JP2003099077A - Electronic watermark embedding device, and extraction device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an electronic watermark embedding device capable of embedding electronic watermark information corresponding to a sound production structure by noticing a voice generation mechanism of a voice generation source and extracting of electronic dispersively embedding the information in many frequency bands by the extraction, and dealling with many kinds of codings so that the information is not lost even though resampling and compression and restoration of voice data are repeated many times. SOLUTION: The electronic watermark embedding device provided with a voice decomposing section which extracts voice articulatory components and sound source components from inputted voice, an electronic watermark embedding section 13 which embeds the information into the articulatory components, and an electronic watermark voice output section 14 that composites the sound source components and the articulatory components, to which the electronic watermark information is embedded, and outputs electronic watermark embedded voice.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital watermark embedding device, an extracting device and a method.

[0002]

2. Description of the Related Art Conventionally, computer technology has been rapidly developed, and along with it, the development of information and communication infrastructure has been advanced and the communication cost has become very low. Currently, active exchange of information in the digital world is in progress. Is being done. Information such as video and audio to be exchanged has been digitized, and is exchanged as digital content using the Internet, satellite broadcasting, or the like.
As a result, the demand for multimedia has increased, and the amount of information exchanged so far as to be out of the eyes of the author has increased the problem of illegal copying of digital contents, particularly recently.

Characters, figures, music, sound, still images as electronic media,
Since various multimedia contents such as moving images are basically software, it is possible to copy them almost as easily as the original because of their characteristics. In order to protect intellectual property rights such as copyright against such illegal copying, security measures for digital information have been studied and implemented in recent years.

As a general method among such security measures, there are a method of sealing the recording medium itself and a method of encrypting the digital content in order to protect the digital content in the distribution process. However, such a method is insufficient as protection of intellectual property rights. That is, in the case of the encryption technology, it is almost useless for illegal copying after re-decoding and re-distribution.

Further, it is difficult for the user to understand the merit of the content by the method of applying the encryption key or sealing the content, and hinders smooth distribution of excellent digital content. After all, concealment is not always a good idea in the Internet world where the value of content changes depending on how well PR is done. Considering such characteristics and the current situation, it is desirable to embed the identification code for identifying the intellectual property right of the content in the content itself and use it without being perceived by the user in the easiest way possible. .

With such a method, the intellectual property right is protected, and at the same time, the use is restricted to convey the convenience of the content to the user, and the user is asked to purchase the content. It also becomes possible. If an illegal copy or illegal use is found, the identification code can be restored, the intellectual property right of the user can be claimed, and infringement of the intellectual property right can be appealed.

[0007] Due to such advantages, the technique of protecting the intellectual property right of the digital contents is attracting attention as "digital watermark" which is a technique of invisible signature. In the future, many users of electronic media will be able to master this digital watermark technology, embed signatures in their own copyrighted digital content and publish the content in their own unique way. is expected.

On the other hand, under such circumstances, it is conceivable that the "voice" of a human being, that is, the voice itself has a commercial value as a new electronic medium and is subject to intellectual property rights. For example, it can be considered that a general consumer purchases voices of actors, talents, etc., arranges them as they like and synthesizes phrases, and uses them for ringing sounds of mobile phones and alarm clocks. Therefore, in recent years, digital watermarks for voice have been studied, and methods for embedding digital watermark information in voice or extracting digital watermark information from voice have been proposed.

For example, for uncompressed audio data,
A time masking method, a frequency masking method, a dither signal method, an adaptive PCM method, etc. have been proposed. Further, for uncompressed music data, an echo method, a spread spectrum method, a modified DCT method and the like have been proposed. Furthermore, for compressed ones, vector quantization method, polar code method,
Source pulse method, parity bit method, etc. have been proposed.

