Audio information hiding method and device based on GAN network

Through a GAN network-based method, the speech samples and image samples are spliced ​​and time-domain transformed, and the model is trained to generate synthetic image samples, solving the limitations of hidden objects and carrier selection in the existing technology, and achieving the effective speech information Hide and transmit.

CN119049488APending Publication Date: 2024-11-29GUANGDONG UNIV OF TECH
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202411191517.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-28
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

The existing data steganography technology has limitations in the selection of hidden objects and carriers. Especially when images are used as carriers, text or images are usually used as hidden objects, and audio is selected as hidden objects.

Method used

Using a GAN network-based method, frequency domain conversion is performed by obtaining speech samples and image samples, spectrum stitching and time domain conversion are performed, and the GAN network model is trained to generate synthetic image samples to realize the hiding of speech information.

Benefits of technology

It solves the limitations of hidden object and carrier selection, and can effectively embed voice information into images, generate synthetic images containing steganography, and improves the security of information transmission.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049488A_ABST
    Figure CN119049488A_ABST
Patent Text Reader

Abstract

The invention discloses an audio information hiding method and device based on a GAN network, and the method comprises the steps: carrying out the frequency domain conversion based on an obtained voice sample as a hiding object and an image sample as a carrier, obtaining a corresponding frequency spectrum, carrying out the splicing of a voice frequency spectrum and an image frequency spectrum, and carrying out the time domain transformation processing, thereby obtaining the hidden audio information. The method comprises the following steps: acquiring a voice sample, acquiring a synthesized image sample, training a GAN network model by using the voice sample, the image sample and the synthesized image sample, so as to minimize the comparison loss of a decoded voice acquired by the synthesized image sample and an original voice, and acquiring a data steganography processing model based on the GAN network after the training is completed. According to the method, the actual voice data and the image data can be input into the trained data steganography processing model, so that the synthetic image containing the steganography voice is generated, and the technical problem of limitation of an existing data steganography technology in hidden object and carrier selection is solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data steganography technology, and in particular to a method and device for hiding audio information based on a GAN network. Background Art

[0002] With the widespread use of digital communications, the need for concealed transmission of sensitive information is growing. In some cases, encryption alone is not enough to prevent the enemy from suspecting the existence of sensitive information, so steganography provides an additional layer of security by embedding voice information into seemingly ordinary communication media to reduce the risk of detection.

[0003] Image steganography is the main content of information hiding. The sender hides the secret information in the original image, generates a cryptic image carrying the secret information, and transmits the secret information on the public channel through the cryptic image. The receiver of the transmission can decode the secret information carried in the cryptic image. Other listeners on the public channel cannot detect the secret information in the cryptic image. However, the existing data steganography technology has obvious limitations in the selection of hidden objects and carriers. If the image is used as the carrier, the text or image is generally used as the hidden object. Only when the carrier is audio, audio is selected as the hidden object. Summary of the invention

[0004] The present application provides a method and device for hiding audio information based on a GAN network, which is used to solve the technical problems of limitations of existing data steganography technology in the selection of hidden objects and carriers.

[0005] In order to solve the above technical problems, the first aspect of the present application provides an audio information hiding method based on a GAN network, comprising:

[0006] Acquire preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively;

[0007] Inputting the speech sample spectrum and the image sample spectrum into a GAN network model, performing spectrum splicing through the GAN network model to obtain a synthetic sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then converting the synthetic sample spectrum into a synthetic image sample through time domain conversion;

[0008] According to each group of speech samples, image samples and synthetic image samples, the GAN network model is trained, and when the convergence degree or the number of iterations of the GAN network model meets the preset conditions, a data steganography processing model is obtained;

[0009] Acquire the voice data and image data to be processed, perform frequency domain conversion on the voice data and the image data respectively to obtain a voice spectrum and an image spectrum, and then input the voice spectrum and the image spectrum into the data steganography processing model to obtain a synthetic image through the data steganography processing model.

[0010] Preferably, before inputting the speech sample spectrum and the image sample spectrum as training samples into the GAN network model, the method further includes:

[0011] Obtain key data, where the key data is used to be input into a GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to obtain an encrypted synthetic image sample.

[0012] Preferably, the loss function of the GAN network model is specifically:

[0013]

[0014] In the formula, is the loss function output of the encoding module, is the loss function output of the decoding module, is the loss function output of the attack module, Represents the original speech input to the encoding module, The decoded speech representing the output of the decoding module, is the decoded speech representing the output of the attack module, represents the composite image output by the encoding module, K represents the key data, , are the weights of the attack module and the decoding module respectively, and d represents the distance value.

[0015] Preferably, performing frequency domain conversion on the speech samples and the image samples to obtain corresponding speech sample spectra and image sample spectra respectively specifically includes:

[0016] The speech samples and the image samples are subjected to wavelet transform processing, and the speech samples and the image samples are decomposed into a plurality of wavelet sub-bands of different frequency bands to obtain speech sample frequency spectra and image sample frequency spectra.

[0017] Preferably, the wavelet transform processing is specifically FWT transform, and the expression of the FWT transform is:

[0018]

[0019] In the formula, is the wavelet coefficient, is a discrete signal, is the discretized wavelet function.

[0020] At the same time, the second aspect of the present application provides an audio information hiding device based on a GAN network, comprising:

[0021] A sample spectrum generating unit, used to obtain preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively;

[0022] A data synthesis unit, used to input the speech sample spectrum and the image sample spectrum into a GAN network model, perform spectrum splicing through the GAN network model to obtain a synthesized sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then convert the synthesized sample spectrum into a synthesized image sample through time domain conversion;

[0023] An adversarial model training unit is used to perform model adversarial training on the GAN network model according to each group of training samples and synthetic image samples, and obtain a data steganography processing model when the convergence degree or the number of iterations of the GAN network model meets a preset condition;

[0024] The data steganography processing unit is used to obtain the voice data and image data to be processed, perform frequency domain conversion on the voice data and the image data respectively to obtain a voice spectrum and an image spectrum, and then input the voice spectrum and the image spectrum into the data steganography processing model to obtain a synthetic image through the data steganography processing model.

[0025] Preferably, it also includes:

[0026] A key data acquisition unit is used to acquire key data, and the key data is used to input into the GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to generate an encrypted synthetic image sample.

[0027] Preferably, the loss function of the GAN network model is specifically:

[0028]

[0029] In the formula, is the loss function output of the encoding module, is the loss function output of the decoding module, is the loss function output of the attack module, Represents the original speech input to the encoding module, The decoded speech representing the output of the decoding module, is the decoded speech representing the output of the attack module, represents the composite image output by the encoding module, K represents the key data, , are the weights of the attack module and the decoding module respectively, and d represents the distance value.

[0030] Preferably, the sample spectrum generating unit is specifically used for:

[0031] Preset speech samples and image samples are obtained, and wavelet transform processing is performed on the speech samples and the image samples to decompose the speech samples and the image samples into a plurality of wavelet sub-bands of different frequency bands to obtain speech sample spectrum and image sample spectrum.

[0032] Preferably, the wavelet transform processing is specifically FWT transform, and the expression of the FWT transform is:

[0033]

[0034] In the formula, is the wavelet coefficient, is a discrete signal, is the discretized wavelet function.

[0035] It can be seen from the above technical solutions that this application has the following advantages:

[0036] The scheme of the present application is first based on the acquired voice samples as hidden objects and image samples as carriers, and then the voice samples and image samples are converted in the frequency domain to obtain corresponding spectra, and then the voice spectra and image spectra are spliced ​​and transformed in the time domain to obtain synthetic image samples, and then the GAN network model is trained using the voice samples, image samples and synthetic image samples to minimize the contrast loss between the decoded speech obtained from the synthetic image samples and the original speech. After the training is completed, a data steganalysis processing model based on the GAN network can be obtained. In practical applications, actual voice data and image data, or synthetic images containing steganalysis speech can be input into the trained data steganalysis processing model to generate a synthetic image containing steganalysis speech or decrypt the steganalysis speech from the synthetic image, thereby solving the technical problems of limitations of existing data steganography technology in the selection of hidden objects and carriers. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0038] Figure 1A schematic diagram of a flow chart of an embodiment of an audio information hiding method based on a GAN network provided in this application.

[0039] Figure 2 A schematic diagram of the overall framework of an embodiment of an audio information hiding method based on a GAN network provided in this application.

[0040] Figure 3 This is a schematic diagram of the architecture of the GAN network model of this application.

[0041] Figure 4 A schematic diagram of the structure of an embodiment of an audio information hiding device based on a GAN network provided in this application. DETAILED DESCRIPTION

[0042] The embodiments of the present application provide a method and device for hiding audio information based on a GAN network, which are used to solve the technical problems of limitations of existing data steganography technologies in the selection of hidden objects and carriers.

[0043] In order to make the purpose, features, and advantages of the invention of this application more obvious and easy to understand, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the embodiments described below are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0044] First, a detailed description of an embodiment of an audio information hiding method based on a GAN network provided by this application is as follows:

[0045] See also Figure 1 and Figure 2 , this embodiment provides an audio information hiding method based on a GAN network, comprising:

[0046] Step 101: Obtain preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively;

[0047] It should be noted that according to the method provided in this embodiment, first, corresponding voice samples and image samples are obtained for constructing the steganographic processing model mentioned in the subsequent training steps, wherein the voice samples are used as hidden objects and the image samples are used as carriers, and then frequency domain conversion is performed based on the obtained sample data to obtain corresponding voice sample spectra and image sample spectra.

[0048] More specifically, the spectrum conversion method of this embodiment specifically includes: performing wavelet transform processing on the speech samples and image samples, decomposing the speech samples and image samples into multiple wavelet sub-bands of different frequency bands, and obtaining speech sample spectra and image sample spectra.

[0049] It should be noted that the original audio and picture are input into the wavelet discrete transform module and decomposed into low-frequency and high-frequency wavelet sub-bands. In this embodiment, FWT transform is preferably used to reduce the amount of calculation and accelerate the speed of data preprocessing by using the orthogonal properties of wavelets. In some scenarios with low processing speed requirements, DWT transform can also be used instead. The formula is as follows:

[0050]

[0051] in: The scale is , translated to The wavelet coefficients of is a discrete signal, is the discretized wavelet function.

[0052] is the original wavelet function, and here we choose the Haar wavelet function with good localization characteristics.

[0053] Step 102: input the speech sample spectrum and the image sample spectrum into the GAN network model, perform spectrum splicing through the GAN network model to obtain a synthetic sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then convert the synthetic sample spectrum into a synthetic image sample through time domain conversion;

[0054] Step 103: training the GAN network model according to each group of speech samples, image samples and synthetic image samples. When the convergence degree or the number of iterations of the GAN network model meets the preset conditions, a data steganography processing model is obtained.

[0055] It should be noted that in steps 102 to 103, based on the speech sample spectrum and image sample spectrum obtained in step 101, combined with the preset GAN network model including an encoding module and a decoding module, these two modules can be classified as the generator in the GAN architecture from the perspective of the standard GAN network architecture, wherein the encoding module is responsible for shaping through the fully connected layer, and its function is to splice the image and audio information, so as to obtain the synthetic sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then convert the synthetic sample spectrum into a synthetic image sample containing hidden speech information through time domain conversion; the decoding module is responsible for decoding and restoring the original speech sample from the synthetic image sample, and after obtaining the synthetic image sample through the encoding link, the corresponding decoded speech is obtained by decoding the synthetic image sample, and the contrast loss between the decoded speech and the original speech is gradually minimized through the back-propagation training of the discriminator. When the convergence degree or the number of iterations of the GAN network model meets the preset conditions, the training can be stopped and the current GAN network model can be output as a data steganography processing model.

[0056] In some embodiments, the following optional steps may be included before step 102:

[0057] Step 1011, obtaining key data;

[0058] The key data is used to input into the GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to obtain an encrypted synthetic image sample.

[0059] It should be noted that when step 1011 is added to the method of this embodiment, the corresponding step 102 can also be adjusted to input the speech sample spectrum, the image sample spectrum and the key data into the GAN network model, and finally output the encrypted synthetic image. When the key data is introduced into the method, the overall architecture of the GAN network model mentioned in this embodiment can be referred to. Figure 3 , Figure 3 The figure shows a three-party adversarial architecture consisting of an encoder, a decoder, and an attacker. Alice (Encoder), Bob (Decoder), and Eve (Attacker) are provided. Alice and Bob want to communicate securely, while Eve wants to eavesdrop on their communication. The desired security property is confidentiality (not integrity), so Eve can intercept the communication.

[0060] Each of the three parties is a neural network with different goals. The Encoder uses 9 convolutional layers, and the Decoder and Attacker use 7 convolutional layers. Alice and Bob networks hope to communicate with maximum clarity while hiding their communication as much as possible from Eve, an eavesdropper in the system. In order to communicate with Bob, Alice sends Bob a confidential message P (picture). Voice , carrier image and key K is Alice and input. When Alice processes the input speech 、Carrier picture and key K and generate a secret image When both Bob and Eve receive (Bob also receives the key K) and tries to recover the original image information from Alice The outputs of Bob and Eve networks are and Alice and Bob share the same secret key K, which gives them a greater advantage over Eve, regenerating the key K for each plaintext (the above pictures and voices use the spectrum generated by the preprocessing step).

