimated
It surveys model and respectively corresponds to different age brackets；
Calculate the multiple age estimation corresponding posterior probability of result；
Result is estimated using the maximum age estimation result of posterior probability as the final age.
8. the method according to claim 1, wherein before the voice data for obtaining user, further includes:
Model training is respectively carried out to multiple age training sample set and obtains multiple age estimation models；Wherein, the multiple
Age training sample set respectively corresponds to different age brackets；And/or
Model training, which is respectively carried out, according to multiple wake-up word training sample set obtains multiple wake-up word identification models；Wherein, institute
It states multiple wake-up word training sample set and respectively corresponds to different age brackets.
9. a kind of training device of vocal print individual model, which is characterized in that described device includes:
Microphone, for obtaining the voice data of user；
Assessment unit obtains age estimation result for estimating the voice data progress age；
Query unit, for selecting the corresponding target vocal print universal model of the age estimation result；
Training unit, for being trained to obtain the user's to the target vocal print universal model according to the voice data
Vocal print individual model；Wherein, whether the user identity of vocal print individual model voice data to be identified for identification is described
User.
10. a kind of computer storage medium, which is characterized in that the computer storage medium is stored with a plurality of instruction, the finger
It enables and is suitable for being loaded by processor and being executed the method and step such as claim 1~8 any one.
11. a kind of electronic equipment characterized by comprising processor, memory and microphone；Wherein, the memory storage
There is computer program, the computer program is suitable for being loaded by the processor and being executed such as claim 1~8 any one
Method and step.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910537129.9A CN110265040B (en) | 2019-06-20 | 2019-06-20 | Voiceprint model training method and device, storage medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910537129.9A CN110265040B (en) | 2019-06-20 | 2019-06-20 | Voiceprint model training method and device, storage medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110265040A true CN110265040A (en) | 2019-09-20 |
| CN110265040B CN110265040B (en) | 2022-05-17 |
Family
ID=67919848
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910537129.9A Active CN110265040B (en) | 2019-06-20 | 2019-06-20 | Voiceprint model training method and device, storage medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110265040B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110797032A (en) * | 2020-01-06 | 2020-02-14 | 深圳中创华安科技有限公司 | Voiceprint database establishing method and voiceprint identification method |
| CN110956966A (en) * | 2019-11-01 | 2020-04-03 | 平安科技（深圳）有限公司 | Voiceprint authentication method, voiceprint authentication device, voiceprint authentication medium and electronic equipment |
| CN110995943A (en) * | 2019-12-25 | 2020-04-10 | 携程计算机技术（上海）有限公司 | Multi-user streaming voice recognition method, system, device and medium |
| CN111161728A (en) * | 2019-12-26 | 2020-05-15 | 珠海格力电器股份有限公司 | Awakening method, device, equipment and medium for intelligent equipment |
| CN111179944A (en) * | 2020-01-03 | 2020-05-19 | 达闼科技成都有限公司 | Voice awakening and age detection method and device and computer readable storage medium |
| CN111261196A (en) * | 2020-01-17 | 2020-06-09 | 厦门快商通科技股份有限公司 | Age estimation method, device and equipment |
| CN111369992A (en) * | 2020-02-27 | 2020-07-03 | Oppo（重庆）智能科技有限公司 | Instruction execution method and device, storage medium and electronic equipment |
| CN111524521A (en) * | 2020-04-22 | 2020-08-11 | 北京小米松果电子有限公司 | Voiceprint extraction model training method, voiceprint recognition method, voiceprint extraction model training device, voiceprint recognition device and voiceprint recognition medium |
| CN111599350A (en) * | 2020-04-07 | 2020-08-28 | 云知声智能科技股份有限公司 | Command word customization identification method and system |
| CN112669836A (en) * | 2020-12-10 | 2021-04-16 | 鹏城实验室 | Command recognition method and device and computer readable storage medium |
| CN112908312A (en) * | 2021-01-30 | 2021-06-04 | 云知声智能科技股份有限公司 | Method and equipment for improving awakening performance |
| CN113126859A (en) * | 2019-12-31 | 2021-07-16 | 宇龙计算机通信科技(深圳)有限公司 | Contextual model control method, contextual model control device, storage medium and terminal |
| CN113593584A (en) * | 2021-09-27 | 2021-11-02 | 深圳市羽翼数码科技有限公司 | Electronic product voice control system capable of effectively restraining response time delay |
| CN113593582A (en) * | 2021-06-24 | 2021-11-02 | 青岛海尔科技有限公司 | Control method and device of intelligent device, storage medium and electronic device |