[0010]

However, in the conventional method of embedding the digital watermark information in the voice or extracting the digital watermark information from the voice, it is considered that the digital watermark disappears due to a filtering process or the like. To be That is, the above-mentioned conventional method focuses on the speech waveform or coding parameter after utterance, but when focusing on the speech waveform, even if the digital watermark information is embedded, the digital watermark information disappears by the filtering process or the like. Can be considered. Also, when focusing on encoding, there is a possibility that the digital watermark information can be removed, for example, by encoding the reproduced voice again using another method, and it is applied to the voice encoded by another method. I can't.

When the digital watermark information is embedded in a specific frequency band of the audio wavelength, it tends to be erased by repeating resampling or audio data compression and decompression. Furthermore, it is difficult for the above-mentioned conventional method to deal with a large number of types of encoding of voice data.

The present invention solves the above conventional problems,
Focusing on the sound generation mechanism at the sound source,
Digital watermark information is embedded according to the pronunciation structure, and
By extracting, the digital watermark information is distributed and embedded in many frequency bands, does not disappear even if resampling and compression / decompression of audio data are repeated, and supports many types of encoding. It is an object of the present invention to provide a digital watermark embedding device, an extracting device and a method that can be performed.

[0013]

To this end, in the digital watermark embedding device of the present invention, a voice decomposing unit for extracting an articulatory component and a sound source component from an input voice and an electronic watermarking information embedded in the articulatory component. The digital watermark embedding unit includes a digital watermark embedding voice output unit that synthesizes the sound source component and the articulatory component in which the digital watermark information is embedded, and outputs a digital watermark embedding voice.

In another digital watermark embedding device of the present invention, the articulation component is a parameter expressing the articulation of voice, and the sound source component is a pitch component of voice.

In still another digital watermark embedding device of the present invention, the parameter is LSP.

In still another digital watermark embedding apparatus of the present invention, it further corresponds to an original codebook in which numerical values of vectors corresponding to the LSP are stored, and an information code having a logical value of 0 or 1. And a change codebook in which the numerical values of the vector changed to are stored.

In still another digital watermark embedding device of the present invention, the digital watermark embedding unit further accesses the change code book to obtain a numerical value corresponding to the information code of the digital watermark information, The numerical value of the vector corresponding to the LSP extracted from the input voice is changed to create the LSP in which the digital watermark information is embedded.

In the digital watermark extracting apparatus of the present invention,
A voice decomposition unit that extracts articulatory components from the input voice,
And a digital watermark extraction unit that extracts digital watermark information from the articulatory component.

In another digital watermark extracting apparatus of the present invention, the articulation component is a parameter expressing the articulation of voice.

In still another digital watermark extracting device of the present invention, the parameter is LSP.

[0021] In still another digital watermark extracting apparatus of the present invention, the apparatus further corresponds to an original codebook in which the numerical value of the vector corresponding to the LSP is stored and an information code having a logical value of 0 or 1. It has a modified codebook in which the modified values of the vector are stored.

In still another digital watermark extracting apparatus of the present invention, the digital watermark extracting section further accesses the original codebook and the modified codebook, and corresponds to the LSP extracted from the input voice. The numerical value of the vector is compared with the numerical values stored in the original codebook and the modified codebook to extract the information code of the digital watermark information.

In the digital watermark embedding method of the present invention,
An articulation component and a sound source component are extracted from the voice, electronic watermark information is embedded in the articulation component, and the sound source component and the articulation component in which the electronic watermark information is embedded are combined to output a digital watermarked voice.

In another digital watermark embedding method of the present invention, the articulation component is a parameter expressing a voice articulation, and the sound source component is a voice pitch component.

In still another digital watermark embedding method of the present invention, the parameter is LSP.

In still another method of embedding a digital watermark of the present invention, a change code book in which the numerical value of the vector changed so as to correspond to the information code is stored is accessed, and the information code of the digital watermark information is accessed. Is obtained and the value of the vector corresponding to the LSP extracted from the voice is changed to create the LSP in which the digital watermark information is embedded.

In the digital watermark extracting method of the present invention,
An articulation component is extracted from the voice, and electronic watermark information is extracted from the articulation component.

In another digital watermark extracting method of the present invention, the articulation component is a parameter expressing the articulation of voice.

In still another digital watermark extracting method of the present invention, the parameter is LSP.

In still another digital watermark extracting method of the present invention, the original codebook in which the numerical value of the vector corresponding to the LSP is stored, and the numerical value of the vector changed so as to correspond to the information code are The stored modification codebook is accessed, the numerical value of the vector corresponding to the LSP extracted from the voice is compared with the numerical values stored in the original codebook and the modified codebook, and the information code of the digital watermark information is extracted. To do.

[0031]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 2 is a conceptual diagram showing a basic concept in the embodiment of the present invention, and FIG. 3 is a conceptual diagram showing a voice generation model in the embodiment of the present invention.

In the present embodiment, the digital watermark information is embedded, paying attention to the sound generation mechanism. In this case, the electronic watermark information is embedded in the articulation part of the voice model.

That is, the voice that is a human voice is
As shown in FIG. 2, it is generated by a combination of sound source, articulation, and radiation. Here, the sound source is generated mainly by the interruption of the expiratory flow due to the vibration of the vocal cords. Articulation is the action of adjusting the shape of the part called the vocal tract above the larynx in order to emit various speech sounds. The radiation is that the sound wave formed in the vocal tract by the sound source and the articulation is emitted from the lips to the space as an acoustic sound waveform.

Therefore, in the present embodiment, the embedding of digital watermark information is performed for articulation, which is the process of radiation. That is, the electronic watermark information is embedded in the articulatory component. This can be realized by creating a voice generation model of a person as a speaker, changing the voice generation model itself, and creating a voice generation model with a digital watermark. Then, the sound radiated by the voice generation model without the operation of embedding the digital watermark information becomes the original voice without the digital watermark, and the sound radiated by the voice generation model with the digital watermark becomes the voice with the digital watermark. Become.

By the way, the operation of embedding the digital watermark information in the present embodiment can be applied to any kind of voice of any speaker. However, due to the nature of the digital watermark, the source is identified. It is suitable for targeting voices that are highly required to be identified. Here, the voice that has a high necessity of identifying the source and certifying the identity is, for example, voice of voice actors such as movies and animations, voices of popular idols, popular singers, and the like. These voices are considered to have commercial value in their own right. Furthermore, these sounds can be appropriately synthesized and used as a calling sound for an alarm clock, a mobile phone, or the like.

The voices of witnesses and appraisers in court proceedings and similar procedures are also suitable for embedding digital watermark information. in this case,
By embedding the electronic watermark information, it is considered that the evidence value of the testimony becomes high. Further, it is considered that the value of the evidence is also high in the case of the voice of the conversation intercepted in the investigation process of the police or the like.

Here, when creating a voice generation model,
A database that expresses the articulation of each speaker is required. In the present embodiment, LSP (Line Spec) is used as a parameter expressing articulation which is an articulation component of voice.
rum Pair). Then, a codebook for each speaker is created in advance using the LSP,
Model articulation based on the codebook. Also,
The codebook is a representative value indicating the articulation of the speaker, and the articulation of the speaker can be changed by changing the codebook.

The LSPs appear as a pair before and after a formant (formant: a mountain where energy is concentrated in a specific frequency band on a frequency spectrum for voice) which is a resonance frequency of the vocal tract. . Here, since the speech waveform has a correlation on the time axis, it is assumed that the sample value of the speech has linear predictability, and the sample values before and after the sample value before and after so as to minimize the root mean square error with the original speech waveform. The prediction coefficient is coded (parameterized). In this case, there are several means for encoding the prediction coefficient, and the LSP is one of them.

