JP2023067365A - Concealed voice transmitter, concealed voice receiver, concealed voice transmission system, concealed voice transmission method, and concealed voice transmission program - Google Patents

Abstract

To provide a concealed voice transmitter, a concealed voice receiver, a concealed voice transmission system, a concealed voice transmission method, and a concealed voice transmission program that can deliver voice with high confidentiality and restore it simply.SOLUTION: A secret voice transmitter 12 generates a mask M based on presence or absence of frequency bins in a spectrogram obtained by short-time Fourier transforming a mixed voice of the secret voice and i disturbance sounds where the frequency component of the secret voice is larger than the disturbance sounds by at least a predetermined threshold value θ. Then, n (2≤t≤n) share masks M'n, which are different from one another, are generated based on the mask M to restore the secret voice by superposing t or more. Then, the secret voice transmitter 12 generates n share voices, different from one another, by masking the spectrogram obtained by short-time Fourier transform of the mixed voice with n share masks M'n, and outputs the inverse Fourier transformed n share voices.SELECTED DRAWING: Figure 2

Description

The present invention relates to a confidential voice transmission device, a confidential voice receiving device, a confidential voice transmission system, a confidential voice transmission method, and a confidential voice transmission program.

There are many uses for concealing messages in voices distributed to an unspecified number of people so that only specific people can receive them, and several methods have been proposed for this purpose.

For example, Patent Literature 1 discloses a method of embedding encrypted speech in sound. The method of Patent Document 1 extracts identification information from a sound pickup signal obtained by picking up sound emitted by a sound emitting device of the facility, and receives a plurality of pieces of related information about the facility. Used for facility voice guidance. Patent Document 1 also discloses a method of embedding identification information in sound by modulating it in a non-audible band.

In general, the security of cryptography is ensured by the computational complexity involved in decryption, but the method of Patent Document 1 requires means for obtaining related information using identification information, resulting in a complicated configuration. , the encryption method that embeds the identifying information in the sound becomes insecure if compromised. Further, even if the identification information is modulated in a non-audible band in order to conceal the identification information, it is easily decoded if the modulated frequency is known, so it is not suitable for confidential voice communication.

On the other hand, in the field of image processing, a visual cryptography scheme (hereinafter referred to as "VCS") shown in Non-Patent Document 1 is known as a technique that is safe even if the encryption method is leaked. there is VCS generates n share images from one secret image, and collects at least t share images among them to restore the original secret image. On the other hand, since the VCS cannot restore the confidential image from t-1 shared images, the secret cannot be leaked unless t or more shared images are collected. Such a scheme is also called (t,n)-VCS.

VCS is characterized by the fact that only t or more share images need to be superimposed in order to restore the secret image, and complicated decoding processing is not required. FIG. 11 shows an example of (t,n)=(3,4)-VCS. In the example of FIG. 11, three or more shared images are superimposed to restore the white or black pixels of the confidential image. Note that FIG. 11A shows a case where the pixel with the confidential image is white (0), and FIG. 11B shows the case where the pixel with the confidential image is black (1). In the example of FIG. 11, there are four share images, and one pixel of the confidential image is expanded three times horizontally and twice vertically to six pixels. That is, the share image in FIG. 11 corresponds to one pixel of the secret image. In order to restore the secret image, a share image corresponding to all pixels of the secret image is generated.

In the example of FIG. 11, the number of pixels that become black by combining two arbitrary share images is 4 out of 6 pixels regardless of whether the original pixels of the confidential image are white (0) or black (1). be. For this reason, it is impossible to distinguish between the combination of arbitrary two of the share images shown in FIG. 11A and the combination of arbitrary two of the share images shown in FIG. 11B. On the other hand, when it comes to a combination of arbitrary three shares, in FIG. In some FIG. 11B, 5 out of 6 pixels are black. As a result, there is a difference in the brightness of the pixels forming the restored images between FIG. 11A and FIG. 11B, and it is possible to distinguish whether the pixels forming the original confidential image were white or black. In this way, the VCS generates n share images after enlarging each pixel of the secret image, so that the secret image cannot be recognized unless t or more share images are collected.

Non-Patent Literatures 2 and 3 propose a method of generating a share image that satisfies conditions for restoring a confidential image for arbitrary t and n. In addition, Patent Document 2 discloses a method of visual encryption using augmented reality. Visual encryption is used to prevent unauthorized viewing of distributed content. can be prevented.

Japanese Patent Application Laid-Open No. 2020-021101 Japanese translation of PCT publication No. 2017-538152 M. Naor and A. Shamir, “Visual cryptography,” Advances in Cryptology - EUROCRYPT '94, Workshop on the Theory and Application of Cryptographic Techniques, 1994, Proceedings, vol.950, pp.1--12, Lecture Notes in Computer Science, Springer, 1994. S.J. Shyu and M.C. Chen, “Optimum pixel expansions for threshold visual secret sharing schemes,” IEEE Transactions on Information Forensics and Security, vol.6, no.3, pp.960--969, 2011. M. Iwamoto, “A weak security notion for visual secret sharing schemes,” IEEE Transactions on Information Forensics and Security, vol.7, no.2, pp.372--382, 2012.

As described above, VCS is effective in distributing anonymized images and restoring these images, but there are no examples of its application to audio.

Therefore, in view of the above background, the present invention provides a confidential voice transmission device, a confidential voice receiving device, a confidential voice transmission system, a confidential voice transmission method, and a confidential voice transmission program that can distribute voice with high confidentiality and easily restore it. intended to

The confidential voice transmission apparatus of the present invention includes a mask generation unit that generates a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice, and t A share mask generation unit that generates n (2 ≤ t ≤ n) share masks that are different from each other for restoring the confidential voice by superimposing the above, based on the mask; A shared voice generation unit that generates n shared voices that are different from each other by masking a spectrogram obtained by short-time Fourier transforming a mixed voice mixed with voices with the n share masks, and an inverse Fourier transform. and a shared audio output unit configured to output the n shared audios.

According to the confidential voice transmission device of the present invention, the other voices are i interfering sounds, and the mask generator generates mixed voices obtained by mixing the i interfering sounds with the confidential voices using a short-time Fourier filter. The mask may be generated based on the presence or absence of a frequency bin in which the frequency component of the concealed voice is greater than that of the interfering sound by a predetermined threshold or more in the spectrogram obtained by the conversion.

According to the confidential voice transmission device of the present invention, the other voices are n cover voices, and the mask generator performs a short-time Fourier transform on the mixed voice obtained by mixing the confidential voices with the n cover voices. In the spectrogram obtained by the above, the mask is generated based on the presence or absence of frequency bins in which the frequency components of the concealed audio are larger than the cover audio by a predetermined threshold or more, and the frequency components of the cover audio are the concealed audio and other generating j (1 ≤ j ≤ n) cover masks based on the presence or absence of frequency bins larger than the cover audio of the The n share masks may be generated based on .

The confidential voice transmission system of the present invention restores the confidential voice by superimposing the confidential voice transmission device described above and the t or more of the n shared voices output from the confidential voice transmission device. and a confidential voice receiving device for

The confidential voice transmission apparatus of the present invention includes a mask generation unit that generates a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice, and t A share mask generation unit that generates, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing the above, and the n share masks. and a voice output unit for outputting a mixed voice obtained by mixing the confidential voice with i interfering sounds.

According to the confidential voice transmission device of the present invention, the mask generation unit generates the frequency The mask may be generated based on the presence or absence of frequency bins whose components are greater than the interfering sound by a predetermined threshold or more.

According to the confidential voice transmission device of the present invention, the share mask output unit converts the spectrogram obtained by short-time Fourier transform of noise whose intensity is equal to or greater than a predetermined value in a predetermined frequency range to the n share masks. After masking, the n shared voices generated by inverse Fourier transform may be output as sounds.

According to the confidential voice transmission device of the present invention, the share mask output unit may output each of the n share masks as digital data.

A confidential voice transmission system according to the present invention includes the confidential voice transmission device described above, and restores the mask from the t or more share masks out of the n share masks output from the share mask output unit. and a hidden voice receiving device that restores the hidden voice by masking the restored mask to the mixed voice output from the voice output unit.

According to the confidential audio transmission device of the present invention, the mask generation unit may generate the mask by setting 1 to frequency bins equal to or greater than the threshold and 0 to frequency bins less than the threshold.

According to the confidential audio transmission device of the present invention, the share mask generator multiplies the mask by m by increasing the number of frequency bins and the number of time frames to satisfy the basic matrix of VCS (Visual Cryptography Scheme). The n share masks may be generated to be a determinant.

According to the confidential voice transmission system of the present invention, the confidential voice receiving device may include a similarity determination unit that determines similarity between the restored voice and the confidential voice.

The confidential voice transmission method of the present invention includes a first step of generating a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice; A second step of generating, based on the masks, n (2≤t≤n) share masks each different for restoring the confidential voice by superimposing the above, and another voice on the confidential voice A third step of masking the spectrogram obtained by short-time Fourier transforming the mixed voice mixed with the n share masks with the n share masks to generate n share voices that are different from each other; A fourth step of outputting n shared voices, and a fifth step of restoring the confidential voice by superimposing the t or more out of the n shared voices that have been output.

The confidential voice transmission method of the present invention includes a first step of generating a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice; a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing the above; a third step of outputting a mixed voice obtained by mixing i interfering sounds with the confidential voice; and restoring the mask from the t or more of the output n share masks. and a fourth step of reconstructing the concealed voice by masking the reconstructed mask to the mixed voice.

The confidential voice transmission program of the present invention provides a computer with a first step of generating a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice; , a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more; a third step of masking a spectrogram obtained by short-time Fourier transforming a mixed voice mixed with other voices with the n share masks to generate n share voices that are different from each other; A fourth step of outputting the converted n shared voices, and a fifth step of restoring the confidential voice by superimposing the t or more of the output n shared voices. Let

The confidential voice transmission program of the present invention provides a computer with a first step of generating a mask based on the presence or absence of a frequency bin having a frequency component larger than a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice; , a second step of generating, based on the masks, n (2≤t≤n) different share masks for reconstructing the concealed speech by superimposing t or more; a third step of outputting a share mask and outputting a mixed sound obtained by mixing i interfering sounds with the confidential sound; a fourth step of reconstructing a mask and masking the reconstructed mask to the mixed audio to reconstruct the concealed audio;

According to the present invention, voice can be distributed with high secrecy and restored easily.

1 is a schematic configuration diagram of a confidential voice transmission system according to a first embodiment; FIG. It is a functional block diagram of the confidential voice transmission device of 1st Embodiment. 2 is a functional block diagram of the confidential voice receiving device of the first embodiment; FIG. FIG. 4 is a schematic diagram of decoding in the first embodiment; FIG. 11 is a functional block diagram of the confidential voice receiving device of the second embodiment; FIG. 11 is a functional block diagram of a confidential voice transmission device according to a third embodiment; FIG. 11 is a functional block diagram of a confidential voice receiving device according to a third embodiment; It is a functional block diagram of the confidential voice transmission device of 4th Embodiment. FIG. 11 is a functional block diagram of a confidential voice transmission device according to a fifth embodiment; FIG. 11 is a functional block diagram of a confidential voice receiving device according to a fifth embodiment; It is a schematic diagram of the share image of VCS, (A) is a case where the pixel of a secret image is white, (B) is a case where the pixel of a secret image is black.

Embodiments of the present invention will be described below with reference to the drawings. It should be noted that the embodiment described below is an example of the case of carrying out the present invention, and the present invention is not limited to the specific configuration described below. In carrying out the present invention, a specific configuration according to the embodiment may be appropriately adopted.

(basic matrix of VCS)
The embodiments described below apply a VCS (Visual cryptography scheme) to distribution and decoding of confidential voice. Therefore, an outline of VCS will be explained.