| CN113707154A (en) * | 2021-09-03 | 2021-11-26 | 上海瑾盛通信科技有限公司 | Model training method and device, electronic equipment and readable storage medium |
| CN113948084A (en) * | 2021-12-06 | 2022-01-18 | 北京声智科技有限公司 | Voice data processing method and device, electronic equipment, storage medium and product |
| CN115083390A (en) * | 2021-03-10 | 2022-09-20 | Oppo广东移动通信有限公司 | Sound source distance sorting method and related product |
| CN117012189A (en) * | 2022-04-29 | 2023-11-07 | 荣耀终端有限公司 | Voice recognition method and electronic equipment |
| CN117975971A (en) * | 2024-04-02 | 2024-05-03 | 暨南大学 | Voiceprint age group estimation method and system based on privacy protection |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102968990A (en) * | 2012-11-15 | 2013-03-13 | 江苏嘉利德电子科技有限公司 | Speaker identifying method and system |
| CN103310788A (en) * | 2013-05-23 | 2013-09-18 | 北京云知声信息技术有限公司 | Voice information identification method and system |
| US9160849B1 (en) * | 2009-08-07 | 2015-10-13 | West Corporation | System, method, and computer-readable medium that facilitate voice biometrics user authentication |
| CN105489221A (en) * | 2015-12-02 | 2016-04-13 | 北京云知声信息技术有限公司 | Voice recognition method and device |
| CN105895105A (en) * | 2016-06-06 | 2016-08-24 | 北京云知声信息技术有限公司 | Speech processing method and device |
| CN108806696A (en) * | 2018-05-08 | 2018-11-13 | 平安科技（深圳）有限公司 | Establish method, apparatus, computer equipment and the storage medium of sound-groove model |
| CN108958810A (en) * | 2018-02-09 | 2018-12-07 | 北京猎户星空科技有限公司 | A kind of user identification method based on vocal print, device and equipment |
| CN109102813A (en) * | 2017-06-21 | 2018-12-28 | 北京搜狗科技发展有限公司 | Method for recognizing sound-groove, device, electronic equipment and storage medium |
| CN109545218A (en) * | 2019-01-08 | 2019-03-29 | 广东小天才科技有限公司 | Voice recognition method and system |
-
2019
- 2019-06-20 CN CN201910537129.9A patent/CN110265040B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9160849B1 (en) * | 2009-08-07 | 2015-10-13 | West Corporation | System, method, and computer-readable medium that facilitate voice biometrics user authentication |
| CN102968990A (en) * | 2012-11-15 | 2013-03-13 | 江苏嘉利德电子科技有限公司 | Speaker identifying method and system |
| CN103310788A (en) * | 2013-05-23 | 2013-09-18 | 北京云知声信息技术有限公司 | Voice information identification method and system |
| CN105489221A (en) * | 2015-12-02 | 2016-04-13 | 北京云知声信息技术有限公司 | Voice recognition method and device |
| CN105895105A (en) * | 2016-06-06 | 2016-08-24 | 北京云知声信息技术有限公司 | Speech processing method and device |
| CN109102813A (en) * | 2017-06-21 | 2018-12-28 | 北京搜狗科技发展有限公司 | Method for recognizing sound-groove, device, electronic equipment and storage medium |
| CN108958810A (en) * | 2018-02-09 | 2018-12-07 | 北京猎户星空科技有限公司 | A kind of user identification method based on vocal print, device and equipment |
| CN108806696A (en) * | 2018-05-08 | 2018-11-13 | 平安科技（深圳）有限公司 | Establish method, apparatus, computer equipment and the storage medium of sound-groove model |
| CN109545218A (en) * | 2019-01-08 | 2019-03-29 | 广东小天才科技有限公司 | Voice recognition method and system |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956966A (en) * | 2019-11-01 | 2020-04-03 | 平安科技（深圳）有限公司 | Voiceprint authentication method, voiceprint authentication device, voiceprint authentication medium and electronic equipment |
| CN110956966B (en) * | 2019-11-01 | 2023-09-19 | 平安科技（深圳）有限公司 | Voiceprint authentication method, voiceprint authentication device, voiceprint authentication medium and electronic equipment |
| CN110995943A (en) * | 2019-12-25 | 2020-04-10 | 携程计算机技术（上海）有限公司 | Multi-user streaming voice recognition method, system, device and medium |
| CN111161728A (en) * | 2019-12-26 | 2020-05-15 | 珠海格力电器股份有限公司 | Awakening method, device, equipment and medium for intelligent equipment |
| CN113126859A (en) * | 2019-12-31 | 2021-07-16 | 宇龙计算机通信科技(深圳)有限公司 | Contextual model control method, contextual model control device, storage medium and terminal |
| CN111179944A (en) * | 2020-01-03 | 2020-05-19 | 达闼科技成都有限公司 | Voice awakening and age detection method and device and computer readable storage medium |
| CN111179944B (en) * | 2020-01-03 | 2022-11-08 | 达闼机器人股份有限公司 | Voice awakening and age detection method and device and computer readable storage medium |
| CN110797032A (en) * | 2020-01-06 | 2020-02-14 | 深圳中创华安科技有限公司 | Voiceprint database establishing method and voiceprint identification method |
| CN111261196A (en) * | 2020-01-17 | 2020-06-09 | 厦门快商通科技股份有限公司 | Age estimation method, device and equipment |
| CN111369992A (en) * | 2020-02-27 | 2020-07-03 | Oppo（重庆）智能科技有限公司 | Instruction execution method and device, storage medium and electronic equipment |
| CN111599350A (en) * | 2020-04-07 | 2020-08-28 | 云知声智能科技股份有限公司 | Command word customization identification method and system |
| CN111524521A (en) * | 2020-04-22 | 2020-08-11 | 北京小米松果电子有限公司 | Voiceprint extraction model training method, voiceprint recognition method, voiceprint extraction model training device, voiceprint recognition device and voiceprint recognition medium |