In the present embodiment, parameters for realizing the basic concept as shown in FIG. 2 are extracted from the voice waveform of the original voice and are resynthesized as voice with digital watermark. There are various parameters, but in the present embodiment, LSP is used as a parameter expressing articulation, which is the articulation component of voice. This is because LSP is optimal as a parameter for expressing vocal tract information when synthesizing and restoring speech. Further, the voice also includes a pitch component as a sound source component, as shown in FIG. Therefore, in order to restore the voice, not only the LSP but also a parameter called pitch as a pitch component is necessary. It should be noted that the pitch is used only for voice synthesis, and no embedding operation of digital watermark information for the pitch is performed.

Here, the pitch is a parameter representing the pitch of a voice. When the tension of the vocal cords is high and the air pressure from the lungs is high, the opening / closing cycle of the vocal cords, that is, the vibration cycle is shortened and the pitch of the sound source is increased, so that the pitch is increased. In the opposite case, the pitch becomes low. In addition, the vibration cycle of the vocal cords is called the fundamental cycle, and the reciprocal of this is called the fundamental frequency. Then, an accent or intonation sensation is added by the temporal change of the basic cycle.

Note that the sound source, articulation, and radiation of the voice and the parameters have the relationship shown in FIG.

Next, the digital watermark embedding device and its operation will be described.

FIG. 1 is a block diagram showing the operation of the digital watermark embedding device according to the embodiment of the present invention, FIG. 4 is a block diagram showing the configuration of the digital watermark embedding device according to the embodiment of the present invention, and FIG. FIG. 6 is a diagram showing a codebook changing method in the embodiment of the present invention, and FIG. 6 is a diagram showing an order in which the codebook is used in the vector quantization in the embodiment of the present invention.

In FIG. 4, reference numeral 10 denotes a digital watermark embedding device, which includes a CPU, an MPU or other arithmetic means, a semiconductor memory,
Storage means such as magnetic disk, keyboard, mouse, input means such as microphone, CRT, liquid crystal display,
The computer is equipped with a printer, an output unit such as a loudspeaker, a communication interface, and the like. From the viewpoint of function, the digital watermark embedding device 10 decomposes the original voice as the input voice, the original voice input unit 11 into which the original voice of the speaker is input, and the LSP as the articulatory component. A voice decomposition unit 12 for extracting a pitch component as a sound source component,
It has a digital watermark embedding unit 13 that embeds digital watermark information in the extracted LSP, and a digital watermark embedded audio output unit 14 that combines the LSP embedded with digital watermark information and a pitch component to output a digital watermark embedded audio.

Here, the digital watermark embedding device 10 is
It has a codebook database 20 for storing the codebook. The codebook database 20
May be a database built in the storage means of the digital watermark embedding device 10 or may be a database built in an external storage means. The codebook database 20 is a file that stores an original codebook as an original codebook created by using the LSP extracted from the speaker's voice.
21, a CBK _w0 22 which is a file storing a modified code book expressing “0” created by modifying the original code book stored in the CBK 21, and an original code book stored in the CBK 21. It is provided with a CBK _w1 23 that is a file that stores a change codebook that expresses "1" created by making changes. Here, the "0" and "1" are information codes of digital watermark information. It is to be noted that the CBK21, CBK _w0 2
2 and CBK _w1 23 respectively store the original codebook and the modified codebook created for each speaker.

Then, the original codebook and the modified codebook are prepared in advance for each speaker, as shown in FIG. First, as preprocessing, using a learning voice of a speaker i (1 ≦ i ≦ n),
Extracting an LSP, creating an original codebook CBK _{i in} which numerical values of vectors corresponding to the LSP are stored,
Store in CBK21. Then, change the values stored in the original codebook CBK _i ,
A modified codebook CBK _iw0 representing "0", and
A modified codebook CBK _iw1 representing “1” is created and _stored in CBK _w0 22 and CBK _w1 23, respectively.

Here, in the case of extracting the LSP, first, the processing unit (frame) is determined by windowing the voice waveform of the speaker i. Then, a linear prediction analysis (L
LS is performed from the linear prediction coefficient obtained as a result of performing the inward predictive coding).
Extract P. Then, LSPs are extracted for all the frames, and a codebook is created using the LBG + splitting algorithm. This is a work for determining a representative value of a vector corresponding to an LSP calculated from each frame, that is, an LSP vector, and a set of representative vectors for the number of dimensions is the original codebook C.
BK _i .

