CN110033778A - One kind state of lying identifies update the system in real time - Google Patents
One kind state of lying identifies update the system in real time Download PDFInfo
- Publication number
- CN110033778A CN110033778A CN201910376615.7A CN201910376615A CN110033778A CN 110033778 A CN110033778 A CN 110033778A CN 201910376615 A CN201910376615 A CN 201910376615A CN 110033778 A CN110033778 A CN 110033778A
- Authority
- CN
- China
- Prior art keywords
- module
- word
- voice
- lying
- voice signal
- 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
- 238000004891 communication Methods 0.000 claims abstract description 64
- 238000012937 correction Methods 0.000 claims abstract description 51
- 238000012545 processing Methods 0.000 claims description 26
- 238000013135 deep learning Methods 0.000 claims description 14
- 238000000034 method Methods 0.000 claims description 13
- 239000004615 ingredient Substances 0.000 claims description 10
- 238000001514 detection method Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 8
- 230000004048 modification Effects 0.000 claims description 7
- 238000012986 modification Methods 0.000 claims description 7
- 230000009467 reduction Effects 0.000 claims description 7
- 230000000692 anti-sense effect Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 6
- 238000010801 machine learning Methods 0.000 claims description 6
- 230000000295 complement effect Effects 0.000 claims description 4
- 238000009432 framing Methods 0.000 claims description 4
- 239000003607 modifier Substances 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 230000008859 change Effects 0.000 description 5
- 230000005236 sound signal Effects 0.000 description 5
- 230000008921 facial expression Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000007405 data analysis Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000002996 emotional effect Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L17/00—Speaker identification or verification techniques
- G10L17/26—Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
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)
- Machine Translation (AREA)
Abstract
The present invention relates to one kind states of lying to identify update the system, including voice input module, sample classification module, semantics identity module, lie state recognition module, voice dictionary, mark module, correction module and output module in real time；The voice input module is for acquiring voice signal；The sample classification module communication is connected on voice input module；The semantics identity module communication is connected in sample classification module；The state recognition module communication of lying is connected in sample classification module；The voice dictionary communication connection is between semantics identity module and state recognition module of lying；The mark module respectively in semantics identity module and lie in state recognition module by communication connection；The correction module difference communication connection is in mark module and voice dictionary；The output module communication connection can identify and identify the confidence level of language in correction module, and correct and lead to the problem of new, and effective strong, the information of acquisition is accurate.
Description
Technical field
The present invention relates to voice and state-detections of lying, and in particular to one kind state of lying identifies update the system in real time.
Background technique
It usually will appear some lies in life, white lie can allow people to avoid falling into bigger trouble or danger
In, but the lie of malice can bring huge harm for other people or society.Lying is a kind of typical psychological condition, be will affect
Expressed to the behavior of people, facial expression, physiological change and language etc..For people higher for psychological quality, this shadow
Sound be it is extremely subtle, ordinary person is difficult to discover, thus their lie also can not be exposed quickly and efficiently.With science and technology
Development, partial information can accurately be collected by equipment, this lays the foundation for the machine recognition state of lying.
Since last century the nineties, equipment of detecting a lie has significant progress in China, and successfully auxiliary is tracked down
More than 2000 difficult cases.Therefore, this project is included in " eight or five " key project, the enforcement period of the ninth five-year plan by the Ministry of Public Security, and is put into
National science and technology research project.But equipment of so far, detecting a lie also rests on " device context acquisition information, artificial backstage judgement letter
The stage of breath ", poor in timeliness judge that the period is long, are easy to miss an opportunity because of a delay；When there is the lower information of confidence level, also it is not easy to
It is putd question to again for these information；And when judging, height is required to the specialized capability of personnel, cannot achieve extensive
It is universal.
Summary of the invention
The technical problem to be solved in the present invention is to provide one kind states of lying to identify update the system in real time, which can know
Not and mark language confidence level, and to the lower language of confidence level correct and targetedly lead to the problem of it is new,
Effective strong, the information of acquisition is accurate.
In order to solve the above-mentioned technical problems, the present invention provides one kind states of lying to identify update the system, including language in real time
Sound input module, sample classification module, speech recognition module, lie state recognition module, voice dictionary, mark module, amendment
Module and output module；The voice input module is for acquiring voice signal；The sample classification module communication is connected to language
On sound input module, the sample classification module is used to distinguish the voice signal of quizmaster He the person of being asked, and will after noise reduction
The voice signal for the person of being asked is transported to semantics identity module and state recognition module of lying；The semantics identity module communication connects
Connect in sample classification module, the meaning of one's words of semantics identity module meaning of one's words signal for identification, and by whole sentence split into word and
Word；The state recognition module communication of lying is connected in sample classification module, described to lie state recognition module for identification
The confidence level of meaning of one's words signal, and confidence level classification is carried out to each word, word, sentence and situation；The voice dictionary communication connection exists
Between semantics identity module and state recognition module of lying, the voice dictionary is for storing acoustic model and language model；Institute
Stating mark module, communication connection and is being lied in state recognition module at semantics identity module respectively, the mark module can when
The confidence levels of the words of semantics identity module output and state recognition module output of lying are coupled in sequence, to know each word
With the confidence levels of word；The correction module difference communication connection is in mark module and voice dictionary, the correction module
The lower verb of confidence levels and adjective for exporting mark module carry out antisense replacement, and confidence levels are lower
Whole sentence and situation modify；The output module communication connection is in correction module, and the output module is for defeated in real time
The voice signal of correction module out.