| CN111524521B (en) * | 2020-04-22 | 2023-08-08 | 北京小米松果电子有限公司 | Voiceprint extraction model training method, voiceprint recognition method, voiceprint extraction model training device and voiceprint recognition device |
| CN112669836A (en) * | 2020-12-10 | 2021-04-16 | 鹏城实验室 | Command recognition method and device and computer readable storage medium |
| CN112669836B (en) * | 2020-12-10 | 2024-02-13 | 鹏城实验室 | Command recognition method and device and computer readable storage medium |
| CN112908312A (en) * | 2021-01-30 | 2021-06-04 | 云知声智能科技股份有限公司 | Method and equipment for improving awakening performance |
| CN112908312B (en) * | 2021-01-30 | 2022-06-24 | 云知声智能科技股份有限公司 | Method and equipment for improving awakening performance |
| CN115083390A (en) * | 2021-03-10 | 2022-09-20 | Oppo广东移动通信有限公司 | Sound source distance sorting method and related product |
| CN113593582B (en) * | 2021-06-24 | 2024-05-24 | 青岛海尔科技有限公司 | Control method and device of intelligent device, storage medium and electronic device |
| CN113593582A (en) * | 2021-06-24 | 2021-11-02 | 青岛海尔科技有限公司 | Control method and device of intelligent device, storage medium and electronic device |
| CN113707154A (en) * | 2021-09-03 | 2021-11-26 | 上海瑾盛通信科技有限公司 | Model training method and device, electronic equipment and readable storage medium |
| CN113707154B (en) * | 2021-09-03 | 2023-11-10 | 上海瑾盛通信科技有限公司 | Model training method, device, electronic equipment and readable storage medium |
| CN113593584A (en) * | 2021-09-27 | 2021-11-02 | 深圳市羽翼数码科技有限公司 | Electronic product voice control system capable of effectively restraining response time delay |
| CN113948084A (en) * | 2021-12-06 | 2022-01-18 | 北京声智科技有限公司 | Voice data processing method and device, electronic equipment, storage medium and product |
| CN117012189A (en) * | 2022-04-29 | 2023-11-07 | 荣耀终端有限公司 | Voice recognition method and electronic equipment |
| CN117975971A (en) * | 2024-04-02 | 2024-05-03 | 暨南大学 | Voiceprint age group estimation method and system based on privacy protection |
| CN117975971B (en) * | 2024-04-02 | 2024-07-05 | 暨南大学 | Voiceprint age group estimation method and system based on privacy protection |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110265040B (en) | 2022-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110265040A (en) | Training method, device, storage medium and the electronic equipment of sound-groove model | |
| CN110838286B (en) | Model training method, language identification method, device and equipment | |
| CN112259106B (en) | Voiceprint recognition method and device, storage medium and computer equipment | |
| CN110853618B (en) | Language identification method, model training method, device and equipment | |
| WO2021208287A1 (en) | Voice activity detection method and apparatus for emotion recognition, electronic device, and storage medium | |
| CN110310623B (en) | Sample generation method, model training method, device, medium, and electronic apparatus | |
| CN106486131B (en) | A kind of method and device of speech de-noising | |
| US11475897B2 (en) | Method and apparatus for response using voice matching user category | |
| CN102509547B (en) | Method and system for voiceprint recognition based on vector quantization based | |
| CN110288077A (en) | A kind of synthesis based on artificial intelligence is spoken the method and relevant apparatus of expression | |
| CN110808034A (en) | Voice conversion method, device, storage medium and electronic equipment | |
| CN110853617B (en) | Model training method, language identification method, device and equipment | |
| CN108281137A (en) | A kind of universal phonetic under whole tone element frame wakes up recognition methods and system | |
| CN110570873B (en) | Voiceprint wake-up method and device, computer equipment and storage medium | |
| CN111819625A (en) | System and method for language model personalization | |
| CN107257996A (en) | The method and system of environment sensitive automatic speech recognition | |
| CN108711429A (en) | Electronic equipment and apparatus control method | |
| CN102404278A (en) | Song requesting system based on voiceprint recognition and application method thereof | |
| CN112309365A (en) | Training method and device of speech synthesis model, storage medium and electronic equipment | |
| CN111312222A (en) | Awakening and voice recognition model training method and device | |
| CN107316635B (en) | Voice recognition method and device, storage medium and electronic equipment | |
| CN107610706A (en) | The processing method and processing unit of phonetic search result | |
| CN106341539A (en) | Automatic evidence obtaining method of malicious caller voiceprint, apparatus and mobile terminal thereof | |
| CN110136726A (en) | A kind of estimation method, device, system and the storage medium of voice gender | |
| CN110781327B (en) | Image searching method and device, terminal equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |