CN110428846A - Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network - Google Patents
Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network Download PDFInfo
- Publication number
- CN110428846A CN110428846A CN201910609734.2A CN201910609734A CN110428846A CN 110428846 A CN110428846 A CN 110428846A CN 201910609734 A CN201910609734 A CN 201910609734A CN 110428846 A CN110428846 A CN 110428846A
- Authority
- CN
- China
- Prior art keywords
- voice
- over
- net
- steganography
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 230000002457 bidirectional effect Effects 0.000 title claims abstract description 33
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 32
- 239000013598 vector Substances 0.000 claims abstract description 71
- 238000012360 testing method Methods 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000013139 quantization Methods 0.000 claims abstract description 38
- 238000001514 detection method Methods 0.000 claims abstract description 25
- 239000000284 extract Substances 0.000 claims abstract description 4
- 238000003780 insertion Methods 0.000 claims description 12
- 230000037431 insertion Effects 0.000 claims description 12
- 238000004458 analytical method Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005303 weighing Methods 0.000 claims description 3
- 230000007774 longterm Effects 0.000 claims 1
- 230000006835 compression Effects 0.000 description 6
- 238000007906 compression Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000005540 biological transmission Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007787 long-term memory Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/018—Audio watermarking, i.e. embedding inaudible data in the audio signal
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Human Computer Interaction (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Multimedia (AREA)
- Acoustics & Sound (AREA)
- Signal Processing (AREA)
- Compression, Expansion, Code Conversion, And Decoders (AREA)
Abstract
The invention discloses a kind of voice-over-net stream steganalysis methods and device based on bidirectional circulating neural network, wherein this method comprises: S1, obtains voice-over-net stream training set sample and voice-over-net current test collection sample；S2 is extracted voice-over-net load from voice-over-net stream training set sample by sliding window method and is handled to obtain quantization index vector；S3 extracts the code word associated vector in quantization index vector using code word correlation model；S4 classifies to code word associated vector by tagsort model；S5 is iterated step S2-S4 by measure of supervision, generates Stego-detection model；S6, by voice-over-net current test collection sample input Stego-detection model detected to obtain steganography probability, according to steganography probabilistic determination voice-over-net current test collection sample whether steganography.This method can quickly and efficiently judge whether raw tone stream passes through information steganography.
Description
Technical field
It is the present invention relates to digital audio Stego-detection technical field, in particular to a kind of based on bidirectional circulating neural network
Voice-over-net stream steganalysis method and device.
Background technique
Cryptography and Information Hiding Techniques are the Liang Ge branches of information security field.Cryptography is mainly with information theory and close
Based on code theory, says that raw information carries out scramble by certain mode, obtain ciphertext.Information steganography is then to believe secret
Breath is embedded in initial carrier according to certain way, under the premise of not discovered by third party, sends containing information to reception
Side.Compared to cryptography mode, the mode of Information hiding is not easy to cause third-party to be attacked note that can significantly reduce by third party
The probability hit, safety greatly improve.Also just because of this, Information hiding has obtained the concern of many people.
Steganography and steganalysis are two importances of Information hiding.Steganography is in order to ensure the information of communicating pair is pacified
Entirely.Steganalysis is then to judge whether there is encrypted message in carrier.Since steganography has high confidentiality, hold very much
It is easily utilized by criminal, to endanger public security.Then, effective Steganalysis can be examined effectively rapidly
The secret information contained in carrier is measured, to prevent the progress of communication in time.
Common steganography carrier specifically includes that image, text and voice.Image-carrier since hidden capacity is larger, by
To more concern, steganography and Steganalysis are relatively mature.Voice and text hidden capacity are relatively small, steganography
It is also relatively large with steganalysis difficulty.Particularly, voice signal due to its randomness it is strong, steganography and steganalysis are more tired
It is difficult.
Compared to static carrier, dynamic carrier steganography increases time dimension, is to use a kind of more steganography now
Carrier.Among these, more representational is stream media technology, with the continuous development of network, based on Streaming Media on network
Using more and more, typical application such as voice communication technology (VoIP).
Steganography based on Streaming Media unit be broadly divided into the steganography based on agreement and the steganography based on load or.It is based on
The Information hiding of network protocol.Mainly make full use of the head of agreement is some not use, or optional region into
Row.This mode is since agreement discloses, and compare preferable detection.Information hiding based on load mainly passes through modification Streaming Media
The redundancy of load is in the middle to achieve the purpose that covert communications.
Summary of the invention
The present invention is directed to solve at least some of the technical problems in related technologies.
For this purpose, an object of the present invention is to provide a kind of voice-over-net stream steganography based on bidirectional circulating neural network
Analysis method, this method can quickly and efficiently judge whether raw tone stream passes through information steganography.
It is another object of the present invention to propose a kind of voice-over-net stream steganography based on bidirectional circulating neural network point
Analysis apparatus.
In order to achieve the above objectives, one aspect of the present invention embodiment proposes a kind of network based on bidirectional circulating neural network
Voice flow steganalysis method, comprising:
The voice-over-net stream steganalysis method based on bidirectional circulating neural network of the embodiment of the present invention, passes through sliding window
Mouth method obtains quantization index vector from Web compression voice flow, is input with quantization index vector, passes through bidirectional circulating mind
The correlative character vector of code word in quantization index vector is extracted through network struction code word network of relation, and uses feature point
Class network classifies to obtained correlative character vector, whether can quickly and efficiently judge raw tone stream
By steganography.
In addition, the voice-over-net stream steganalysis side according to the above embodiment of the present invention based on bidirectional circulating neural network
Method can also have following additional technical characteristic:
Further, in one embodiment of the invention, the voice-over-net according to the steganography probabilistic determination
Current test collection sample whether steganography, comprising:
The steganography probability is more than or equal to default decision threshold, then there are steganography in the voice-over-net current test collection sample
Information；
The steganography probability is less than default decision threshold, then there is no steganography to believe in the voice-over-net current test collection sample
Breath.
Further, in one embodiment of the invention, before step S2 further include:
The half voice-over-net stream training set sample in the voice-over-net stream training set sample is selected to carry out information insertion,
Take default modulation system that required hiding information is embedded in the half voice-over-net stream training set sample of selection.
Further, in one embodiment of the invention, the voice-over-net stream training set sample and the network language
It is acquired in the voice-over-net stream sample database that sound current test collection sample is constructed in internet database or in advance.
Further, in one embodiment of the invention, the voice-over-net in the voice-over-net stream training set sample
Stream has K frame, then the quantization index vector is expressed as:
X=[x1,x2,…,xK]。
Further, in one embodiment of the invention, the code word correlation model is to bidirectional circulating neural network
What training generated, the code word correlation model are as follows:
Wherein, h is node state vector, and W is weighing vector, and b is bias vector, and y is output vector
H is shot and long term memory network, and x is the quantization index vector.
Further, in one embodiment of the invention, the tagsort model are as follows:
st=tanh (Wt·yt+bt)
Zk=Vk*sk+bk
Wherein, stIt is the middle layer output of the tagsort model, ZkIt is the output valve of corresponding a certain classification, Z is final
The normalization probability of obtained each classification, h are node state vectors.
In order to achieve the above objectives, another aspect of the present invention embodiment proposes a kind of net based on bidirectional circulating neural network
Network voice flow hidden information analysis device, comprising:
Module is obtained, for obtaining voice-over-net stream training set sample and voice-over-net current test collection sample；
Generation module, for extracting voice-over-net from the voice-over-net stream training set sample by sliding window method
Load is handled to obtain quantization index vector；
Abstraction module, for extracting the code word associated vector in the quantization index vector using code word correlation model；
Categorization module, for being classified by tagsort model to the code word associated vector；
Modeling module generates Stego-detection model for being iterated by measure of supervision；
Detection module detect for the voice-over-net current test collection sample to be inputted the Stego-detection model
To steganography probability, according to voice-over-net current test collection sample described in the steganography probabilistic determination whether steganography.