VCS is a technology that generates n share images from one secret image and collects at least t (2≦t≦n) share images to restore the original secret image. Note that t and n have a relationship of 2≦t≦n. As an example of application of VCS, game participants (also called users) acquire shared image data at different locations or at different times, and only participants who have acquired t or more shared images can acquire confidential images. , you can proceed to the next stage of the game.

In order to generate a share image for arbitrary (t, n), it is first necessary to define the access structure Γ of the participants to the share image. When there is ^a set P={1, 2, . Then, a set of shared images whose secret image can be restored by superimposing the shared images is defined as a qualified set _ΓQ .

Since the secret image can be restored by collecting t share images, Γ _Q is minimized when t share images are collected, and this case is defined as a minimal qualified set Γ ^* _Q . Conversely, a set in which no information about the confidential image can be obtained from the shared image is defined as a forbidden set _ΓF . Therefore, Γ _F is the complement of Γ _Q to the power set 2 ^P of the whole, and Γ=(Γ _Q , Γ _F ). The case where t−1 share images are gathered at this time is called a maximal forbidden set Γ ^* _F .

Here, when the secret image is a binary image, the access structure Γ is realized when the set of n*m Boolean matrices (X ₀ , X ₁ ) satisfies the following two conditions, where m is the magnification of pixel enlargement. is said to be the basic matrix for Note that 0 represents a white pixel and 1 represents a black pixel.

(Condition 1) Restorable conditions for confidential images:
There exists a constant α>0 for all S∈Γ ^* _Q such that HW(OR(X ₀ [S]))+αM≦HW(OR(X ₁ [S])). Here, X.[S] is an operation for extracting only the row corresponding to S from X., OR is OR for each column, and HW is Hamming weight.

(Condition 2) Safety conditions:
For all S∈Γ ^* _F , X ₀ [S] and X ₁ [S] can be made equal with proper column permutation.

A share image is generated based on the basic matrices X _{0 and} X ₁ that satisfy the above conditions. For example, the basic matrices X _{0 and} X ₁ can be obtained by solving an integer programming problem with the criterion of minimizing the pixel enlargement ratio m. This solution method is described in detail in Documents 1 and 2 below. Then, the obtained basic matrices X _{0 and} X ₁ represent the pixel values of the share image pixel-enlarged corresponding to the pixel values {0, 1} of the secret image, respectively.

As an example, referring to FIG. 11 where t=3, n=4, ie (3,4)-VCS, pixel magnification m=6, the determinant of FIG.
_X0 [1] = [0,1,1,1,0,0]
_X0 [2] = [1,0,1,1,0,0]
_X0 [3] = [1,1,0,1,0,0]
_X0 [4] = [1,1,1,0,0,0]
_X1 [1] = [1,0,0,0,1,1]
_X1 [2] = [0,1,0,0,1,1]
_X1 [3] = [0,0,1,0,1,1]
_X1 [4] = [0,0,0,1,1,1]

Reference 1 SJ Shyu and MC Chen, “Optimum pixel expansions for threshold visual secret sharing schemes,” IEEE Transactions on Information Forensics and Security, vol.6, no.3, pp.960--969, 2011.
Reference 2 M. Iwamoto, “A weak security notion for visual secret sharing schemes,” IEEE Transactions on Information Forensics and Security, vol.7, no.2, pp.372--382, 2012.

(First embodiment)
FIG. 1 is a schematic configuration diagram of a confidential voice transmission system 10 in which the VCS described above is applied to confidential voice transmission.

The confidential voice transmission system 10 of this embodiment is used for games and leisure as an example. For example, it is assumed that participants can listen to a secret message (secret voice) that only the target person can understand from voices (shared voices) recorded by actually going to a plurality of predetermined locations. This indicates that the participants who have heard the confidential voice have followed the originally determined course. In addition, the confidential voice transmission system 10 can be used for proof of existence. For example, multiple participants simultaneously record shared audio that is distributed at the same time, and when a certain number of shared audios are collected, a secret message (hidden audio) can be heard. This indicates that a plurality of participants who have heard the confidential voice were at a predetermined place at a predetermined time.

As shown in FIG. 1, a confidential voice transmission system 10 of this embodiment includes a confidential voice transmitting device 12 and a confidential voice receiving device 14 .

The confidential voice transmission device 12 generates n different shared voices from a message to be confidential (hereinafter referred to as "confidential voice"), and outputs the shared voices from the speaker 16 installed in the public space, thereby improving confidentiality. deliver the message while preserving Although it is not possible to recognize what the hidden audio is from each shared audio, the hidden audio can be restored by superimposing t (2≦t≦n) or more shared audio. That is, the confidential voice transmission device 12 encrypts the confidential voice into n shared voices and distributes them.

Note that the t shared voices may be output from one speaker 16 at different times, or may be output separately from a plurality of speakers 16 located at different locations.

The confidential voice receiving device 14 is, for example, a mobile terminal device owned by a game participant, and the mobile terminal device is provided with a function (application) for decoding confidential voices from n or more shared voices. Note that the mobile terminal device is a smartphone, a tablet terminal, or the like. The participants can listen to the confidential voices by collecting and decoding t or more shared voices using the mobile terminal device, which is the confidential voice receiving device 14 . However, if the number of shared voices collected is less than t, the participant cannot restore the confidential voice and cannot listen to the confidential voice. The restored confidential voice may be output from the speaker of the mobile terminal device, which is the confidential voice receiving device 14, or may be stored as digital data.

Here, a sound source separation technique for separating a sound signal for each sound source from voice in which a plurality of different sound signals are mixed is being researched. In this sound source separation technique, under an underdetermined condition in which the number of observed sound channels is less than the number of sound sources, it is difficult to separate the sound signal of one sound source originally desired to be observed from the sound signals of other sound sources. In particular, when there is one observation channel and information about the position of the sound source cannot be obtained, it is quite difficult to separate the sound source. Therefore, in the confidential voice transmission system 10 of the present embodiment, using this fact, a mixed voice obtained by mixing the confidential voice with other sounds (i interfering sounds in the present embodiment) is output from the speaker 16, A shared audio is generated based on the processing using VCS for this mixed audio.

FIG. 2 is a functional block diagram of the confidential voice transmission device 12. As shown in FIG.

The confidential voice transmission device 12 includes a Fourier transform unit 20 , a mask generation unit 22 , a share mask generation unit 24 , a masking unit 26 and an inverse Fourier transform unit 28 in addition to the speaker 16 . Note that the processes executed by the Fourier transform unit 20, the mask generator 22, the share mask generator 24, the masking unit 26, and the inverse Fourier transform unit 28 are executed by a program stored in a recording medium included in the confidential audio transmission device 12. executed. Also, by executing this program, a method corresponding to the program is executed.

The Fourier transform unit 20 is an FFT (Fast Fourier Transform) analyzer that performs a short-time Fourier transform on a mixed voice obtained by mixing confidential voice with i interfering sounds. The i interfering sounds mixed with the hidden voice are different sounds, and the interfering sounds use meaningless sounds or voices of the same or different speakers as the hidden voice that speaks other contents.

The mask generation unit 22 generates a mask M based on the presence or absence of frequency bins (time-frequency bins) in which the frequency components of the concealed voice are greater than the interfering sound by a predetermined threshold θ or more in the spectrogram obtained by Fourier transforming the mixed voice. Generate. Note that the mask generation unit 22 of the present embodiment generates the mask M by setting 1 to frequency bins equal to or greater than the threshold θ and 0 to frequency bins less than the threshold θ.

The share mask generation unit 24 generates, based on the mask M, n (2≦t≦n) different share masks M′ _n for restoring the confidential voice by superimposing t or more. As will be described later in detail, the share mask generation unit 24 of this embodiment multiplies the mask by m by increasing the number of frequency bins and the number of time frames _{, and} the _matrix Generate n share masks M' _n so as to satisfy the following equation.

The masking unit 26 masks the spectrogram obtained by short-time Fourier transforming the mixed voice with n share masks M′ _n to generate n different share voices.

The inverse Fourier transform unit 28 inverse Fourier transforms the n shared voices generated by the masking unit 26 .

The speaker 16 outputs the n shared voices that have undergone the inverse Fourier transform as sounds and distributes them to the user.

FIG. 3 is a functional block diagram of the confidential voice receiving device 14. As shown in FIG. The confidential voice receiving device 14 includes a microphone 30 , a synchronization section 32 and a voice restoring section 34 . The processing executed by the synchronization unit 32 and the audio restoration unit 34 is executed by a program stored in the recording medium of the confidential audio reception device 14. FIG. Also, by executing this program, a method corresponding to the program is executed.

The microphone 30 receives an input of shared voice output from the speaker 16 of the confidential voice transmission device 12 .

The synchronizing unit 32 performs synchronization processing for aligning the starting points of the n shared voices in order to restore the confidential voice from the n shared voices.

The audio restoration unit 34 restores the confidential audio by superimposing t or more of the synchronized n shared audios. Note that when the number of shared voices collected by the confidential voice receiving device 14 is less than t, the voice restoring unit 34 cannot restore the confidential voice. The confidential voice restored by the voice restoring unit 34 is output from the speaker of the mobile terminal device, which is the confidential voice receiving device 14 .

Next, the details of the generation of the share mask M' _n according to this embodiment will be described.

The Fourier transform unit 20 performs a short-time Fourier transform on the concealed speech and i (1≤i≤I) interfering sounds to obtain the concealed speech spectrum s(τ,f) and the interfering sound spectrum σ _i (τ,f). to get Note that τ is a time frame and 1≦τ≦T, and f is a frequency bin ID and 1≦f≦F.

The mask generation unit 22 sets the mask M to 1 for time-frequency bins in which the sound level |s| of the concealed sound is greater than the sound level |σ _i | of the interfering sound by a threshold value θ or more, and for other time-frequency bins set mask M to 0. That is, the mask M corresponding to the concealed sound whose sound level is relatively higher than the sound level of the interfering sound is set to 1. This is expressed in Equation 1.

Then, by multiplying the mask M(τ, f) by the spectrum X of the mixed speech represented by Equation 2 below, the main part of the spectrum of the concealed speech can be extracted. That is, the mask M(τ, f) represents the spectrum of the hidden voice in {0, 1}.

Therefore, the share mask generation unit 24 generates a share mask M' _n by regarding the mask M generated by the mask generation unit 22 as a confidential image (binary image) in the VCS. That is, the confidential voice transmission device 12 generates the share mask M' _n using the basic matrices X ₀ and X ₁ of the VCS. Then, the confidential voice transmission device 12 performs masking processing on the mixed voice, and collects t or more to create shared voice that can restore the confidential voice.

Also, in order to generate the share mask _M'n , it is necessary to perform the same processing as pixel enlargement in the VCS on the mask M. FIG. Therefore, the share mask generator 24 of the present embodiment multiplies the mask M by m (m is an integer) by increasing the number of frequency bins and the number of time frames to obtain a matrix that satisfies the basic matrices X _{0 and} X ₁ of the VCS. A share mask M' _n is generated so as to satisfy the following equation.

An example of the method for multiplying the mask M by m is as follows.
(1) Increase the number of frequency bins by m times.
(2) Increase the number of time frames by m times.
(3) Multiply the number of frequency bins by _m1 and multiply the number of time frames by _m2 ( _m1 * _m2 =m).

An example of increasing the number of frequency bins is as follows.
(1) By multiplying the sampling frequency of the short-time Fourier transform for generating the shared voice from the mixed voice by m times, the band of the shared voice is expanded to increase the number of frequency bins within the same time width by m times.
(2) Short-time Fourier transform is performed by multiplying the window length by m.

An example of increasing the number of time frames is as follows.
(1) By multiplying the sampling frequency by m and performing a short-time Fourier transform, the number of time frames within the same shift width is increased by m.
(2) A short-time Fourier transform is performed by multiplying the frame shift by 1/m.