Preferably, the semantics identity module includes the successively Audio Processing Unit of communication connection, acoustical match list
Member and language processing unit；The Audio Processing Unit is used to extract the word of voice signal and the acoustic feature of word, and assigns every
A word and the unique label of word；The acoustical match unit is used to match the acoustic feature and voice word of Audio Processing Unit extraction
Acoustic feature in allusion quotation, to obtain the ingredient and paraphrase of each word and word；The language processing unit be used for according to syntactic structure,
The ingredient and paraphrase of language model correction word and word in semantic structure and voice dictionary.
Preferably, the ingredient is to include that subject, predicate, object, dynamic language, attribute, the adverbial modifier, complement and head exist
Interior sentence element；The paraphrase is specific meaning representated by word and word.
Preferably, the state recognition module of lying includes the pretreatment unit, rarefaction representation of successively communication connection
Unit, deep learning unit and dynamic sparse Bayesian time series modeling unit；The pretreatment unit be used for voice signal according to
Secondary progress framing, adding window, preemphasis, end-point detection and PCA dimension-reduction treatment；The rarefaction representation unit will be for that will pass through pretreatment
Voice signal afterwards was decomposed into complete dictionary and corresponding sparse coefficient matrix；The deep learning unit is the building of more hidden layers
Machine learning model, for extracting the depth structure information of voice signal；The dynamic sparse Bayesian time series modeling unit
For extracting the global feature information and minutia information of voice signal.
Preferably, the excessively complete sparse dictionary is the proper subspace of the characteristic set of voice signal；It is described dilute
Sparse coefficient matrix is the characteristic parameter of projection immanent structure of the voice signal on proper subspace.
Preferably, the method for the rarefaction representation of the rarefaction representation unit can be based on quick K-SVD sparse table
Show algorithm.
Preferably, the depth structure information includes that language feature, physiological characteristic, acoustic feature and psychological response are special
Sign.
One kind state of lying identifies update the system, including voice input module, sample classification module, speech recognition mould in real time
Block, state recognition module of lying, voice dictionary, mark module, correction module, output module, manual correction module, large database concept
With problem generation module；The voice input module is for acquiring voice signal；The sample classification module communication is connected to language
On sound input module, the sample classification module is used to distinguish the voice signal of quizmaster He the person of being asked, and will after noise reduction
The voice signal for the person of being asked is transported to semantics identity module and state recognition module of lying, and the voice of all inputs is believed
It number is transported in large database concept；The semantics identity module communication is connected in sample classification module, the semantics identity module
The meaning of one's words of meaning of one's words signal for identification, and whole sentence is split into word and word；The state recognition module communication of lying is connected to sample
In this categorization module, the confidence level of the state recognition module meaning of one's words signal for identification of lying, and to each word, word, sentence and
Situation carries out confidence level classification；The voice dictionary communication connection between semantics identity module and state recognition module of lying,
The voice dictionary is for storing acoustic model and language model；The mark module difference communication connection is in semantics identity module
With lie in state recognition module, the mark module can be coupled in timing semantics identity module output words and lie
The confidence levels of state recognition module output, to know the confidence levels of each word and word；The correction module is led to respectively
News are connected in mark module and voice dictionary, and the confidence levels that the correction module is used to export mark module are lower
Verb and adjective carry out antisense replacement, and the lower whole sentence of confidence levels and situation are modified；The output module is logical
News are connected in correction module, and the output module for exporting the voice signal of correction module and problem generation module in real time；
The large database concept respectively in sample classification module and lie in state recognition module by communication connection, and the large database concept is for depositing
The voice signal and the person's of the being asked information obtained from outside that storage generates during puing question to；The manual correction module communication connects
It connects between correction module and large database concept, the manual correction module can for the artificially voice signal of the modification person of being asked
Certainty；Described problem generation module communication connection between large database concept and output module, use by described problem generation module
In combining big data to extract the lower word of confidence levels, word, sentence and situation, and lead to the problem of it is corresponding after, by
Output module is exported to quizmaster.
Preferably, the person's of being asked information includes but is not limited to that personal essential information, activity time and range are believed
Breath, credit information and social relationships information.
