JP2002169579A - Device for embedding additional data in audio signal and device for reproducing additional data from audio signal - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device for embedding additional data in an audio data with small possibility that the additional data are destroyed when a voice signal is processed. SOLUTION: This device has a polarity inversion part 13 which inverts the polarity of the audio signal inputted to an input terminal 11 and a border detection part 14 which detects a syllable border of the audio signal, and outputs the audio signal having its polarity inverted by syllables corresponding to the additional data inputted to an input terminal 12 from an output terminal 17.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to a device for embedding additional data in an audio signal such as a voice signal or a tone signal, and a device for reproducing additional data from an audio signal in which the additional data is embedded.

[0002]

2. Description of the Related Art As a method of embedding another data (additional data) in an audio signal such as a voice signal or a musical sound signal so as to be inaudible, a method using an auditory masking characteristic has been proposed. Auditory masking characteristics
The low-frequency component whose frequency or time is close to the high-level component is a property of human hearing that it is difficult to hear.

Accordingly, by assigning the former component called a masker as a main audio signal to be transmitted, and assigning additional data to the latter component called a masky, the additional data can be embedded in the audio signal. According to this method, additional data can be embedded at a relatively high bit rate, and application to data hiding, digital watermarking, and the like can be expected.

[0004]

The method of embedding the additional data in the audio signal using the auditory masking characteristic has a problem that the additional data is destroyed by the processing of the audio signal. For example, MP
MPE such as 3 (MPEG-1 Audio Layer-III)
G audio or ATRAC (Adaptive Transform
In high-quality and high-efficiency audio compression encoding technology such as Acoustic Coding, compression is performed using the auditory masking characteristic. Therefore, if such audio compression encoding processing is performed on an audio signal in which additional data is embedded by the above-described method, the additional data will be destroyed.

[0005] Further, processing not utilizing the auditory masking characteristic, for example, A / D (analog-digital conversion) -D /
Even when the processing of A (digital-analog conversion) is performed on the audio signal in which the additional data is embedded, the additional data may be similarly destroyed.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a device for embedding additional data in an audio signal and a device for reproducing additional data from an audio signal, in which the additional data is less likely to be destroyed by processing into the audio signal.

[0007]

In order to solve the above-mentioned problems, the present invention utilizes the property that human hearing is insensitive to the inversion of the polarity of an audio signal. By embedding additional data and detecting the polarity inversion of the audio signal,
Play the embedded additional data.

That is, the present invention basically embeds the binary additional data in the audio signal by inverting the polarity of the audio signal in accordance with the binary additional data for each predetermined unit and outputting the inverted signal. Features. When the audio signal is an audio signal, for example, a syllable boundary of the audio signal is detected, and the polarity of the audio signal is inverted for each syllable unit based on the syllable boundary and output.

For detecting a syllable boundary, for example, an input audio signal is divided into frames, and the autocorrelation of a residual signal obtained by linear prediction analysis of the audio signal of each frame is obtained to obtain a modified autocorrelation function. The voiced section and the unvoiced section are determined from the modified autocorrelation function and the low-frequency energy of each frame of the audio signal, and the syllable boundary is determined from the energy of the frame determined as the unvoiced section of the audio signal.

In the present invention, the polarity of the audio signal in which the additional data is embedded is determined for each predetermined unit by inverting the polarity according to the binary additional data for each predetermined unit as described above. The basic feature is to reproduce the embedded additional data. If the audio signal in which the additional data is embedded is a voice signal and the polarity of the audio signal is inverted in accordance with the additional data for each syllable unit, a syllable boundary of the audio signal is detected, and based on the syllable boundary. The additional data can be reproduced by determining the polarity of the audio signal for each syllable unit.

Here, the syllable boundary is detected by dividing an input audio signal into frames and performing auto-correlation of a residual signal obtained by performing linear prediction analysis on the audio signal of each frame, as in the case of embedding additional data. The voiced section and the unvoiced section are determined from the modified autocorrelation function and the low-frequency energy of each frame of the audio signal, and the syllable is determined from the energy of the frame determined as the unvoiced section of the audio signal. This is done by determining the boundaries. In reproducing the additional data, the polarity of the audio signal is determined for each syllable unit by taking a majority decision of the polarity of the residual signal of the frame determined to be a voiced section for each syllable sandwiched between the determined syllable boundaries. It is performed by

[0012]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a schematic configuration of an apparatus for embedding additional data in an audio signal according to an embodiment of the present invention. In FIG. 1, an input terminal 11 receives an audio signal such as a digitized voice signal (language voice) or a tone signal, and another input terminal 12 adds a binary signal to be embedded in the audio signal. The data is entered.

An audio signal from an input terminal 11 is input to a polarity inversion section 13 and a boundary detection section 14. The polarity inverting unit 13 inverts the polarity of the input audio signal (multiplies the amplitude value by -1). Boundary detector 14
Is for detecting a predetermined boundary of the input audio signal. Specifically, when the audio signal is a voice signal, for example, a syllable boundary is detected as described later, and when the audio signal is a musical tone signal, , A silent section in the music, a space between the music, and the like.

The additional data from the input terminal 12 is temporarily stored in the data buffer 15. The additional data held in the data buffer 15 is read out one bit at a time each time a boundary is detected by the boundary detection unit 14. The additional data read from the data buffer 15 is supplied to the switch 16 as a control signal.

The switch 16 receives the audio signal from the input terminal 11 and the polarity-inverted audio signal from the polarity inversion unit 13 as inputs, and outputs the signal to the boundary detection unit 1.
Each time a boundary is detected in step 4, switching is performed in accordance with the value of each bit of the additional data read from the data buffer 15 until the next boundary is detected. The signal is selected and output to the output terminal 17.

For example, if the bit of the additional data read from the data buffer 15 is “1” when the boundary is detected by the boundary detector 14, the switch 16 selects the audio signal from the input terminal 11. Is done. When the bit of the additional data read from the data buffer 15 is “0”, the switch 16 selects the audio signal whose polarity has been inverted by the polarity inverting unit 13.

In order to minimize the effect of the discontinuity at each boundary when the switch 16 is switched, for example, when the switch 16 is switched, the level of the signal selected so far is gradually reduced, and The level of the signal to be transmitted may be gradually increased. Specifically, a trapezoidal window having linear rising and falling characteristics in a section of a predetermined time length (for example, about 1 to 10 ms) at both ends of a section between the boundaries of the audio signal is set to the switch 16. It can be realized by applying to

In FIG. 1, the polarity inversion section 13 and the switch 16 are described separately for easy understanding. However, the switch 16 may be omitted and the polarity inversion section 13 may have a gain control function. In this case, the polarity inverting unit 1 is switched according to the additional data read from the data buffer 15.
3 has a linear rising and falling characteristic at the start and end of the polarity inversion, that is, at a predetermined time length at both ends between the audio signal boundaries. The configuration may be such that the gain is controlled as described above.

From the output terminal 17 in this manner, an audio signal whose polarity is inverted according to the additional data in units of segments (for example, syllables) sandwiched between the boundaries according to the value of each bit of the additional data. Is output. That is, the waveform of the segment corresponding to the “0” bit of the additional data is not inverted, and the waveform of the segment corresponding to the “1” bit of the additional data is inverted. It is embedded in the audio signal (audio signal) as the presence / absence of the polarity inversion of one waveform.

The audio signal in which the additional data is embedded as to the presence or absence of the polarity reversal of the unit waveform has no particularly large deterioration in human auditory characteristics, and the audio signal is more compared with the method using the auditory masking characteristics. There is very little possibility that embedded additional data will be destroyed and become unreproducible by various processing processes such as compression encoding and A / D-D / A.

FIGS. 2A and 2B show this situation in the case where the input audio signal is an audio signal. FIG. 2A shows the original audio signal waveform, and FIG. 2B shows the polarity inversion in syllable units according to the additional data. 3 shows the obtained audio signal waveforms.
In FIG. 2B, the polarity of the syllable waveform with an arrow is inverted. If the input audio signal is a tone signal, the polarity is inverted in accordance with the additional data, for example, for each song or for each interval between silent intervals in the song, thereby similarly inverting the polarity. Is embedded in the audio signal as presence or absence.

FIG. 3 shows a schematic configuration of an apparatus for reproducing additional data from an audio signal according to an embodiment of the present invention.
This additional data reproducing apparatus corresponds to the additional data embedding apparatus shown in FIG. 1, and an input terminal 21 receives an audio signal in which the additional data is embedded as to whether or not the polarity is inverted.

The audio signal in which the additional data is embedded from the input terminal 21 is input to the polarity determination unit 22 and the boundary detection unit 23. The boundary detecting section 23 detects the boundary of the input audio signal in the same manner as the boundary detecting section 14 in FIG. 1. When the audio signal is a voice signal, for example, it detects a syllable boundary as described later. If the audio signal is a tone signal, a silent section in the music, a space between the music, and the like are detected. The polarity determination unit 22 determines the polarity of the input audio signal each time the boundary is detected by the boundary detection unit 23, and outputs the polarity determination result to the output terminal 25 as reproduced additional data. The audio signal from the input terminal 21 is also guided to the output terminal 24.

Next, the case where the audio signal is an audio signal will be described more specifically. As described above, in the present invention, an audio signal is divided into segments sandwiched between certain boundaries, and one bit of additional data is assigned to each segment, and each segment is assigned according to the assigned bits. The additional data is embedded by inverting or not inverting the polarity of. Here, when determining the length of the segment of the audio signal, there are two conditions that 1) the amplitude of the waveform is sufficiently small at both ends of the segment and 2) the length of the segment is sufficiently long. It is desirable to satisfy

The condition 1) is desired to reduce the discontinuity of the waveform when the polarity is inverted. In the case of audio signals, it is ideal to perform polarity reversal in silent periods or in closed periods of unvoiced closed consonants. It is also possible to reverse.

With respect to the condition 2), there is a trade-off. That is, while it is desired that the length of the segment is long enough to determine the polarity, it is better to be as short as possible in order to maximize the bit rate of the additional data to be embedded. For example, when a sentence is used as a segment, the sentence contains vowels sufficient for estimating the polarity, so that it is possible to embed 1 bit of additional data reliably, but the bit rate is considerably reduced. Resulting in. On the other hand, syllables are the most efficient as a unit for performing polarity reversal in terms of both ease of polarity determination and bit rate of additional data.

A normal syllable consists of a nucleus, a start and a tail, where the power of the waveform is usually small at the start and the end. This satisfies the above condition 1). Further, the syllable has an optimal length in the sense that the polarity is estimated in the stationary part of the vowel. For example, the duration of a syllable in US English is 200-250 ms, and the duration of a vowel is about 100 ms.

Therefore, in this embodiment, when the target audio signal is an audio signal, the boundary detection unit 1 shown in FIG.
4 and 3 detects a syllable boundary. Known examples of algorithms for automatically detecting syllable boundaries include S. Wu, ML Shire, and S. Greenberg an.
d N. Morgan, “Integrating Syllable Boundary Inform
ation into Speech Recognition, ”Proc. ICASSP (IEE
A paper entitled E), 1997 describes an automatic syllable start detection algorithm based on subband energy trajectories. On the other hand, in the present embodiment, by using a method in which the full-band energy trajectory and the voicing detection by the modified autocorrelation function are used together, a voiced section (particularly, a vowel) always exists between two consecutive syllable boundaries. Made it exist.

FIG. 4 shows the structure of a syllable boundary used as the boundary detector 14 in FIG. 1 and the boundary detector 23 in FIG. The audio signal from the input terminal 11 is first input to the frame division unit 31 and is divided into frames having a predetermined frame length, and then the windowing processing unit 32 performs windowing processing.

When the input audio signal is a digital audio signal of 16 kHz sampling, the frame division is, for example, frame length: 256 points = 16 ms, frame period: 1
28 points = 8 ms, that is, continuous frames overlap by 50% (128 points) in time. Here, the frame period is a shift width between adjacent frames. In the windowing processing unit 32, windowing is performed with a windowing length equal to the frame length, for example, using a hamming window. The audio signal of each frame after windowing is
It is input to a linear prediction (LPC: Linear Predictive Coding) analyzer 33. The analysis order in the LPC analysis unit 33 is, for example, 16th order.

The LPC obtained by the LPC analyzer 33
An inverse filter 34 having coefficients as filter coefficients is configured,
With this inverse filter 34, a residual signal, which is an LPC prediction error, is obtained for the audio signal of each frame from the frame division unit 31. The residual signal is calculated by the autocorrelation operation unit 35
And its autocorrelation function (this is called a modified autocorrelation function) is obtained. FIGS. 5A and 5B show this state. FIG. 5A shows a vowel waveform of a typical sound signal, and FIG.
Denotes a residual signal obtained by linear prediction analysis, and (c) denotes a modified autocorrelation function.

The modified autocorrelation function is input to the voiced / unvoiced discriminating section 36. The voiced / non-voiced discriminating unit 36 receives the low-frequency energy obtained by the low-frequency energy calculating unit 37. Low frequency energy calculator 3
In step 7, in the audio signal of each frame from the frame division unit 31, the energy of a low-frequency component of, for example, 1 kHz or less (this is referred to as intra-frame low-frequency energy) is calculated.

The voiced / non-voiced discriminating section 36 discriminates whether each frame is a voiced section or a non-voiced section from the modified autocorrelation function and the low-frequency energy in the frame. Specifically, when the modified autocorrelation function is normalized, for example, there is a normalized peak having a peak value of 0.2 or more where the lag is 0.2 ms or more and a low-frequency energy within the frame exists. When the energy is larger than the entire average low-frequency energy, the voiced section is determined, and the other sections are determined as non-voiced sections.

The discrimination result of the voiced / non-voiced discriminating section 36 is calculated by an intra-frame energy calculating section 38.
1 together with the intra-frame energy obtained for the audio signal of each frame, and is input to the syllable boundary determination unit 39. The syllable boundary determination unit 39 basically determines, as a syllable boundary, a point at which the energy in the frame becomes minimum for each continuous non-voiced section sandwiched between voiced sections.

More specifically, a frame whose energy in a frame is the minimum (Emin) is found in a certain unvoiced section, and the time difference Δt from the immediately preceding syllable boundary where the energy Emin in the frame is equal to or less than a predetermined threshold value Eth is found. 100m
Points equal to or longer than s, preferably equal to or longer than 200 ms are determined as syllable boundaries. FIG. 6 shows a voiced section (indicated by a rectangle) obtained by the voiced / non-voiced discriminating unit 36 for a vowel sound waveform of the audio signal and a syllable boundary (indicated by a circle) obtained by the syllable boundary determining unit 39. An example is shown.

Next, a specific example of the polarity determining section 22 in FIG. 2 will be described. The source of the power of the audio signal is
It is a respiratory system in most languages, and air is exhaled from the lungs. Air from the lungs passes through the trachea and at the pharynx between the vocal cords. Since voice is usually uttered when exhaling, the glottal airflow during vocalization is unidirectional, so that the polarity of the voice signal waveform is always constant.

One method of measuring the polarity of the audio signal waveform is to linearly predict the glottal airflow (LPC: Linear PPC) of the audio signal.
This is a method of estimating using an inverse filter method such as redictive coding (analysis). Speech sounds are generally composed of voiced sounds and unvoiced sounds, and voiced sounds and unvoiced sounds are modeled as responses from a vocal tract filter system when a plurality of sound sources are input.

In the linear prediction analysis, the voiced sound source is assumed to be a quasi-periodic pulse, and the unvoiced sound source is assumed to be random noise. Since the residual signal in the linear prediction analysis for the voiced sound becomes impulse-like, the linear prediction as shown in FIG. 5B is performed rather than examining the polarity of the original voice signal waveform shown in FIG. It can be seen that by examining the direction of the pulse of the residual signal by analysis, the directionality of the sound source, that is, the polarity of the original audio signal waveform can be examined. In the example of this figure, the direction of the pulse of the residual signal is upward, and the polarity of the audio signal waveform is positive. The polarity of the audio signal waveform is constant even if the type of vowel changes, as long as various conditions, for example, utterance when exhaling, or the condition that the polarity of the device at the time of recording does not change with time, do not change. The polarity determination unit 22 will be described based on the above points.

FIG. 7 is a block diagram showing the configuration of the polarity judging unit 22 and the boundary detecting unit 23 in FIG. 3 when the input audio signal is an audio signal. The input terminal 21 receives an audio signal (audio signal) in which additional data is embedded by the additional data embedding device shown in FIG.

In FIG. 7, a frame dividing section 31, a windowing section 32, an LPC analyzing section 33, an inverse filter 34, an autocorrelation calculating section 35, a voiced / unvoiced discriminating section 36, a low-frequency energy calculating section 37, an intra-frame Energy calculator 38
And the syllable boundary determination unit 39 is the same as in FIG.
FIG. 7 shows a configuration in which a polarity calculation unit 41 is further added. As shown in correspondence with FIG. 3, the frame division unit 31, the windowing processing unit 32, the LPC analysis unit 33, the inverse filter 34,
Autocorrelation calculation unit 35, voiced / unvoiced discrimination unit 36, low-frequency energy calculation unit 37, intra-frame energy calculation unit 3
8 and the syllable boundary determination unit 39 are shared by the polarity determination unit 22 and the boundary detection unit 23 in FIG. 3, and the polarity calculation unit 41 is further combined with the polarity determination unit 22 to configure the polarity determination unit 22.