In this way, the share mask generation unit 24 performs an increase process for increasing the mask M in at least one of the frequency direction and the time direction in order to generate the share mask _M'n , and the spectrogram of the mask M is m times as large as the spectrogram. Generate a share mask _M'n . Then, the masking unit 26 generates n shared voices by masking the short-time Fourier-transformed mixed voice using the share mask _M'n . Note that the spectrumgram of the mixed speech to be masked is also made m times larger.

A shared audio Y generated by the masking process is represented by Equation 3.

Then, the inverse Fourier transform unit 28 performs an inverse short-time Fourier transform on the shared audio to obtain audio that can be output from the speaker 16, and the speaker 16 outputs the shared audio.

For example, one user can record at least t shared voices out of n shared voices by recording the shared voices output from the speaker 16 multiple times with the microphone 30 of the confidential voice receiving device 14 such as a mobile terminal device. get. In addition, in order to prevent other users from acquiring the shared audio because they do not want to hear the confidential audio, by specifying a time that only a predetermined user can know and outputting the shared audio from the speaker 16, It is possible to reduce the possibility that the user will hear the confidential voice.

The confidential audio receiving device 14 performs synchronization processing to match the starting points of t or more recorded shared audio, and as shown in FIG. Restore hidden audio. Synchronization processing can be realized, for example, by correlating each signal of a plurality of shared voices. Synchronous addition of t shared voices is represented by the following formula 4, and when j consists of elements of Γ _Q , the sum Σ of the share masks M' _n in formula 4 is a mask that leaves the concealed voice among the mixed voices. , the hidden voice is restored.

Note that, unlike images, audio has redundancy associated with short-time Fourier transform between adjacent components of the spectrogram. For this reason, the share mask _M'n is blurred and {0, 1} is not retained, and there is a possibility that the S/N will not be improved by the synchronous addition of the share audio. Therefore, the shared audio may be synchronously multiplied so that 1 and 0 are more clearly separated.

As described above, the confidential voice transmission system 10 of the present embodiment applies visual cryptography (VCS) to the transmission of confidential voice, and embeds the share mask M′ _n generated from the mask M in the spectrogram of the confidential voice. Generate n share voices. When the user collects t or more shared sounds out of n shared sounds, the user can restore the confidential sound, but cannot restore the secret sound when the collected shared sounds are less than t. In addition, the confidential voice transmission system 10 of this embodiment utilizes the fact that it is difficult to solve the problem of separating a signal in which a plurality of sound sources are mixed from a monaural observed signal that does not have spatial information. For encryption, the secret voice transmission system 10 only needs to have an FFT analyzer that performs a short-time Fourier transform on voice. No complex decoder is required.

In this way, the confidential voice transmission system 10 of the present embodiment can distribute voice with high confidentiality and easily restore it.

(Second embodiment)
The confidential voice transmission system 10 of this embodiment determines the similarity between the confidential voice restored by the confidential voice receiving device 14 and the original distributed message. This makes it possible to detect whether the user has acquired the share mask _M'n consisting of the correct qualified set. This makes it possible to detect that the user has been at a specified location for a certain period of time.

FIG. 5 is a functional block diagram of the confidential voice receiving device 14 of this embodiment. The confidential voice receiving device 14 of this embodiment includes a similarity determination unit 36 in addition to the configuration of the confidential voice receiving device 14 of the first embodiment.

The similarity determining unit 36 determines similarity between the voice restored by the voice restoring unit 34 and the message that is the confidential voice. Although this message is previously acquired as digital data from a server (not shown) or the like by the confidential voice receiving device 14 and stored in the storage means, it is made unrecognizable to the user. Note that the similarity determination unit 36 determines the similarity using, for example, an index used for evaluating sound source separation, such as the S/N ratio of the restored voice and message, the voice-to-distortion ratio, or a correlation coefficient. .

(Third Embodiment)
In the first embodiment in which the mixed voice is masked with the share mask _M'n and output from the speaker 16, the redundancy of the short-time Fourier transform may reduce the reconstruction accuracy of the concealed voice. For example, human voices, sounds of musical instruments, etc. have a small number of active time-frequency bins, and the sparsity of sound signals to be transmitted is high. The estimation accuracy of the mask M generated based on such a highly sparsity sound signal may be degraded.

Therefore, in this embodiment, the share mask _M'n and the mixed voice are separately transmitted. For this reason, the confidential voice transmission device 12 of the present embodiment uses noise, such as white noise, whose intensity is equal to or greater than a predetermined value in a predetermined frequency range as a carrier, and shares the voice obtained by masking this noise with the share mask M' _n . It is output from the speaker 16 as voice. For example, when white noise is used as noise, all frequency components are active when averaged over time, so the spectrogram of the share voice is stochastically the same as the share mask _M'n . Also, for example, if the frequency of the hidden voice is limited to 500 to 1 kHz, white noise limited to that band can be used as noise, and pink noise can be used if the hidden voice does not contain much high-frequency components. be done.

FIG. 6 is a functional block diagram of the confidential voice transmission device 12 of this embodiment. , an inverse Fourier transform unit 28 , a speaker 16 , a noise generator 40 , and a Fourier transform unit 42 .

The mask generating unit 22 generates a mask M′ based on the presence or absence of frequency bins in which the frequency components of the concealed audio are greater than the interference sound by a threshold θ or more in the spectrogram obtained by performing the short-time Fourier transform of the concealed audio.

The share mask generation unit 24 generates, based on the mask M, n (2≦t≦n) different share masks M′ _n for restoring the confidential voice by superimposing t or more.

The noise generator 40 generates and outputs noise such as white noise.

The Fourier transform unit 42 performs a short-time Fourier transform on the noise output from the noise generator 40 and outputs the result.

The masking unit 26 masks the spectrogram obtained by performing a short-time Fourier transform on noise with n share masks M′ _n to generate n different share voices.

The inverse Fourier transform unit 28 inverse Fourier transforms the n shared voices generated by the masking unit 26 . Then, the inverse Fourier-transformed n shared voices are output from the speaker 16 .