Preferably, the large database concept includes raw data base and amendment database, the raw data base is used for
The voice messaging generated in storage dialog procedure, the amendment database are used to store the voice letter of the revised person of being asked
Breath, and retain the confidence level of all voice messagings.
Compared with prior art, the beneficial effects of the present invention are:
1, every words can saying to the person of being asked of the present invention carry out state-detection of lying, and mark each word,
The confidence level of word, sentence and situation, word lower for confidence level, word, sentence and situation are modified, and keep it more true
Property；And the higher information of confidence level can be exported in real time after the completion of modification, make quizmaster be easier to know which is just
True, which is wrong, the authenticity of the information of raising enquirement efficiency and acquisition；
2, the present invention can in conjunction with the person of being asked all relevant informations, for the lower word of confidence level, word, sentence and
Situation extracts, then generates corresponding new problem, is exported by output module to quizmaster, can be in system and outside system
Circulation is generated, the confidence level of all information is gradually increased, repeatedly after circulation, required all real informations can be obtained, obtained
Information quickly, accurately, saves a large amount of time, avoids the interference of error message.
Detailed description of the invention
Fig. 1 is the structural schematic diagram of embodiment 1；
Fig. 2 is the structural schematic diagram of embodiment 2.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Whole description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art without creative labor it is obtained it is all its
His embodiment, shall fall within the protection scope of the present invention.
Embodiment 1:
Shown in referring to Fig.1, present embodiment discloses one kind states of lying to identify update the system in real time, including voice inputs mould
Block, sample classification module, speech recognition module, state recognition module of lying, voice dictionary, mark module, correction module and defeated
Module out.
Voice input module:
Above-mentioned voice input module is microphone.Voice input module is for acquiring voice signal.
Sample classification module:
Above-mentioned sample classification module and voice input module communication connection.Sample classification module is used for quizmaster, is mentioned
The voice signal for the person of asking distinguishes, while carrying out noise reduction.The voice signal for the person of being asked that sample classification module extracts
It can be transported to semantics identity module simultaneously and lie in state recognition module.
Specifically, sample classification module can distinguish voice signal sample by way of manual key, timing and/or tone color
This.
Semantics identity module:
Above-mentioned semantics identity module is connected with sample classification module communication.Semantics identity module voice signal for identification
The meaning of one's words, and whole sentence is split into word and word.
Specifically, above-mentioned semantics identity module include successively the Audio Processing Unit of communication connection, acoustical match unit and
Language processing unit.
Above-mentioned Audio Processing Unit be used for extract the whole sentence of voice signal word and word acoustic feature, and assign each word and
The unique label of word.
Above-mentioned acoustical match unit is used to match the acoustics in the acoustic feature and voice dictionary of Audio Processing Unit extraction
Feature, to obtain the ingredient and paraphrase of each word and word.
Above-mentioned language processing unit is used for according to the language model correction word in syntactic structure, semantic structure and voice dictionary
With the ingredient and paraphrase of word.
Mentioned component be sentence including subject, predicate, object, dynamic language, attribute, the adverbial modifier, complement and head at
Point；Paraphrase is specific meaning representated by word and word.
It lies state recognition module:
Above-mentioned state recognition module of lying is connected with sample classification module communication.State recognition unit of lying language for identification
The confidence level of sound signal, and be classified.Status identification module lie when carrying out confidence level identification, classification, it can also be with
External sensor acquires facial expression information, emotional change information and the physiological change information of the person of being asked, further to mention
The accuracy of high ratings.
Specifically, above-mentioned state recognition module of lying include successively the pretreatment unit of communication connection, rarefaction representation unit,
Deep learning unit and dynamic sparse Bayesian time series modeling unit.
Above-mentioned pretreatment unit is for successively carrying out framing, adding window, preemphasis, end-point detection and PCA drop to voice signal
Dimension processing.
Above-mentioned rarefaction representation unit by pretreated voice signal for that will be decomposed into complete dictionary and corresponding
Sparse coefficient matrix.Cross the proper subspace for the characteristic set that complete sparse dictionary is voice signal；Sparse coefficient matrix is language
The characteristic parameter of projection immanent structure of the sound signal on proper subspace.Rarefaction representation unit can be by the original of voice signal
Data carry out rarefaction representation, and clear in structure, high-efficient, processing cost is low, and provides convenience for subsequent processing.Sparse table
Show that the method for the rarefaction representation of unit can be based on quick K-SVD rarefaction representation algorithm.
Above-mentioned deep learning unit is the machine learning model of more hidden layers building, for extracting the depth structure of voice signal
Information.Depth structure information includes language feature, physiological characteristic, acoustic feature and psychological response feature.Deep learning unit
Extracting method are as follows: the deep structure mapped using the good multidimensional data linear decomposition ability of rarefaction representation and multilayered nonlinear
Advantage completes approaching for complicated function；The input data of deep learning network is passed through by non-supervisory sparse autocoding algorithm
It crosses and gradually maps；Extract voice signal depth structure information.Deep learning unit improves the accuracy of speech data analysis
And timeliness.Machine learning model, which can be, utilizes signal data and sparse storehouse self-encoding encoder network (SSAE) model, structure
It builds thought and network structure is simple, operation is easily achieved.