The created original codebook
CBK_iChange the content of the
Create a modified codebook for embedding information. In this case, each
Make one change location in the level and add two types of changes
Change Codebook CBK _iw0, CBK_iw1Create
It In addition, changes will be made for all levels. And change
Original codebook CBK
_iThe original code with the same value as
Dobook CBK_iRepresents digital watermark information "none"
To use.

For example, as shown in FIG. 5, with respect to all the levels of the original codebook CBK _i , a change is made in a specific dimension to change the codebooks CBK _iw0 , CBK.
A case of creating _iw1 will be described. In this case, LS
It is assumed that P is 16-dimensional and has levels of 1 to 256, and the numerical value of the third dimension of the original codebook CBK _i is moved by 0.2 in the negative direction to change the codebook CB.
It is _assumed that K _iw0 is created and moved in the plus direction by 0.2 to create a modified codebook CBK _iw1 .

As a result, as shown in FIG.
Original code book CBK_iNumerical value of the third dimension of 1.
0 is changed to 0.8, and the change as shown in FIG.
Sara Codebook CBK_iw0Is created. In addition, the above
Original Code Book CBK _iNumerical value of the 3rd dimension of 1.0
Is changed to 1.2, and the change is as shown in Fig. 5 (c).
Codebook CBK_iw1Is created. Therefore,
The distribution of the LSP spectrum as shown in (d) is
5 (e) is changed.

[0052] In this way, the learning for the voice of a speaker i, the original code book CBK _i, change the codebook _CBK iw0 to represent a "0", and, to change the codebook CBK _iw1 to represent a "1" It is created and _stored in advance in the CBK 21, CBK _w0 22 and CBK _w1 23, respectively.

Next, when the original voice of the speaker i is input from the original voice input unit 11, the voice decomposition unit 12 extracts the LSP and the pitch component from the original voice. Then, the digital watermark embedding unit 13 performs a digital watermark information embedding operation on the LSP. In this case, the electronic watermark embedding unit 13 accesses the codebook database 20, a vector corresponding to the extracted the LSP, vector quantization is performed using the original codebook CBK _i, representing the original codebook CBK _i Replace with a representative value.

Then, the digital watermark embedding unit 13 uses a bit string as an information code of the digital watermark information, for example, as shown in FIG.
, The modified codebook CBK _iw0 that represents “0” in the LSP sequence that has been vector-quantized by the original codebook CBK _i , and “1” are represented in accordance with “1010110 ...”. Vector quantize using the modified codebook CBK _iw1 . In this case, the third-dimensional numerical value 1.0 of the original codebook CBK _i as shown in FIG. 6A is 0.8 and 1.2.
Change code book CBK expressing "0" changed to
_iw0 and modified codebook CBK representing "1"
_iw1 is used in the order as shown in FIG.
As a result, the digital watermark information embedding operation is completed, and LSP ′ as a digital watermark embedded LSP in which the digital watermark information is embedded is created.

The vector quantization means the target vectors x ₁ (0), x ₁ (1), ..., X ₁ (L) and the original codebook CBK _i and the modified codebook CBK.
_iw0 , modified codebook CBK _iw1, etc., Euclidean distance d _i with a vector registered in the codebook,

[0056]

[Formula 1]

Is to calculate the index i that minimizes. Here, c _i is a vector registered in the codebook.

Then, the digital watermark embedded audio output unit 1
4 synthesizes the LSP 'and the pitch component to create a digital watermarked voice. The pitch component is the pitch component extracted by the voice decomposing unit 12 and is not changed. Finally, the electronic watermark embedded voice is output by the electronic watermark embedded voice output unit 14 as the electronic watermark embedded voice of the speaker i.

Next, the digital watermark extracting device and its operation will be described.