In this way, when the noise is white noise, the n share voices output from the speaker 16 of the present embodiment are obtained by masking the spectrogram obtained by short-time Fourier transform with n share masks _M'n . It is later generated by inverse Fourier transform. If the noise is pink noise or the like, it can be processed in the same way as white noise by performing filter processing that matches the characteristics of the noise. Note that the speaker 16 outputs the mixed sound separately from the shared sound.

Thus, the speaker 16 of the confidential voice transmission device 12 of the present embodiment functions as a share mask output unit that outputs the share mask M' _n as sound (n share voices) by using noise as a carrier. It also functions as an audio output unit that outputs mixed audio separately from shared audio. For example, the shared audio and the mixed audio may be output from one speaker 16 at different times, or may be output separately from multiple speakers 16 located at different locations.

FIG. 7 is a functional block diagram of the confidential voice receiving device 14 of this embodiment. The confidential voice receiving device 14 restores the mask M from t or more shared voices among the n shared voices output from the speaker 16, and superimposes the restored mask M on the mixed voice output from the speaker 16. By doing so, the confidential voice is restored.

As shown in FIG. 7, the confidential voice receiving device 14 of this embodiment includes a microphone 30, a synchronization section 32, a mask restoration section 50 and a masking section 52. FIG.

The mask restoration unit 50 restores the mask M by superimposing t or more shared voices input from the microphone 30 and synchronized by the synchronization unit 32 . Note that when the number of shared voices collected by the confidential voice receiving device 14 is less than t, the mask restoring unit 50 cannot restore the mask M.

The masking unit 52 performs masking processing on the mixed voice input to the microphone 30 with the mask M restored by the mask restoring unit 50, thereby extracting the confidential voice.

In this way, the confidential voice transmission system 10 of the present embodiment outputs the share mask M' _n as shared voice, and extracts the confidential voice by masking t or more shared voices and mixed voice. As a result, the confidential voice transmission system 10 of the present embodiment can accurately restore the confidential voice from the distributed shared voice and mixed voice.

(Fourth embodiment)
In this embodiment, an extended visual cryptography scheme (EVCS) is applied to secure voice transmission. EVCS displays a cover image for a single share image, but displays a confidential image when t share images are collected, and is proposed in Document 3 below. When a share image is generated using this EVCS, since the share image displays the cover image, there is an advantage that it is difficult for a person who sees the share image to notice that the share image has encrypted the confidential image.
Reference 3 G. Ateniese, C. Blundo, AD Santis, and DR Stinson, “Extended capabilities for visual cryptography,” Theoretical Computer Science, vol.250, no.1, pp.143--161, 2001.

(fundamental matrix of EVCS)
Next, the basic matrices X0 and X1 of EVCS will be explained. In the EVCS, in addition to the pixel values of the secret image, if the _pixel values ( ^binary ) of the cover images C ₁ ^, _. ^,Cn , _X1C1 ^,...,Cn ) is said to be a basic matrix that realizes the access structure Γ when the following three conditions are satisfied.

(Condition 1) Restorable conditions for confidential images:
For all S∈Γ ^* _Q , there exist integers l _s , h _s satisfying 0≦l _s ≦h _s ≦m, and Equation 5 holds for all C ₁ , . . . , C _n .

(Condition 2) Safety conditions:
For all S∈Γ ^* _F and C ₁ ,...,C _n , X ₀ ^C1,...,Cn [S] and X ₁ ^C1,...,Cn [S] can be equal by permuting appropriate columns. .

(Condition 3) Cover image viewing conditions:
For all j=1,...,n, there exist integers l _j , h _j satisfying 0≦l _j <h _j ≦m, and for all C ₁ ,...,C _n except j, the formula 6 holds.

A method of optimizing the fundamental matrices X ₀ and X ₁ that satisfies the above three conditions is proposed in the following documents 4 and 5. Document 4 SJ Shyu, “Threshold visual cryptographic scheme with meaningful shares,” IEEE Signal Processing Letters, vol. 21, no.12, pp.1521--1525, 2014.
Reference 5 K. Sekine and H. Koga, “Optimal basis matrices of a visual cryptography scheme with meaningful shares and analysis of its security,” 2020 International Symposium on Information Theory and Its Applications (ISITA), pp.422--426, 2020 .

(Configuration of the fourth embodiment)
FIG. 8 is a functional block diagram of the confidential voice transmission device 12 of this embodiment, in which the Fourier transform unit 20, the mask generator 22, the share mask generator 24, the masking unit 52, the inverse Fourier transform unit 28, and the speaker 16 are Prepare.

The Fourier transform unit 20 performs a short-time Fourier transform on a mixed voice obtained by mixing j (1≤j≤n) cover voices with the confidential voice. The j cover voices mixed with the concealed voice are different voices.

The mask generation unit 22 generates a mask M based on the presence or absence of frequency bins in which the frequency component of the concealed voice is greater than the cover voice by a predetermined threshold θ or more in the spectrogram obtained by performing the short-time Fourier transform of the mixed voice, j cover masks M ^C _j are generated based on the presence or absence of frequency bins in which the frequency components of the cover audio are larger than those of the concealed audio and other cover audio by a predetermined threshold θ or more.

The share mask generation unit 24 generates n share masks M′ _n based on the mask M generated by the mask generation unit 22 and the j cover masks M ^C _j .

The masking unit 26 generates n different shared voices by masking the short-time Fourier-transformed mixed voice with n share masks _M'n . The inverse Fourier transform unit 28 inverse Fourier transforms the n shared voices generated by the masking unit 26 and outputs them from the speaker 16 .

The confidential voice receiving device 14 of the present embodiment is the same as the confidential voice receiving device 14 of the first embodiment, and superimposes t or more of the n shared voices output from the confidential voice transmitting device 12. By doing so, the confidential voice is restored.

Next, the details of the generation of the share mask M' _n according to this embodiment will be described.

The Fourier transform unit 20 performs a short-time Fourier transform on the mixed voice of the concealed voice and the j cover voices, and obtains the spectrum s(τ,f) of the concealed voice and the spectrum κ _j (τ,f) of the cover voice. .