Above-mentioned dynamic sparse Bayesian time series modeling unit is used to extract global feature information and the details spy of voice signal
Reference breath.The extracting method of dynamic sparse Bayesian time series modeling unit are as follows: dynamic by the network topology structure of sparse Bayesian
State reflects the case where psychological condition of lying changes over time；Thus the association in network between the corresponding feature of each variable is calculated to close
Sequential relationship between system and feature.The feature that dynamic sparse Bayesian time series modeling unit passes through extraction different time scales
The state of lying can be profoundly given expression to, Detection accuracy is high.
Voice dictionary:
Above-mentioned voice dictionary communication connection is between semantics identity module and state recognition module of lying.Voice dictionary is used for
Acoustic model and language model are stored, to provide data basis.
Mark module:
Above-mentioned mark module respectively in semantics identity module and lie in state recognition module by communication connection.Mark module energy
The confidence levels of the enough words that the output of semantics identity module is coupled in timing and state recognition module output of lying, to obtain
Know the confidence level of each word and each word.
Correction module:
Above-mentioned correction module difference communication connection is in mark module and voice dictionary.Correction module can be by mark module
The lower verb of the confidence levels of output and adjective carry out antisense replacement, by the lower whole sentence of confidence levels and situation into
Row modification, and corresponding voice signal is obtained from voice dictionary.
Output module:
Above-mentioned output module communication connection is in correction module.Output module for exporting the voice letter of correction module in real time
Number.The output signal of output module can be voice signal or letter signal.
The state of lying of the voice signal for the person of being asked accurately can identify and be identified by the structure optimized above, into
The authenticity of Huo Qu ti quizmaster's language when enabling quizmaster Shi to quizmaster is exported after row amendment, and then more targetedly
Continue put question to, to obtain true information, accuracy rate and timeliness are high.
Embodiment 2:
Referring to shown in Fig. 2, present embodiment discloses one kind states of lying to identify that update the system, including voice input mould in real time
Block, sample classification module, speech recognition module, lie state recognition module, voice dictionary, mark module, correction module, output
Module, manual correction module, large database concept and problem generation module.
Voice input module:
Above-mentioned voice input module is microphone.Voice input module is for acquiring voice signal.
Sample classification module:
Above-mentioned sample classification module and voice input module communication connection.Sample classification module is used for quizmaster, is mentioned
The voice signal for the person of asking distinguishes, while carrying out noise reduction.Sample classification module can be by the language of the person of being asked extracted
Sound signal is transported to semantics identity module and lies in state recognition module, and all voice signals of input are transported to greatly
In database.
Specifically, sample classification module can distinguish voice signal sample by way of manual key, timing and/or tone color
This.
Semantics identity module:
Above-mentioned semantics identity module is connected with sample classification module communication.Semantics identity module voice signal for identification
The meaning of one's words, and whole sentence is split into word and word.
Specifically, above-mentioned semantics identity module include successively the Audio Processing Unit of communication connection, acoustical match unit and
Language processing unit.
Above-mentioned Audio Processing Unit be used for extract the whole sentence of voice signal word and word acoustic feature, and assign each word and
The unique label of word.
Above-mentioned acoustical match unit is used to match the acoustics in the acoustic feature and voice dictionary of Audio Processing Unit extraction
Feature, to obtain the ingredient and paraphrase of each word and word.
Above-mentioned language processing unit is used for according to the language model correction word in syntactic structure, semantic structure and voice dictionary
With the ingredient and paraphrase of word.
Mentioned component be sentence including subject, predicate, object, dynamic language, attribute, the adverbial modifier, complement and head at
Point；Paraphrase is specific meaning representated by word and word.
It lies state recognition module:
Above-mentioned state recognition module of lying is connected with sample classification module communication.State recognition unit of lying language for identification
The confidence level of sound signal, and be classified.Status identification module lie when carrying out confidence level identification, classification, it can also be with
External sensor acquires facial expression information, emotional change information and the physiological change information of the person of being asked, further to mention
The accuracy of high ratings.
Specifically, above-mentioned state recognition module of lying include successively the pretreatment unit of communication connection, rarefaction representation unit,
Deep learning unit and dynamic sparse Bayesian time series modeling unit.
Above-mentioned pretreatment unit is for successively carrying out framing, adding window, preemphasis, end-point detection and PCA drop to voice signal
Dimension processing.