The voice-over-net stream hidden information analysis device based on bidirectional circulating neural network of the embodiment of the present invention, passes through sliding window
Mouth method obtains quantization index vector from Web compression voice flow, is input with quantization index vector, passes through bidirectional circulating mind
The correlative character vector of code word in quantization index vector is extracted through network struction code word network of relation, and uses feature point
Class network classifies to obtained correlative character vector, whether can quickly and efficiently judge raw tone stream
By steganography.
In addition, the voice-over-net stream steganalysis dress according to the above embodiment of the present invention based on bidirectional circulating neural network
Following additional technical characteristic can also be had by setting:
Further, in one embodiment of the invention, the voice-over-net according to the steganography probabilistic determination
Current test collection sample whether steganography, comprising:
The steganography probability is more than or equal to default decision threshold, then there are steganography letters for the voice-over-net current test collection sample
Breath；
The steganography probability is less than default decision threshold, then there is no steganography letters for the voice-over-net current test collection sample
Breath.
Further, in one embodiment of the invention, further includes: insertion module；
The insertion module, for selecting the half voice-over-net stream training set in the voice-over-net stream training set sample
Sample carries out information insertion, takes default modulation system that required hiding information is embedded in the half voice-over-net stream of selection
Training set sample.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partially become from the following description
Obviously, or practice through the invention is recognized.
Detailed description of the invention
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, in which:
Fig. 1 is the voice-over-net stream steganalysis side based on bidirectional circulating neural network according to one embodiment of the invention
Method flow chart；
Fig. 2 is the code word correlation model overall framework schematic diagram according to one embodiment of the invention；
Fig. 3 is the door recursive unit structural schematic diagram according to one embodiment of the invention；
Fig. 4 is the voice-over-net stream steganalysis side based on bidirectional circulating neural network according to one embodiment of the invention
Method application scenario diagram；
Fig. 5 is the steganalysis model schematic based on bidirectional circulating neural network according to one embodiment of the invention；
Fig. 6 is the voice coding flow chart according to one embodiment of the invention；
Fig. 7 is the quantization index modulation flow chart according to one embodiment of the invention；
Fig. 8 is to be filled according to the voice-over-net stream steganalysis based on bidirectional circulating neural network of one embodiment of the invention
Set structural schematic diagram.
Specific embodiment
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to attached
The embodiment of figure description is exemplary, it is intended to is used to explain the present invention, and is not considered as limiting the invention.
The voice-over-net based on bidirectional circulating neural network proposed according to embodiments of the present invention is described with reference to the accompanying drawings
Flow steganalysis method and device.
The network language based on bidirectional circulating neural network proposed according to embodiments of the present invention is described with reference to the accompanying drawings first
Sound stream steganalysis method.
Fig. 1 is the voice-over-net stream steganalysis side based on bidirectional circulating neural network according to one embodiment of the invention
Method flow chart.
As shown in Figure 1, should voice-over-net stream steganalysis method based on bidirectional circulating neural network the following steps are included:
Step S1 obtains voice-over-net stream training set sample and voice-over-net current test collection sample.
Further, in one embodiment of the invention, voice-over-net stream training set sample and voice-over-net current test
It is acquired in the voice-over-net stream sample database that collection sample is constructed in internet database or in advance.
Specifically, acquisition or other way construct a voice-over-net stream sample database from network.Therefrom randomly choose one
Subnetwork voice flow is as training sample set, and remaining voice-over-net stream is as test sample collection.
Step S2, through sliding window method from extraction voice-over-net load carries out in voice-over-net stream training set sample
Reason obtains quantization index vector.
Further, in one embodiment of the invention, before step S2 further include:
It selects the half voice-over-net stream training set sample in voice-over-net stream training set sample to carry out information insertion, takes
Required hiding information is embedded in the half voice-over-net stream training set sample of selection by default modulation system.
Specifically, it in training sample set, selects half voice-over-net stream therein to do information insertion, takes quantization rope
Required hiding information is embedded in voice flow by the mode for drawing modulation or gene debugging, and it is steganography that sample label, which is arranged,
Voice.To the other half voice-over-net stream, embedding information, setting sample label are not non-steganography voice.To test sample collection using same
The method of sample is handled.
Further, it, which has K frame, is assumed for one section of voice-over-net stream, then quantization index vector can indicate are as follows:
X=[x1,x2,…,xK]
G.729 or G.723 particularly for for voice compression coding, wherein xjFor the column vector of a 3*1, j ∈
[1,K]。
Step S3 extracts the code word associated vector in quantization index vector using code word correlation model.
Further, code word correlation model is to generate to bidirectional circulating neural metwork training, as shown in Fig. 2, illustrating
Overall framework schematic diagram.It can be expressed as by formula:
Wherein, h is node state vector, and W is weighing vector, and b is bias vector, and y is output vector.
Wherein, H is that usually shot and long term memory network (LSTM), also can choose similar to door recursive unit (GRU),
Structural representation with formula as shown in figure 3, can be stated are as follows:
it=σ (Wi·[ht-1,xt]+bi)
ft=σ (Wf·[ht-1,xt]+bf)
qt=tanh (Wq·[ht-1,xt]+bq)
ot=σ (Wo·[ht-1,xt]+bo)
ct=ft⊙ct-1+it⊙qt
ht=ot⊙tanh(ct)
Wherein W be weight vector, i be input gate vector, f be forget door vector, o be out gate vector, x be input to
Amount, b are linear bias, and h is node state, and σ is usually sigmoid function, and ⊙ is point-by-point operation.
Step S4 classifies to code word associated vector by tagsort model.
Further, tagsort model classifies the correlated characteristic vector that code word correlation model extracts, it is assumed that
The correlated characteristic vector extracted is y, then tagsort network can be described as:
st=tanh (Wt·yt+bt)
Zk=Vk*sk+bk
Wherein, s is the middle layer output of tagsort network, ZkIt is the output valve of corresponding a certain classification, Z is to finally obtain
Each classification normalization probability.
Further, feature extraction network and tagsort network, which all pass through, has the learning framework training of supervision to obtain.
Step S5 is iterated step S2-S4 by measure of supervision, generates Stego-detection model.
Further, it is iterated by using measure of supervision, available trained Stego-detection model.
Step S6 is detected voice-over-net current test collection sample input Stego-detection model to obtain steganography probability, root
According to steganography probabilistic determination voice-over-net current test collection sample whether steganography.
Voice-over-net current test collection sample is inputted into trained Stego-detection model, exports voice-over-net current test collection sample
The probability of this steganography whether there is steganography content according to probabilistic determination voice-over-net current test collection sample.
Further, in one embodiment of the invention, according to steganography probabilistic determination voice-over-net current test collection sample
Whether steganography, comprising:
Steganography probability is more than or equal to default decision threshold, then there are steganography information in voice-over-net current test collection sample；
Steganography probability is less than default decision threshold, then steganography information is not present in voice-over-net current test collection sample.
Specifically, for Stego-detection, have:
Wherein, z is the probability value exported before this, and threshold is given judgment threshold, and stego speech is steganography
Label, cover speech are non-steganography label.
As shown in Figure 4 and Figure 5, model designed by the embodiment of the present invention is primarily directed to the Information hiding based on load,
Detection carrier is the voice payload of Streaming Media unit and mainly seems VoIP voice flow.
In general, in order to efficiently carry out network transmission, common voice-over-net standard can all compress voice
Transmission, for G.723, voice coding process is as shown in Figure 6.
Wherein, voice sequence is compressed into a series of by linear predictive coding (Linear Predictive Coding, LPC)
Voice coefficient, then these coefficients, which are quantized into, is more suitable the quantization parameter bit stream of transmission and is transmitted on network.In
In voice steganography, popular steganography method is realized usually in the process of compression and quantization parameter and is write to information
Enter, such as quantization index modulation and pitch period steganography.Particularly, as shown in Figure 7 for quantization index modulation steganography.
Quantization index modulation steganography passes through the code book for dividing vector quantization, by Information hiding in code book classification.It receives
Side is by judging code book belonging to code word, thus the information flow that judgement is transmitted.Pitch period modulation is similar with this mode.This
The method of invention is able to detect the steganography including the voice-over-nets stream such as quantization index modulation and pitch period modulation, has certain
Versatility.
Since voice itself has very special property, for common audio steganalysis method to voice-over-net stream simultaneously
It is not particularly suitable.It is common such as directly in the detection mode of tim e- domain detection and transform domain since compress speech itself has
It is extraordinary can not perceive and voice-over-net stream required by real-time so that these methods and being not suitable for.
Particularly, the existing steganalysis for voice-over-net stream is mainly with statistical method and deep learning method
It is main.Statistics-Based Method mainly passes through manual extraction low order code word feature, is then classified using support vector machines, this
It is lower that kind mode generally calculates very fast but accuracy rate, and can not capture the information of high-order.Voice-over-net based on deep learning
Most of flow algorithm all real-times are poor, and most of correlation information all not accounted between voice sequence code word, so that
The Detection accuracy of complex condition is not high.
The voice-over-net stream steganalysis method based on bidirectional circulating neural network proposed according to embodiments of the present invention leads to
It crosses sliding window method and obtains quantization index vector from Web compression voice flow, be input with quantization index vector, by double
Code word network of relation is constructed to Recognition with Recurrent Neural Network to extract the correlative character vector of code word in quantization index vector, and is made
Classified with tagsort network to obtained correlative character vector, it being capable of the original language of judgement quickly and efficiently
Whether sound stream passes through steganography.
The voice-over-net based on bidirectional circulating neural network proposed according to embodiments of the present invention is described referring next to attached drawing
Flow hidden information analysis device.
Fig. 8 is to be filled according to the voice-over-net stream steganalysis based on bidirectional circulating neural network of one embodiment of the invention
Set structural schematic diagram.
As shown in figure 8, being somebody's turn to do the voice-over-net stream hidden information analysis device based on bidirectional circulating neural network includes: acquisition module
100, generation module 200, abstraction module 300, categorization module 400, modeling module 500 and detection module 600.
Wherein, module 100 is obtained, for obtaining voice-over-net stream training set sample and voice-over-net current test collection sample.
Generation module 200, for extracting voice-over-net from voice-over-net stream training set sample by sliding window method
Load is handled to obtain quantization index vector.
Abstraction module 300, for extracting the code word associated vector in quantization index vector using code word correlation model.
Categorization module 400, for being classified by tagsort model to code word associated vector.
Modeling module 500 generates Stego-detection model for being iterated by measure of supervision.
Detection module 600, it is hidden for being detected to obtain voice-over-net current test collection sample input Stego-detection model
Write probability, according to steganography probabilistic determination voice-over-net current test collection sample whether steganography.
Further, in one embodiment of the invention, according to steganography probabilistic determination voice-over-net current test collection sample
Whether steganography, comprising:
Steganography probability is more than or equal to default decision threshold, then there are steganography information for voice-over-net current test collection sample；
Steganography probability is less than default decision threshold, then steganography information is not present in voice-over-net current test collection sample.
Further, in one embodiment of the invention, further includes: insertion module；
It is embedded in module, for selecting the half voice-over-net stream training set sample in voice-over-net stream training set sample to carry out
Information insertion, takes default modulation system that required hiding information is embedded in the half voice-over-net stream training set sample of selection.
It should be noted that aforementioned to the voice-over-net stream steganalysis method embodiment based on bidirectional circulating neural network
Explanation be also applied for the device of the embodiment, details are not described herein again.
The voice-over-net stream hidden information analysis device based on bidirectional circulating neural network proposed according to embodiments of the present invention is led to
It crosses sliding window method and obtains quantization index vector from Web compression voice flow, be input with quantization index vector, by double
Code word network of relation is constructed to Recognition with Recurrent Neural Network to extract the correlative character vector of code word in quantization index vector, and is made
Classified with tagsort network to obtained correlative character vector, it being capable of the original language of judgement quickly and efficiently
Whether sound stream passes through steganography.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Implicitly include at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two, three
It is a etc., unless otherwise specifically defined.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiment or examples.In addition, without conflicting with each other, the skill of this field
Art personnel can tie the feature of different embodiments or examples described in this specification and different embodiments or examples
It closes and combines.
Although the embodiments of the present invention has been shown and described above, it is to be understood that above-described embodiment is example
Property, it is not considered as limiting the invention, those skilled in the art within the scope of the invention can be to above-mentioned
Embodiment is changed, modifies, replacement and variant.
Claims (10)
1. a kind of voice-over-net stream steganalysis method based on bidirectional circulating neural network, which is characterized in that including following step
It is rapid:
S1 obtains voice-over-net stream training set sample and voice-over-net current test collection sample；
S2 is extracted voice-over-net load from the voice-over-net stream training set sample by sliding window method and handle
To quantization index vector；
S3 extracts the code word associated vector in the quantization index vector using code word correlation model；
S4 classifies to the code word associated vector by tagsort model；
S5 is iterated step S2-S4 by measure of supervision, generates Stego-detection model；
The voice-over-net current test collection sample is inputted the Stego-detection model and is detected to obtain steganography probability, root by S6
According to voice-over-net current test collection sample described in the steganography probabilistic determination whether steganography.
2. the method according to claim 1, wherein the voice-over-net according to the steganography probabilistic determination
Current test collection sample whether steganography, comprising:
The steganography probability is more than or equal to default decision threshold, then there are steganography letters in the voice-over-net current test collection sample
Breath；
The steganography probability is less than default decision threshold, then steganography information is not present in the voice-over-net current test collection sample.
3. the method according to claim 1, wherein before step S2 further include:
It selects the half voice-over-net stream training set sample in the voice-over-net stream training set sample to carry out information insertion, takes
Required hiding information is embedded in the half voice-over-net stream training set sample of selection by default modulation system.
4. the method according to claim 1, wherein the voice-over-net stream training set sample and the network language
It is acquired in the voice-over-net stream sample database that sound current test collection sample is constructed in internet database or in advance.
5. the method according to claim 1, wherein the voice-over-net in the voice-over-net stream training set sample
Stream has K frame, then the quantization index vector is expressed as:
X=[x1,x2,…,xK]。
6. the method according to claim 1, wherein the code word correlation model is to bidirectional circulating neural network
What training generated, the code word correlation model are as follows:
Wherein, h is node state vector, and W is weighing vector, and b is bias vector, and y is output vector, and H is that shot and long term remembers net
Network, x are the quantization index vector.
7. the method according to claim 1, wherein the tagsort model are as follows:
st=tanh (Wt·yt+bt)
Zk=Vk*sk+bk
Wherein, stIt is the middle layer output of the tagsort model, ZkIt is the output valve of corresponding a certain classification, Z is to finally obtain
Each classification normalization probability, h is node state vector.
8. a kind of voice-over-net stream hidden information analysis device based on bidirectional circulating neural network characterized by comprising
Module is obtained, for obtaining voice-over-net stream training set sample and voice-over-net current test collection sample；
Generation module, for extracting voice-over-net load from the voice-over-net stream training set sample by sliding window method
It is handled to obtain quantization index vector；
Abstraction module, for extracting the code word associated vector in the quantization index vector using code word correlation model；
Categorization module, for being classified by tagsort model to the code word associated vector；
Modeling module generates Stego-detection model for being iterated by measure of supervision；
Detection module, it is hidden for being detected to obtain the voice-over-net current test collection sample input Stego-detection model
Write probability, according to voice-over-net current test collection sample described in the steganography probabilistic determination whether steganography.
9. device according to claim 6, which is characterized in that the voice-over-net according to the steganography probabilistic determination
Current test collection sample whether steganography, comprising:
The steganography probability is more than or equal to default decision threshold, then there are steganography information for the voice-over-net current test collection sample；
The steganography probability is less than default decision threshold, then steganography information is not present in the voice-over-net current test collection sample.
10. device according to claim 6, which is characterized in that further include: insertion module；
The insertion module, for selecting the half voice-over-net stream training set sample in the voice-over-net stream training set sample
Information insertion is carried out, takes default modulation system that required hiding information is embedded in the half voice-over-net stream training of selection
Collect sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910609734.2A CN110428846A (en) | 2019-07-08 | 2019-07-08 | Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910609734.2A CN110428846A (en) | 2019-07-08 | 2019-07-08 | Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110428846A true CN110428846A (en) | 2019-11-08 |
Family
ID=68410368
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910609734.2A Pending CN110428846A (en) | 2019-07-08 | 2019-07-08 | Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110428846A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111768803A (en) * | 2020-05-15 | 2020-10-13 | 宁波大学 | General audio steganalysis method based on convolutional neural network and multi-task learning |
| CN111899748A (en) * | 2020-04-15 | 2020-11-06 | 珠海市杰理科技股份有限公司 | Audio coding method and device based on neural network and coder |
| WO2024001108A1 (en) * | 2022-06-29 | 2024-01-04 | 浪潮电子信息产业股份有限公司 | Text answer determination method and apparatus, device, and medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102314881A (en) * | 2011-09-09 | 2012-01-11 | 北京航空航天大学 | MP3 (Moving Picture Experts Group Audio Layer 3) watermarking method for improving watermark-embedding capacity in MP3 file |
| US20190087726A1 (en) * | 2017-08-30 | 2019-03-21 | The Board Of Regents Of The University Of Texas System | Hypercomplex deep learning methods, architectures, and apparatus for multimodal small, medium, and large-scale data representation, analysis, and applications |
| CN109522445A (en) * | 2018-11-15 | 2019-03-26 | 辽宁工程技术大学 | A kind of audio classification search method merging CNNs and phase algorithm |
| CN109729070A (en) * | 2018-11-28 | 2019-05-07 | 甘肃农业大学 | A kind of detection method of the concurrent stego-channel of network isomery based on CNN and RNN Fusion Model |
| CN109815496A (en) * | 2019-01-22 | 2019-05-28 | 清华大学 | Based on capacity adaptive shortening mechanism carrier production text steganography method and device |
| CN109817233A (en) * | 2019-01-25 | 2019-05-28 | 清华大学 | Voice flow steganalysis method and system based on level attention network model |
-
2019
- 2019-07-08 CN CN201910609734.2A patent/CN110428846A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102314881A (en) * | 2011-09-09 | 2012-01-11 | 北京航空航天大学 | MP3 (Moving Picture Experts Group Audio Layer 3) watermarking method for improving watermark-embedding capacity in MP3 file |
| US20190087726A1 (en) * | 2017-08-30 | 2019-03-21 | The Board Of Regents Of The University Of Texas System | Hypercomplex deep learning methods, architectures, and apparatus for multimodal small, medium, and large-scale data representation, analysis, and applications |
| CN109522445A (en) * | 2018-11-15 | 2019-03-26 | 辽宁工程技术大学 | A kind of audio classification search method merging CNNs and phase algorithm |
| CN109729070A (en) * | 2018-11-28 | 2019-05-07 | 甘肃农业大学 | A kind of detection method of the concurrent stego-channel of network isomery based on CNN and RNN Fusion Model |
| CN109815496A (en) * | 2019-01-22 | 2019-05-28 | 清华大学 | Based on capacity adaptive shortening mechanism carrier production text steganography method and device |
| CN109817233A (en) * | 2019-01-25 | 2019-05-28 | 清华大学 | Voice flow steganalysis method and system based on level attention network model |
Non-Patent Citations (4)
| Title |
|---|
| HAO YANG ET AL.: "Steganalysis of VoIP Streams with CNN-LSTM Network", 《INFORMATION HIDING AND MULTIMEDIA SECURITY》 * |
| MIKE SCHUSTER: "Bidirectional recurrent neural networks", 《IEEE TRANSACTIONS ON SIGNAL PROCESSING》 * |
| ZINAN LIN ET AL.: "RNN-SM: Fast Steganalysis of VoIP Streams Using Recurrent Neural Network", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 * |
| 李长存: "《多媒体安全与认证》", 31 July 2014, 国防工业出版社 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111899748A (en) * | 2020-04-15 | 2020-11-06 | 珠海市杰理科技股份有限公司 | Audio coding method and device based on neural network and coder |
| CN111899748B (en) * | 2020-04-15 | 2023-11-28 | 珠海市杰理科技股份有限公司 | Audio coding method and device based on neural network and coder |
| CN111768803A (en) * | 2020-05-15 | 2020-10-13 | 宁波大学 | General audio steganalysis method based on convolutional neural network and multi-task learning |
| CN111768803B (en) * | 2020-05-15 | 2024-01-30 | 天翼安全科技有限公司 | General audio steganalysis method based on convolutional neural network and multitask learning |
| WO2024001108A1 (en) * | 2022-06-29 | 2024-01-04 | 浪潮电子信息产业股份有限公司 | Text answer determination method and apparatus, device, and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110428846A (en) | Voice-over-net stream steganalysis method and device based on bidirectional circulating neural network | |
| CN109104441A (en) | A kind of detection system and method for the encryption malicious traffic stream based on deep learning | |
| CN110136731A (en) | Empty cause and effect convolution generates the confrontation blind Enhancement Method of network end-to-end bone conduction voice | |
| CN109817233A (en) | Voice flow steganalysis method and system based on level attention network model | |
| CN108877832A (en) | A kind of audio sound quality also original system based on GAN | |
| CN110533570A (en) | A kind of general steganography method based on deep learning | |
| CN108154096A (en) | A kind of checking method and device of hearing data | |
| CN111966998A (en) | Password generation method, system, medium, and apparatus based on variational automatic encoder | |
| CN110349588A (en) | A kind of LSTM network method for recognizing sound-groove of word-based insertion | |
| KR20200143940A (en) | Method and system for assessing language ability based on artificial intelligence | |
| CN112967725A (en) | Voice conversation data processing method and device, computer equipment and storage medium | |
| Yang et al. | Real-time steganalysis for streaming media based on multi-channel convolutional sliding windows | |
| CN105283916B (en) | Electronic watermark embedded device, electronic watermark embedding method and computer readable recording medium | |
| Zhou et al. | ICRC-HIT: A deep learning based comment sequence labeling system for answer selection challenge | |
| Chen et al. | Distribution-preserving steganography based on text-to-speech generative models | |
| CN116844095A (en) | Video emotion polarity analysis method based on multi-mode depth feature level fusion | |
| Ding et al. | Discop: Provably secure steganography in practice based on" distribution copies" | |
| CN113407663A (en) | Image-text content quality identification method and device based on artificial intelligence | |
| WO2020260869A1 (en) | Anonymization | |
| CN110049034A (en) | A kind of real-time Sybil attack detection method of complex network based on deep learning | |
| Zhang et al. | An encrypted speech retrieval method based on deep perceptual hashing and CNN-BiLSTM | |
| CN116721176B (en) | Text-to-face image generation method and device based on CLIP supervision | |
| CN113571048A (en) | Audio data detection method, device, equipment and readable storage medium | |
| CN111445545B (en) | Text transfer mapping method and device, storage medium and electronic equipment | |
| CN115438753B (en) | Method for measuring security of federal learning protocol data based on generation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191108 |
|
| RJ01 | Rejection of invention patent application after publication |