Then, the mask generation unit 22 sets the mask M to 1 for time-frequency bins in which the sound level |s| of the concealed sound is greater than the sound level |κ _j | of the cover sound by a threshold θ or more, , the mask M is set to 0. This is expressed in Equation 7.

Further, the cover mask M ^C _j is applied to the time-frequency bin whose cover audio level |κ _j | is greater than the other cover audio audio level |κ _j′ | and the confidential audio audio level |s| by a threshold θ or more. 1, and the cover mask M ^C _j is set to 0 for other time-frequency bins. This is expressed in Equation 8.

Here, by setting c _j =M ^C _j (τ, f), the cover mask M ^C _j can be regarded as a cover image (binary image) in EVCS. That is, by using the mask M as the secret voice and the cover mask M ^Cj as the cover voice, the share mask generation unit 24 can generate the share mask by the _same process as the EVCS.

The share mask generation unit 24 multiplies the mask M and the cover mask M ^C _j by m by increasing the number of frequency bins and the number of time frames, and generates n share masks so that the determinant satisfies the basic matrix of the EVCS. Generate.

Then, by performing masking processing on the mixed audio, the cover audio j can be heard as the shared audio j. In other words, if a person listens to only the shared voice alone, they hear a voice different from the hidden voice, and it is difficult to recognize that the hidden voice is encrypted in the shared voice. You can get hidden voice. As a result, the confidential voice transmission system 10 of the present embodiment can distribute shared voice with enhanced confidentiality.

(Fifth embodiment)
The confidential voice transmission system 10 of this embodiment is a system that separately transmits the share mask _M'n and the mixed voice, as in the third embodiment.

The confidential voice transmission device 12 of this embodiment outputs each of the n share masks M' _n as digital data to the confidential voice reception device 14 . The confidential voice receiving device 14 restores the mask M from t or more share masks M′ _n output from the confidential voice transmitting device 12 . As a result, the confidential voice receiving device 14 can directly calculate the mask M from the acquired t share masks _M'n . Then, the confidential voice receiving device 14 restores the confidential voice by masking the mixed voice output from the confidential voice transmitting device 12 on the restored mask M.

FIG. 9 is a functional block diagram of the confidential voice transmission device 12 of this embodiment. A unit 60 is provided. The functions of the Fourier transform unit 20, the mask generation unit 22, and the share mask generation unit 24 are the same as those of the confidential voice transmission device 12 of the third embodiment.

The data transmission unit 60 outputs each of the share masks _M'n generated by the share mask generation unit 24 as digital data. The output destination of the share mask M' _n is the confidential voice receiving device 14 that can communicate with the confidential voice transmitting device 12 . Note that the data transmission unit 60 may communicate with the confidential voice receiving device 14 by short-range communication, for example, and output the share mask M' _n at different times, or may output the share mask M' _n separately at different locations. can be output to

A speaker 16 outputs the mixed sound. For example, the mixed sound may be output from one speaker 16 at different times, or may be output separately from a plurality of speakers 16 positioned at different locations. The time and place at which the mixed sound is output may be the same as the time and place at which the share mask _M'n is output, or may be different.

FIG. 11 is a functional block diagram of the confidential voice receiving device 14 of this embodiment. A conversion unit 76 and a masking unit 78 are provided.

The microphone 30 receives input of mixed voice output from the speaker 16 of the confidential voice transmission device 12 .

The data receiving unit 70 receives the share mask M′ _n transmitted from the data transmitting unit 60 of the confidential voice transmitting device 12 .

The synchronizing unit 72 performs a synchronizing process to align the start points of the n share masks M′ _n in order to restore the mask _M from the n share masks M′ n received by the data receiving unit 70 .

The mask restoration unit 74 restores the mask M by superimposing t or more share masks M′ _n synchronized by the synchronization unit 32 . Note that if the number of share masks M′ _n collected by the confidential audio receiving device 14 is less than t, the mask restoration unit 50 cannot restore the mask M.

The inverse Fourier transform unit 76 inverse Fourier transforms the mask M restored by the mask restore unit 50 .

The masking unit 78 obtains confidential audio by masking the mixed audio input to the microphone 30 with the mask M restored by the mask restoring unit 50 .

Here, if the pixel enlargement ratios m ₁ and m ₂ are known, the mask restoration unit 74 sums m ₁ and m ₂ mask components {0, 1}, and if the sum exceeds a predetermined threshold, The accuracy of estimating the mask M may be further improved by setting the value to 1 if the value does not exceed a predetermined threshold value and to 0 if the value does not exceed the predetermined threshold.

That is, the (τ, f) component in the original mask M is pixel-enlarged to (m ₁ (τ- ₁ )+k ₁ ,m ₂ (f-1)+k ₂ ) in the share mask _M'n . Therefore, the pixel-enlarged mask M is estimated by adding or multiplying t share masks _M'n , and _M'est is generated by the following equation (9). Note that _k1 satisfies _1≤k1≤m1 , _{and k2 satisfies 1≤k2≤m2} .

Note that χ(x) is an indicator function that returns 1 if x is greater than or equal to a threshold and returns 0 if x is less than the threshold. Then the magnitude of M' _est is equal to the original mask M. Also, the threshold value is set to 1 only when all the components in the region are 1, that is, when m ₁ *m ₂ = 1, or to a value where a certain component is 1 by creating a histogram of the entire area, or to a median value There is a method such as

Although the present invention has been described using the above embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. Various changes or improvements can be made to the above-described embodiments without departing from the gist of the invention, and forms with such changes or improvements are also included in the technical scope of the present invention.

For example, the similarity determination unit 36 of the second embodiment is applied to the confidential voice receiving device 14 of the first embodiment, but the present invention is not limited to this. It may be applied to the confidential voice receiving device 14 of the fifth embodiment.