Above-mentioned rarefaction representation unit by pretreated voice signal for that will be decomposed into complete dictionary and corresponding
Sparse coefficient matrix.Cross the proper subspace for the characteristic set that complete sparse dictionary is voice signal；Sparse coefficient matrix is language
The characteristic parameter of projection immanent structure of the sound signal on proper subspace.Rarefaction representation unit can be by the original of voice signal
Data carry out rarefaction representation, and clear in structure, high-efficient, processing cost is low, and provides convenience for subsequent processing.Sparse table
Show that the method for the rarefaction representation of unit can be based on quick K-SVD rarefaction representation algorithm.
Above-mentioned deep learning unit is the machine learning model of more hidden layers building, for extracting the depth structure of voice signal
Information.Depth structure information includes language feature, physiological characteristic, acoustic feature and psychological response feature.Deep learning unit
Extracting method are as follows: the deep structure mapped using the good multidimensional data linear decomposition ability of rarefaction representation and multilayered nonlinear
Advantage completes approaching for complicated function；The input data of deep learning network is passed through by non-supervisory sparse autocoding algorithm
It crosses and gradually maps；Extract voice signal depth structure information.Deep learning unit improves the accuracy of speech data analysis
And timeliness.Machine learning model, which can be, utilizes signal data and sparse storehouse self-encoding encoder network (SSAE) model, structure
It builds thought and network structure is simple, operation is easily achieved.
Above-mentioned dynamic sparse Bayesian time series modeling unit is used to extract global feature information and the details spy of voice signal
Reference breath.The extracting method of dynamic sparse Bayesian time series modeling unit are as follows: dynamic by the network topology structure of sparse Bayesian
State reflects the case where psychological condition of lying changes over time；Thus the association in network between the corresponding feature of each variable is calculated to close
Sequential relationship between system and feature.The feature that dynamic sparse Bayesian time series modeling unit passes through extraction different time scales
The state of lying can be profoundly given expression to, Detection accuracy is high.
Voice dictionary:
Above-mentioned voice dictionary communication connection is between semantics identity module and state recognition module of lying.Voice dictionary is used for
Acoustic model and language model are stored, to provide data basis.
Mark module:
Above-mentioned mark module respectively in semantics identity module and lie in state recognition module by communication connection.Mark module energy
The confidence levels of the enough words that the output of semantics identity module is coupled in timing and state recognition module output of lying, to obtain
Know the confidence level of each word and each word.
Correction module:
Above-mentioned correction module difference communication connection is in mark module and voice dictionary.Correction module can be by mark module
The lower verb of the confidence levels of output and adjective carry out antisense replacement, by the lower whole sentence of confidence levels and situation into
Row modification, and corresponding voice signal is obtained from voice dictionary.
Output module:
Above-mentioned output module communication connection is in correction module.Output module is produced for exporting correction module and problem in real time
The voice signal of raw module.The output signal of output module can be voice signal or letter signal.
Large database concept:
Above-mentioned large database concept respectively in sample classification module and lie in state recognition module by communication connection.Large database concept is used
The voice signal generated during storage is putd question to and the person's of the being asked information obtained from outside.The person's of being asked information include but
It is not limited to personal essential information, activity time and range information, credit information and social relationships information.Large database concept can be language
Consciousness is not and state of lying provides data basis.
Above-mentioned large database concept includes raw data base and amendment database.Raw data base is produced for storing in dialog procedure
Raw voice messaging.Amendment database is used to store the voice messaging of the revised person of being asked, and retains all voice messagings
Confidence level.
Manual correction module:
Above-mentioned manual correction module communication is connected between correction module and large database concept.Manual correction module is for artificial
Modify the confidence level of the voice signal for the person of being asked.It, can be with when quizmaster can be assured that the answer of quizmaster is true
Increase confidence level, manually to accelerate question and answer process.
Problem generation module:
Above problem generation module communication connection is between large database concept and output module.Problem generation module is for combining
Big data extracts the lower word of confidence levels, word, sentence and situation, after generating corresponding new problem, by output mould
Block is exported to quizmaster.Problem generation module helps to obtain the higher information of confidence level.
The structure optimized above in system and can generate circulation outside system, gradually increase the confidence level of all information,
Repeatedly after circulation, required all real informations can be obtained.It obtains information quickly, accurately, saves a large amount of time, avoid
The interference of error message.
The foregoing description of the disclosed embodiments enables those skilled in the art to implement or use the present invention.
Various modifications to these embodiments will be readily apparent to those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, of the invention
It is not intended to be limited to the embodiments shown herein, and is to fit to consistent with principles disclosed herein and novel point
Widest scope
It should be noted that above-described embodiment, is not intended to limit the scope of protection of the present invention, in above-mentioned technical proposal
On the basis of made equivalents or substitution each fall within the range that the claims in the present invention are protected.