FIG. 7 is a block diagram showing the configuration of the digital watermark extracting apparatus according to the embodiment of the present invention, FIG. 8 is a block diagram showing the operation of the digital watermark extracting apparatus according to the embodiment of the present invention, and FIG. 9 is the present invention. FIG. 6 is a diagram showing a procedure of extracting digital watermark information in the embodiment of FIG.

In FIG. 7, reference numeral 30 denotes a digital watermark extraction device, which is a CPU, MPU or other arithmetic means, semiconductor memory,
Storage means such as magnetic disk, keyboard, mouse, input means such as microphone, CRT, liquid crystal display,
The computer is equipped with a printer, an output unit such as a loudspeaker, a communication interface, and the like. The digital watermark extracting device 30 may be configured integrally with the digital watermark embedding device 10. From the viewpoint of function, the digital watermark extracting device 30 decomposes the voice input unit 31 to which the voice to be the target of extracting the digital watermark information is input, the voice into which the digital watermark is input, and the voice to extract the LSP. It has a decomposition unit 32 and a digital watermark extraction unit 33 that extracts digital watermark information from the extracted LSP.

Here, the digital watermark extracting device 30
It has a codebook database 20 for storing the codebook. The codebook database 20
Has the same configuration as the codebook database 20 of the digital watermark embedding device 10 and stores the same data. Therefore, the description of the codebook database 20 and the files CBK21, CBK _w0 22 and CBK _w1 23 will be omitted. The codebook database 20 may be a database built in the storage means of the digital watermark extraction device 30 or a database built in an external storage means. Further, the codebook database 2 of the digital watermark embedding device 10
You may make it share 0.

When extracting the digital watermark information,
When the voice of the speaker i to be inspected as to whether or not the digital watermark information is embedded is input from the voice input unit 31,
The voice decomposing unit 32 extracts the LSP from the voice. In this case, it is not necessary to extract the pitch as in the case of embedding the digital watermark information. Then, the digital watermark extraction unit 33
Accesses the codebook database 20 and performs a digital watermark extraction operation on the LSP.

Here, the extraction of the digital watermark information is performed by extracting LSP 'which is the LSP of the voice from the voice to be inspected and judging by the distance from the codebook. First, as shown in FIG. 9, 16-dimensional LSP ′ is extracted for each frame. Then, the selection level is set in order to check which level of the codebook was used for the synthesis. In this case, 'vector corresponding to, i.e., 16-dimensional LSP' extracted LSP except one dimension out of the vector, the file CBK21, CBK _w0 22 and CB
K _w1 23 on the stored and has three codebooks, i.e., changing the code representing the original codebook CBK _i, a "0" book CBK _IW0, and Euclidean distance between the change codebook CBK _iw1 to represent "1" d
Calculate _i . Note that the calculation is performed for all 16 dimensions. Then, the CBK _i CBK _iw0 and C
_Based on the Euclidean distance d _i with BK _iw1 , the level (1-256) with the shortest distance is determined, and when the level has the same value in the CBK _i CBK _iw0 and CBK _iw1 , the watermark is _{added to} that level. Judge that the information is embedded.

Then calculate the distance between all 16-dimensional LSP's and the original codebook CBK _i ,
When embedding digital watermark information in the digital watermark embedding device 10, a threshold value is set in accordance with a value changed to create the modified codebooks CBK _iw0 and CBK _iw1 , for example, 0.2. To do. And 1 of the voice to be inspected
Numerical value of 6-dimensional LSP 'vector (LSP'0-LS
P′15) and the numerical values (LSP0 to 16) of the 16-dimensional LSP vector stored in the original codebook CBK _i.
As shown in FIG. 9, when the absolute value of the difference between the numerical value of the LSP ′ vector and the numerical value of the LSP vector is larger than a threshold value, it is determined that digital watermark information is included. . Furthermore, when it is determined that the digital watermark information is included, the numerical values (LSP'0 to LSP'15) of the LSP 'vector and the modified codebook CB are included.
16-dimensional LS stored in K _iw0 and CBK _iw1
The distance between the numerical value (LSP0 _{w0 to} LSP15 _w0 ) and (LSP0 _{w1 to} LSP15 _w1 ) of the P vector is calculated, and the distance is replaced with the information represented by the codebook having the shorter distance.
It is determined whether it is "0" or "1".