[0061] (1) Training process: input the preprocessed data into the network to obtain the output. Bob's loss function is defined as follows:

[0062]

[0063] Eve’s loss function is defined in the following formula, similar to Bob’s network, which is and The L1 distance between them.

[0064]

[0065] Finally, Alice's loss function is as follows:

[0066]

[0067] Take Alice's loss function as the total loss function, input the picture and voice into the FWT module, and then input the output of FWT into the adversarial network to minimize Alice's loss function, and then perform back propagation. Finally, the hidden voice can be obtained by scanning the picture with software.

[0068] Step 104, obtaining the voice data and image data to be processed, performing frequency domain conversion on the voice data and image data respectively to obtain a voice spectrum and an image spectrum, and then inputting the voice spectrum and the image spectrum into a data steganography processing model to obtain a synthetic image through the data steganography processing model.

[0069] It should be noted that in the actual application stage corresponding to step 104, based on the trained data steganography processing model, the actual voice data and image data obtained are input into the data steganography processing model according to the same processing flow as the modeling stage, and the data steganography processing model performs frequency domain conversion, spectrum splicing, time domain conversion and other processing to obtain a synthetic image containing the steganographic voice. When the steganographic voice in the synthetic image needs to be read, the corresponding synthetic image can be scanned by a device or APP that can remotely call the data steganography processing model or directly carries the data steganography processing model, so as to obtain the corresponding decoded voice. This method can be applied to places such as museums and hospitals that require explanation and guidance.

[0070] The above is a detailed description of an embodiment of an audio information hiding method based on a GAN network provided by the present application. The following is a detailed description of an embodiment of an audio information hiding device based on a GAN network provided by the present application.

[0071] See also Figure 4 This embodiment provides an audio information hiding device based on a GAN network, comprising:

[0072] The sample spectrum generating unit 201 is used to obtain preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively;

[0073] The data synthesis unit 202 is used to input the speech sample spectrum and the image sample spectrum into the GAN network model, perform spectrum splicing through the GAN network model to obtain a synthesized sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then convert the synthesized sample spectrum into a synthesized image sample through time domain conversion;

[0074] The adversarial model training unit 203 is used to perform model adversarial training on the GAN network model according to each group of training samples and synthetic image samples, and obtain a data steganography processing model when the convergence degree or the number of iterations of the GAN network model meets the preset conditions;

[0075] The data steganography processing unit 204 is used to obtain the voice data and image data to be processed, perform frequency domain conversion on the voice data and image data respectively to obtain a voice spectrum and an image spectrum, and then input the voice spectrum and the image spectrum into the data steganography processing model to obtain a synthetic image through the data steganography processing model.

[0076] Preferably, it also includes:

[0077] The key data acquisition unit 2011 is used to acquire key data, and the key data is used to be input into the GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to generate an encrypted synthetic image sample.

[0078] Preferably, the loss function of the GAN network model is specifically:

[0079]

[0080] In the formula, is the loss function output of the encoding module, is the loss function output of the decoding module, is the loss function output of the attack module, Represents the original speech input to the encoding module, The decoded speech representing the output of the decoding module, is the decoded speech representing the output of the attack module, represents the composite image output by the encoding module, K represents the key data, , are the weights of the attack module and the decoding module respectively, and d represents the distance value.

[0081] Preferably, the sample spectrum generating unit 201 is specifically used for:

[0082] The preset speech samples and image samples are obtained, and the speech samples and image samples are processed by wavelet transform, the speech samples and image samples are decomposed into a plurality of wavelet sub-bands of different frequency bands, and the speech sample spectrum and image sample spectrum are obtained.

[0083] Preferably, the wavelet transform processing is specifically FWT transform, and the expression of FWT transform is:

[0084]

[0085] In the formula, is the wavelet coefficient, is a discrete signal, is the discretized wavelet function.

[0086] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the terminals, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0087] In the several embodiments provided in the present application, it should be understood that the disclosed terminals, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0088] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described herein can, for example, be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0089] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0090] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0091] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0092] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0093] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for hiding audio information based on a GAN network, characterized in that: include: Acquire preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively; Inputting the speech sample spectrum and the image sample spectrum into a GAN network model, performing spectrum splicing through the GAN network model to obtain a synthetic sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then converting the synthetic sample spectrum into a synthetic image sample through time domain conversion; According to each group of speech samples, image samples and synthetic image samples, the GAN network model is trained, and when the convergence degree or the number of iterations of the GAN network model meets the preset conditions, a data steganography processing model is obtained; Acquire the voice data and image data to be processed, perform frequency domain conversion on the voice data and the image data respectively to obtain a voice spectrum and an image spectrum, and then input the voice spectrum and the image spectrum into the data steganography processing model to obtain a synthetic image through the data steganography processing model.