Claims (10)
1. one kind is lied, state identifies update the system in real time, which is characterized in that the update the system includes voice input module, sample
This categorization module, semantics identity module, state recognition module of lying, voice dictionary, mark module, correction module and output mould
Block；
The voice input module is for acquiring voice signal；
The sample classification module communication is connected on voice input module, the sample classification module for distinguish quizmaster and
The voice signal for the person of being asked, and the voice signal for the person of being asked is transported to semantics identity module and state of lying after noise reduction
Identification module；
The semantics identity module communication is connected in sample classification module, semantics identity module voice signal for identification
The meaning of one's words, and whole sentence is split into word and word；
The state recognition module communication of lying is connected in sample classification module, described to lie state recognition module for identification
The confidence level of voice signal, and confidence level classification is carried out to each word, word, sentence and situation；
Between semantics identity module and state recognition module of lying, the voice dictionary is used for the voice dictionary communication connection
Store acoustic model and language model；
The mark module respectively in semantics identity module and lie in state recognition module by communication connection, the mark module energy
The confidence levels of the enough words that the output of semantics identity module is coupled in timing and state recognition module output of lying, to know
The confidence levels of each word and word；
In mark module and voice dictionary, the correction module is used for mark module the correction module difference communication connection
The lower verb of the confidence levels of output and adjective carry out antisense replacement, by the lower whole sentence of confidence levels and situation into
Row modification；
The output module communication connection is in correction module, and the output module for exporting the voice letter of correction module in real time
Number.
2. state according to claim 1 of lying identifies update the system in real time, which is characterized in that the semantics identity module
Audio Processing Unit, acoustical match unit and language processing unit including successively communication connection；
The Audio Processing Unit is used to extract the word of voice signal and the acoustic feature of word, and assigns each word and word is unique
Label；
The acoustical match unit is used to match the acoustic feature in the acoustic feature and voice dictionary of Audio Processing Unit extraction,
To obtain the ingredient and paraphrase of each word and word；
The language processing unit is used for according to the language model correction word and word in syntactic structure, semantic structure and voice dictionary
Ingredient and paraphrase.
3. state according to claim 2 of lying identifies update the system in real time, which is characterized in that the ingredient be include master
Sentence element including language, predicate, object, dynamic language, attribute, the adverbial modifier, complement and head；The paraphrase is representated by word and word
Specific meaning.
4. state according to claim 1 of lying identifies update the system in real time, which is characterized in that the state recognition of lying
Module include successively the pretreatment unit of communication connection, rarefaction representation unit, deep learning unit and dynamic sparse Bayesian when
Sequence modeling unit；
The pretreatment unit is for successively carrying out at framing, adding window, preemphasis, end-point detection and PCA dimensionality reduction voice signal
Reason；
The rarefaction representation unit by pretreated voice signal for that will be decomposed into complete dictionary and corresponding sparse
Coefficient matrix；
The deep learning unit is the machine learning model of more hidden layers building, and the depth structure for extracting voice signal is believed
Breath；
The dynamic sparse Bayesian time series modeling unit is used to extract the global feature information and minutia letter of voice signal
Breath.
5. state according to claim 4 of lying identifies update the system in real time, which is characterized in that the excessively complete sparse word
Allusion quotation is the proper subspace of the characteristic set of voice signal；The sparse coefficient matrix is voice signal on proper subspace
Project the characteristic parameter of immanent structure.
6. state according to claim 4 of lying identifies update the system in real time, which is characterized in that the rarefaction representation unit
The method of rarefaction representation be based on quick K-SVD rarefaction representation algorithm.
7. state according to claim 4 of lying identifies update the system in real time, which is characterized in that the depth structure information
Including language feature, physiological characteristic, acoustic feature and psychological response feature.
8. one kind is lied, state identifies update the system in real time, which is characterized in that the update the system includes voice input module, sample
This categorization module, speech recognition module, state recognition module of lying, voice dictionary, mark module, correction module, output module,
Manual correction module, large database concept and problem generation module；
The voice input module is for acquiring voice signal；
The sample classification module communication is connected on voice input module, the sample classification module for distinguish quizmaster and
The voice signal for the person of being asked, and the voice signal for the person of being asked is transported to semantics identity module and state of lying after noise reduction
Identification module, and the voice signal of all inputs is transported in large database concept；
The semantics identity module communication is connected in sample classification module, semantics identity module voice signal for identification
The meaning of one's words, and whole sentence is split into word and word；
The state recognition module communication of lying is connected in sample classification module, described to lie state recognition module for identification
The confidence level of voice signal, and confidence level classification is carried out to each word, word, sentence and situation；
Between semantics identity module and state recognition module of lying, the voice dictionary is used for the voice dictionary communication connection
Store acoustic model and language model；
The mark module respectively in semantics identity module and lie in state recognition module by communication connection, the mark module energy
The confidence levels of the enough words that the output of semantics identity module is coupled in timing and state recognition module output of lying, to know
The confidence levels of each word and word；
In mark module and voice dictionary, the correction module is used for mark module the correction module difference communication connection
The lower verb of the confidence levels of output and adjective carry out antisense replacement, by the lower whole sentence of confidence levels and situation into
Row modification；
In correction module, the output module produces the output module communication connection for exporting correction module and problem in real time
The voice signal of raw module；
The large database concept respectively in sample classification module and lie in state recognition module by communication connection, and the large database concept is used
The voice signal generated during storage is putd question to and the person's of the being asked information obtained from outside；
The manual correction module communication is connected between correction module and large database concept, and the manual correction module is for artificial
Modify the confidence level of the voice signal for the person of being asked；
Described problem generation module communication connection is between large database concept and output module, and described problem generation module is for combining
Big data extracts the lower word of confidence levels, word, sentence and situation, and lead to the problem of it is corresponding after, by output mould
Block is exported to quizmaster.