As a result, for example, as shown in FIG. 8, the information code of the digital watermark information "1010110 ..." is extracted. Then, the information code string is
It is determined whether the information code is the same as the information code of the digital watermark information embedded by the digital watermark information embedding operation in the digital watermark embedding device 10.

When the digital watermark information embedding operation is performed by changing the codebook like the digital watermark embedding device 10 in the present embodiment, changing the contents of the codebook changes the LSP frequency. Since the change is equivalent to the change, the change of the codebook may change the sound quality of the synthesized voice.

Therefore, in order to "embed digital watermark information without degrading the quality of contents", to what extent can the LSP extracted from the waveform of the original voice be changed, and the changed LSP is used. It is necessary to evaluate to what extent the synthesized speech has a change in sound quality.

When the LSP analysis order is 16 and the 16-dimensional LSP is changed as in the digital watermark embedding device 10 of the present embodiment, of the 16-dimensional LSP,
It is necessary to consider which order the LSP frequency is changed, how to combine and perform the embedding operation, and the direction (+, −) in which the LSP frequency is changed. Further, it is conceivable to control the ratio of embedding in all frames by a function, or a burst embedding operation using the auditory masking characteristic.

As described above, in the present embodiment, the digital watermark embedding device 10 uses the LSP as the articulatory component and the pitch decomposition component as the sound source component from the input voice, and the LSP. It has a digital watermark embedding unit 13 for embedding the digital watermark information, and a digital watermark embedded voice output unit 14 for synthesizing the LSP in which the digital watermark information is embedded and the pitch component to output a digital watermark embedded voice.

Further, in the present embodiment, the digital watermark extracting device 30 uses the input voice as an articulation component L
It has a voice decomposing unit 32 for extracting SP and a digital watermark extracting unit 33 for extracting digital watermark information from the LSP.

Therefore, since the digital watermark information is embedded and extracted corresponding to the pronunciation structure of voice,
It is distributed over many frequency bands. Therefore, even if the sound is resampled many times or the sound is subjected to the compression processing and the decompression processing many times, the digital watermark information is not erased. It is also possible to support encoding of many types of audio signals. Furthermore, the quality of the voice does not deteriorate.

The present invention is not limited to the above-mentioned embodiments, but can be variously modified within the scope of the present invention, and they are not excluded from the scope of the present invention.

[0074]

As described above in detail, according to the present invention, by embedding and extracting the digital watermark information corresponding to the pronunciation structure of the voice, the digital watermark information is dispersed in many frequency bands. Since it is embedded, it does not disappear even if resampling and compression / decompression of audio data are repeated, and it is possible to cope with many types of encoding.

[Brief description of drawings]

FIG. 1 is a block diagram showing an operation of a digital watermark embedding device according to an embodiment of the present invention.

FIG. 2 is a conceptual diagram showing a basic concept in the embodiment of the present invention.

FIG. 3 is a conceptual diagram showing a voice generation model in the embodiment of the present invention.

FIG. 4 is a block diagram showing a configuration of a digital watermark embedding device according to an embodiment of the present invention.

FIG. 5 is a diagram showing a codebook changing method according to the embodiment of the present invention.

FIG. 6 is a diagram showing an order of using a codebook in vector quantization according to the embodiment of the present invention.

FIG. 7 is a block diagram showing a configuration of a digital watermark extracting device according to an embodiment of the present invention.

FIG. 8 is a block diagram showing an operation of the digital watermark extracting device in the embodiment of the present invention.

FIG. 9 is a diagram showing a procedure of digital watermark information extraction according to the embodiment of the present invention.