In addition, in the first embodiment, a form of generating a mask M and a share mask _M'n from a mixed sound in which i interfering sounds are mixed with a confidential sound has been described, but the present invention is not limited to this, and the mask M and the share Only the concealed speech may be used without using the mixed speech to generate the mask _M'n . In the case of this form, when mask M is generated, by setting i=0, mask M is generated only with the confidential voice.

Further, in the above-described embodiment, a form in which mixed sound is output from the speaker 16 has been described, but the present invention is not limited to this. Confidential speech may be extracted from the mixed speech received as data.

10 Confidential Audio Transmission System 12 Confidential Audio Transmitting Device 14 Confidential Audio Receiving Device 16 Speaker (Shared Audio Output Unit)
22 mask generation unit 24 share mask generation unit 26 masking unit (share audio generation unit)

Claims (16)

A mask generation unit that generates a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a share mask generation unit that generates, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
Shared voice generation for generating n shared voices, each of which is different, by masking a spectrogram obtained by performing a short-time Fourier transform on the mixed voice obtained by mixing the confidential voice with other voices with the n share masks. Department and
a shared audio output unit that outputs the n shared audios that have undergone inverse Fourier transform;
Secrecy voice transmission device. The other sounds are i interfering sounds,
In a spectrogram obtained by performing a short-time Fourier transform on a mixed sound in which the i interfering sounds are mixed with the concealed sound, the mask generation unit is configured such that the frequency components of the concealed sound are higher than the interfering sounds by a predetermined threshold or more. 2. The secure audio transmission device according to claim 1, wherein said mask is generated based on the presence or absence of large frequency bins. The other sounds are n cover sounds,
In a spectrogram obtained by performing a short-time Fourier transform on a mixed sound obtained by mixing n cover sounds with the cover sound, the mask generation unit is configured such that frequency components of the cover sound are larger than the cover sound by a predetermined threshold or more. The mask is generated based on the presence or absence of frequency bins, and j (1 ≤ j ≤ j) frequency bins (1 ≤ j ≤ n) generate a cover mask,
The share mask generator generates the n share masks based on the mask and the j cover masks.
The confidential voice transmission device according to claim 1. The confidential voice transmission device according to any one of claims 1 to 3;
a confidential voice receiving device that restores the confidential voice by superimposing the t or more of the n shared voices output from the confidential voice transmitting device;
A secret voice transmission system comprising: A mask generation unit that generates a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a share mask generation unit that generates, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
a share mask output unit that outputs the n share masks;
an audio output unit that outputs a mixed audio obtained by mixing i interfering sounds with the confidential audio;
Secrecy voice transmission device. In a spectrogram obtained by performing a short-time Fourier transform on a mixed sound in which the i interfering sounds are mixed with the concealed sound, the mask generation unit is configured such that the frequency components of the concealed sound are higher than the interfering sounds by a predetermined threshold or more. 6. The secure audio transmission device according to claim 5, wherein said mask is generated based on the presence or absence of large frequency bins. The shear mask output unit generates a spectrogram obtained by performing a short-time Fourier transform on noise whose intensity is greater than or equal to a predetermined value in a predetermined frequency range, and then performs an inverse Fourier transform after masking the spectrogram with the n shear masks. 7. The confidential audio transmission device according to claim 5, wherein said n pieces of shared audio are output as sounds. 7. The confidential voice transmission device according to claim 5, wherein said share mask output unit outputs each of said n share masks as digital data. The confidential voice transmission device according to any one of claims 5 to 8;
restoring the masks from the t or more of the n share masks output from the share mask output unit, and applying the restored masks to the mixed audio output from the audio output unit; a confidential audio receiving device that restores the confidential audio by masking;
A secret voice transmission system comprising: The mask generation unit generates the mask by setting frequency bins equal to or greater than the threshold value to 1 and frequency bins less than the threshold value to 0.
The confidential voice transmission device according to any one of claims 1 to 3 and 5 to 8. The share mask generation unit multiplies the mask by m by increasing the number of frequency bins and the number of time frames, and generates the n shares so as to obtain a determinant that satisfies the basic matrix of VCS (Visual Cryptography Scheme). generate a mask,
11. The confidential voice transmission device according to any one of claims 1 to 3, 5 to 8, and 10. 10. The confidential voice transmission system according to claim 4, wherein said confidential voice receiving device comprises a similarity determination unit for determining similarity between restored voice and said confidential voice. A first step of generating a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
A third step of generating n different shared voices by masking a spectrogram obtained by performing a short-time Fourier transform on the mixed voice obtained by mixing the confidential voice with other voices with the n share masks. and,
a fourth step of outputting the n share voices that have undergone inverse Fourier transform;
A fifth step of restoring the confidential voice by superimposing the t or more of the n shared voices that have been output;
A confidential voice transmission method comprising: A first step of generating a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
a third step of outputting the n share masks and outputting a mixed sound obtained by mixing i interfering sounds with the confidential sound;
A fourth step of restoring the mask from the t or more share masks among the output n share masks, and masking the mixed sound with the restored masks, thereby restoring the confidential voice. and,
A confidential voice transmission method comprising: to the computer,
A first step of generating a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
A third step of generating n different shared voices by masking a spectrogram obtained by performing a short-time Fourier transform on the mixed voice obtained by mixing the confidential voice with other voices with the n share masks. and,
a fourth step of outputting the n share voices that have undergone inverse Fourier transform;
A fifth step of restoring the confidential voice by superimposing the t or more of the n shared voices that have been output;
Secrecy voice transmission program for executing to the computer,
A first step of generating a mask based on the presence or absence of a frequency bin having a frequency component greater than or equal to a predetermined threshold in a spectrogram obtained by performing a short-time Fourier transform on the confidential voice;
a second step of generating, based on the masks, n (2≤t≤n) different share masks for restoring the confidential voice by superimposing t or more;
a third step of outputting the n share masks and outputting a mixed sound obtained by mixing i interfering sounds with the confidential sound;
A fourth step of restoring the mask from the t or more share masks among the output n share masks, and masking the mixed sound with the restored masks, thereby restoring the confidential voice. and,
Secrecy voice transmission program for executing

Images