9. state according to claim 8 of lying identifies update the system in real time, which is characterized in that the person's of being asked information
Including personal essential information, activity time and range information, credit information and social relationships information.
10. state according to claim 9 of lying identifies update the system in real time, which is characterized in that the large database concept packet
Raw data base and amendment database are included, the raw data base is described for storing the voice messaging generated in dialog procedure
Amendment database is used to store the voice messaging of the revised person of being asked, and retains the confidence level of all voice messagings.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376615.7A CN110033778B (en) | 2019-05-07 | 2019-05-07 | Real-time identification and correction system for lie state |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376615.7A CN110033778B (en) | 2019-05-07 | 2019-05-07 | Real-time identification and correction system for lie state |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110033778A true CN110033778A (en) | 2019-07-19 |
| CN110033778B CN110033778B (en) | 2021-07-23 |
Family
ID=67241449
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910376615.7A Active CN110033778B (en) | 2019-05-07 | 2019-05-07 | Real-time identification and correction system for lie state |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110033778B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2596351A (en) * | 2020-06-26 | 2021-12-29 | Lexiqal Ltd | Forensic interaction understanding: A system and method for determining the nature of spoken interactions using acoustic and linguistic markers |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20070276669A1 (en) * | 2006-05-25 | 2007-11-29 | Charles Humble | Quantifying psychological stress levels using voice patterns |
| CN101999902A (en) * | 2009-09-03 | 2011-04-06 | 上海天岸电子科技有限公司 | Voiceprint lie detector and voiceprint lie detecting method |
| CN102104676A (en) * | 2009-12-21 | 2011-06-22 | 深圳富泰宏精密工业有限公司 | Wireless communication device with lie detection function and lie detection method thereof |
| KR20120066276A (en) * | 2010-12-14 | 2012-06-22 | 유비벨록스(주) | Lie detecting response method and terminal device having the same |
| US20120262296A1 (en) * | 2002-11-12 | 2012-10-18 | David Bezar | User intent analysis extent of speaker intent analysis system |
| CN205493847U (en) * | 2016-03-04 | 2016-08-24 | 北京同方神火联合科技发展有限公司 | Pronunciation analytic system |
| CN105976820A (en) * | 2016-06-14 | 2016-09-28 | 上海质良智能化设备有限公司 | Voice emotion analysis system |
| CN107293302A (en) * | 2017-06-27 | 2017-10-24 | 苏州大学 | A kind of sparse spectrum signature extracting method being used in voice lie detection system |
| CN107452405A (en) * | 2017-08-16 | 2017-12-08 | 北京易真学思教育科技有限公司 | A kind of method and device that data evaluation is carried out according to voice content |
| CN107680602A (en) * | 2017-08-24 | 2018-02-09 | 平安科技（深圳）有限公司 | Voice fraud recognition methods, device, terminal device and storage medium |
| US20180160959A1 (en) * | 2016-12-12 | 2018-06-14 | Timothy James Wilde | Modular electronic lie and emotion detection systems, methods, and devices |
| WO2018135303A1 (en) * | 2017-01-18 | 2018-07-26 | ソニー株式会社 | Information processing device, information processing method, and program |
| CN108520753A (en) * | 2018-02-26 | 2018-09-11 | 南京工程学院 | Voice lie detection method based on the two-way length of convolution memory network in short-term |
-
2019
- 2019-05-07 CN CN201910376615.7A patent/CN110033778B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120262296A1 (en) * | 2002-11-12 | 2012-10-18 | David Bezar | User intent analysis extent of speaker intent analysis system |
| US20070276669A1 (en) * | 2006-05-25 | 2007-11-29 | Charles Humble | Quantifying psychological stress levels using voice patterns |
| CN101999902A (en) * | 2009-09-03 | 2011-04-06 | 上海天岸电子科技有限公司 | Voiceprint lie detector and voiceprint lie detecting method |
| CN102104676A (en) * | 2009-12-21 | 2011-06-22 | 深圳富泰宏精密工业有限公司 | Wireless communication device with lie detection function and lie detection method thereof |
| KR20120066276A (en) * | 2010-12-14 | 2012-06-22 | 유비벨록스(주) | Lie detecting response method and terminal device having the same |
| CN205493847U (en) * | 2016-03-04 | 2016-08-24 | 北京同方神火联合科技发展有限公司 | Pronunciation analytic system |