2. According to the GAN network-based audio information hiding method of claim 1, it is characterized in that: Before inputting the speech sample spectrum and the image sample spectrum as training samples into the GAN network model, the following further includes: Obtain key data, where the key data is used to be input into a GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to obtain an encrypted synthetic image sample.

3. According to claim 2, a method for hiding audio information based on a GAN network is characterized in that: The loss function of the GAN network model is specifically: ; In the formula, is the loss function output of the encoding module, is the loss function output of the decoding module, is the loss function output of the attack module, Represents the original speech input to the encoding module, The decoded speech representing the output of the decoding module, is the decoded speech representing the output of the attack module, represents the composite image output by the encoding module, K represents the key data, , are the weights of the attack module and the decoding module respectively, and d represents the distance value.

4. According to the GAN network-based audio information hiding method of claim 1, it is characterized in that: Performing frequency domain conversion on the speech samples and the image samples to obtain corresponding speech sample spectra and image sample spectra respectively specifically includes: The speech samples and the image samples are subjected to wavelet transform processing, and the speech samples and the image samples are decomposed into a plurality of wavelet sub-bands of different frequency bands to obtain speech sample frequency spectra and image sample frequency spectra.

5. According to claim 4, the audio information hiding method based on the GAN network is characterized in that: The wavelet transform process is specifically FWT transform, and the expression of the FWT transform is: ; In the formula, is the wavelet coefficient, is a discrete signal, is the discretized wavelet function.

6. An audio information hiding device based on a GAN network, characterized in that: include: A sample spectrum generating unit, used to obtain preset speech samples and image samples, and perform frequency domain conversion on the speech samples and image samples to obtain corresponding speech sample spectra and image sample spectra respectively; A data synthesis unit, used to input the speech sample spectrum and the image sample spectrum into a GAN network model, perform spectrum splicing through the GAN network model to obtain a synthesized sample spectrum corresponding to the speech sample spectrum and the image sample spectrum, and then convert the synthesized sample spectrum into a synthesized image sample through time domain conversion; An adversarial model training unit is used to perform model adversarial training on the GAN network model according to each group of training samples and synthetic image samples, and obtain a data steganography processing model when the convergence degree or the number of iterations of the GAN network model meets a preset condition; The data steganography processing unit is used to obtain the voice data and image data to be processed, perform frequency domain conversion on the voice data and the image data respectively to obtain a voice spectrum and an image spectrum, and then input the voice spectrum and the image spectrum into the data steganography processing model to obtain a synthetic image through the data steganography processing model.

7. The audio information hiding device based on the GAN network according to claim 6 is characterized in that: Also includes: A key data acquisition unit is used to acquire key data, and the key data is used to input into the GAN network model to combine the speech sample spectrum and the image sample spectrum in the GAN network model to generate an encrypted synthetic image sample.

8. The audio information hiding device based on the GAN network according to claim 7 is characterized in that: The loss function of the GAN network model is specifically: ; In the formula, is the loss function output of the encoding module, is the loss function output of the decoding module, is the loss function output of the attack module, Represents the original speech input to the encoding module, The decoded speech representing the output of the decoding module, is the decoded speech representing the output of the attack module, represents the composite image output by the encoding module, K represents the key data, , are the weights of the attack module and the decoding module respectively, and d represents the distance value.

9. The audio information hiding device based on GAN network according to claim 6 is characterized in that: The sample spectrum generating unit is specifically used for: Preset speech samples and image samples are obtained, and wavelet transform processing is performed on the speech samples and the image samples to decompose the speech samples and the image samples into a plurality of wavelet sub-bands of different frequency bands to obtain speech sample spectrum and image sample spectrum.

10. The audio information hiding device based on GAN network according to claim 9, characterized in that: The wavelet transform process is specifically FWT transform, and the expression of the FWT transform is: ; In the formula, is the wavelet coefficient, is a discrete signal, is the discretized wavelet function.

Citation Information

Patent Citations

  • Wavelet transform and Arnold based adaptive gray-scale watermark embedded method

    CN102142258A

  • Video-based digital holographic information hiding and processing method

    CN110135134A

  • Secure non-embedded steganography method based on generative adversarial network

    CN112243005A

  • Information hiding method and device

    CN113515753A

  • Neural network training method, image coding and decoding method, device and medium

    CN113705806A