[Explanation of symbols]

10 Digital watermark embedding device 12, 32 Speech decomposition unit 13 Digital watermark embedding unit 14 Digital watermarked audio output section 30 Digital Watermark Extractor 33 Digital Watermark Extraction Unit

Claims (18)

[Claims] 1. A voice decomposition unit for extracting an articulatory component and a sound source component from an input voice, and a digital watermark embedding unit for embedding digital watermark information in the articulatory component.
(C) A digital watermark embedding device, comprising: a sound source component and an articulatory component in which the digital watermark information is embedded, and a digital watermark embedded audio output unit for outputting a digital watermark embedded audio. 2. The digital watermark embedding device according to claim 1, wherein the articulation component is a parameter expressing a voice articulation, and the sound source component is a voice pitch component. 3. The digital watermark embedding device according to claim 2, wherein the parameter is LSP. 4. An original codebook in which a numerical value of a vector corresponding to the LSP is stored, and a modified codebook in which a numerical value of the vector modified to correspond to an information code having a logical value 0 or 1 is stored. Claim 3 having
The electronic watermark embedding device as described in 1. 5. The digital watermark embedding unit accesses the modified codebook, obtains a numerical value corresponding to the information code of the digital watermark information, and obtains a vector corresponding to the LSP extracted from the input voice. Change the value of
The digital watermark embedding device according to claim 4, wherein an LSP in which the digital watermark information is embedded is created. 6. A digital watermark comprising: (a) a voice decomposing unit for extracting an articulatory component from the input voice; and (b) a digital watermark extracting unit for extracting digital watermark information from the articulatory component. Extractor. 7. The digital watermark extracting apparatus according to claim 6, wherein the articulation component is a parameter expressing articulation of voice. 8. The digital watermark extracting apparatus according to claim 7, wherein the parameter is LSP. 9. An original codebook in which a numerical value of a vector corresponding to the LSP is stored, and a modified codebook in which a numerical value of the vector modified to correspond to an information code having a logical value 0 or 1 is stored. Claim 8 having
The digital watermark extracting device according to. 10. The digital watermark extracting unit accesses the original codebook and the modified codebook, and sets the numerical value of the vector corresponding to the LSP extracted from the input speech and the original codebook and the modified codebook. The digital watermark extracting device according to claim 9, wherein the information code of the digital watermark information is extracted by comparing it with a stored numerical value. 11. (a) Extracting an articulatory component and a sound source component from a voice, (b) embedding digital watermark information in the articulatory component, and (c) a sound source component and an articulatory component in which the digital watermark information is embedded. A method for embedding a digital watermark, which comprises synthesizing audio and outputting audio with a digital watermark. 12. The digital watermark embedding method according to claim 12, wherein the articulatory component is a parameter expressing the articulation of voice, and the sound source component is a pitch component of voice. 13. The digital watermark embedding method according to claim 12, wherein the parameter is LSP. 14. A modified code book, in which a numerical value of a vector corresponding to the LSP modified to correspond to an information code is stored, is accessed to acquire a numerical value corresponding to the information code of the digital watermark information, and the voice The LSP in which the digital watermark information is embedded is created by changing the numerical value of the vector corresponding to the LSP extracted from the.
The method for embedding a digital watermark according to item 3. 15. (a) Extracting an articulatory component from voice,
(B) A digital watermark extracting method characterized by extracting digital watermark information from the articulatory component. 16. The digital watermark extracting method according to claim 15, wherein the articulation component is a parameter expressing articulation of voice. 17. The digital watermark extracting method according to claim 16, wherein the parameter is LSP. 18. Accessing an original codebook in which a numerical value of a vector corresponding to the LSP is stored and a modified codebook in which a numerical value of the vector modified to correspond to an information code is stored, 18. The digital watermark extracting method according to claim 17, wherein the numerical value of the vector corresponding to the extracted LSP is compared with the numerical values stored in the original codebook and the modified codebook to extract the information code of the digital watermark information.