| CN105976820A (en) * | 2016-06-14 | 2016-09-28 | 上海质良智能化设备有限公司 | Voice emotion analysis system |
| US20180160959A1 (en) * | 2016-12-12 | 2018-06-14 | Timothy James Wilde | Modular electronic lie and emotion detection systems, methods, and devices |
| WO2018135303A1 (en) * | 2017-01-18 | 2018-07-26 | ソニー株式会社 | Information processing device, information processing method, and program |
| CN107293302A (en) * | 2017-06-27 | 2017-10-24 | 苏州大学 | A kind of sparse spectrum signature extracting method being used in voice lie detection system |
| CN107452405A (en) * | 2017-08-16 | 2017-12-08 | 北京易真学思教育科技有限公司 | A kind of method and device that data evaluation is carried out according to voice content |
| CN107680602A (en) * | 2017-08-24 | 2018-02-09 | 平安科技（深圳）有限公司 | Voice fraud recognition methods, device, terminal device and storage medium |
| CN108520753A (en) * | 2018-02-26 | 2018-09-11 | 南京工程学院 | Voice lie detection method based on the two-way length of convolution memory network in short-term |
Non-Patent Citations (5)
| Title |
|---|
| M.E. GADALLAH: "Speech based automatic lie detection", 《PROCEEDINGS OF THE SIXTEENTH NATIONAL RADIO SCIENCE CONFERENCE. NRSC"99 (IEEE CAT. NO.99EX249)》 * |
| 周燕: "基于语音稀疏表示的谎言检测方法", 《中国优秀博士学位论文全文数据库信息科技辑》 * |
| 潘欣裕: "谎言语音检测分析研究", 《中国优秀博士学位论文全文数据库信息科技辑》 * |
| 范程: "基于深度置信网络的汉语语音测谎研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 赵力: "语音测谎技术研究现状与展望", 《数据采集与处理》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2596351A (en) * | 2020-06-26 | 2021-12-29 | Lexiqal Ltd | Forensic interaction understanding: A system and method for determining the nature of spoken interactions using acoustic and linguistic markers |
| GB2596351B (en) * | 2020-06-26 | 2022-10-19 | Lexiqal Ltd | Forensic interaction understanding: A system and method for determining the nature of spoken interactions using acoustic and linguistic markers |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110033778B (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110717018A (en) | Industrial equipment fault maintenance question-answering system based on knowledge graph | |
| CN110910283A (en) | Method, device, equipment and storage medium for generating legal document | |
| CN111694940B (en) | User report generation method and terminal equipment | |
| CN1170239C (en) | Palm acoustic-print verifying system | |
| CN107731233A (en) | A kind of method for recognizing sound-groove based on RNN | |
| CN104538034A (en) | Voice recognition method and system | |
| CN109410911A (en) | Artificial intelligence learning method based on speech recognition | |
| CN112712349A (en) | Intelligent paperless conference data information processing method based on artificial intelligence and big data analysis | |
| CN106971737A (en) | A kind of method for recognizing sound-groove spoken based on many people | |
| CN111581348A (en) | Query analysis system based on knowledge graph | |
| CN111627448A (en) | System and method for realizing trial and talk control based on voice big data | |
| CN117198338B (en) | Interphone voiceprint recognition method and system based on artificial intelligence | |
| CN106934870A (en) | A kind of voice attendance checking system | |
| CN110033778A (en) | One kind state of lying identifies update the system in real time | |
| CN106971730A (en) | A kind of method for recognizing sound-groove based on channel compensation | |
| CN105160229A (en) | Single-soldier system with voice and fingerprint dual authentication | |
| CN109325155A (en) | A kind of novel dialogue state storage method and system | |
| CN109003613A (en) | The Application on Voiceprint Recognition payment information method for anti-counterfeit of combining space information | |
| CN113823267A (en) | Depression automatic identification method and device based on speech recognition and machine learning | |
| CN113053416A (en) | Speech emotion recognition system based on knowledge graph | |
| Lashkari et al. | NMF-based cepstral features for speech emotion recognition | |
| CN110910904A (en) | Method for establishing voice emotion recognition model and voice emotion recognition method | |
| CN110265003A (en) | A kind of method of voiced keyword in identification broadcast singal | |
| Ma et al. | Computer Intelligent Recognition Image Speech Signal Application | |
| CN112820274B (en) | Voice information recognition correction method and system |
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 |