In the polarity calculator 41, the inverse filter 3
The syllabic boundary and the syllable boundary are obtained from the residual signal of the voiced section by inputting the residual signal from the voice signal 4 and the determination result of the voiced / non-voiced determining unit 36 and the syllable boundary determined by the syllable boundary determining unit 39. The polarity of the audio signal waveform is obtained by calculation for each interposed syllable. Specifically, the polarity calculator 41
In, the polarity of the residual signal pulse is determined for each frame of a voiced section in each syllable, whereby the polarity of the audio signal waveform of each syllable is determined. That is, a majority decision is made between a frame estimated to have inverted polarity and a frame estimated to be non-inverted in a voiced section in each syllable.

The input terminal 21 receives a voice signal in which the additional data is embedded by inverting the polarity in accordance with the additional data in syllable units, so that the polarity determination result obtained by the polarity calculator 41 is added to the input terminal 21. This represents data, which is output from the output terminal 25 as reproduced additional data.

It is important that the quality of the audio signal does not deteriorate in the process of data hiding or digital watermarking on the audio signal. In response to this requirement, the present invention utilizes the property that human hearing is insensitive to inversion of the polarity of an audio signal. In order to confirm this point, the inventors tried a discrimination experiment for a plurality of subjects between the original speech signal and the speech signal after embedding the additional data depending on the presence / absence of polarity reversal in syllable units. In the following experiments, syllable boundaries were detected manually.

As the voice signal, 20 sentences of the TIMIT database were used. The subjects were 20 native speakers of Japanese, and all were hearing listeners. In the experiments, the ABX discrimination method (X is either A or B) was used. A is the original audio signal and B is the polarity inversion signal, or vice versa. The order of stimulation was randomly rearranged for each subject. The experiment was performed in a simple soundproof room using a PC (personal computer), and the subjects were headphones (Sennheis
er, HD600), and answered according to the instructions on the PC screen. Each subject could hear the same stimulus up to 10 times. The correct answer rate when X is the original audio signal and when X is the polarity inversion signal is 50
Ideally, the percentage of persons responding to the original audio signal with respect to the same audio signal is equal to the percentage of persons responding to the polarity-determined audio signal.

As a result of this discrimination experiment, the average of the correct answer rate when X is the original voice signal and when X is the polarity-inverted signal is 51.3%. Audio signal is indistinguishable from the original audio signal,
It was confirmed that the embedding of the additional data hardly affected the signal quality.

On the other hand, when a similar discrimination experiment was performed for the case where syllable boundaries were automatically detected with the configuration shown in FIG. 4, the case where X was the original voice signal and the case where X was the polarity inversion The average of the correct answer rate in the case of the signal was 53.0%. From this result, it was found that the human auditory sense could not discriminate the original speech signal from the speech signal after the polarity reversal even for the syllable boundary automatically detected.

Next, an attempt was made to automatically extract and reproduce the additional data embedded in the audio signal as described above with the configuration shown in FIG. This additional data automatic extraction algorithm is based on the above-described LPC analysis. First, syllable boundaries are detected, and a majority decision for positive and negative polarities between syllable boundaries (syllable units) is determined for each voice frame. The polarity was the polarity of the syllable.

As additional data, several binary data strings were used and embedded in 20 sentences of the TIMIT database. At that time, the average bit length was 7.6 bits per sentence. Next, a binary data string as embedded additional data was extracted and reproduced, and compared with the original additional data before embedding, it was correctly reproduced with a probability of 96.78% (5 out of 152 bits). error). The breakdown of the errors is two errors in automatic syllable boundary detection and three errors due to automatic polarity determination. By increasing the accuracy of these detections and determinations, errors can be further reduced.

In the above embodiment, the example has been described in which the syllable boundaries of the audio signal are detected and the polarity of the audio signal is inverted between the syllable boundaries, that is, in units of syllables according to the additional data.
Needless to say, the unit of the polarity inversion based on the additional data is not limited to a syllable unit. For example, when the bit rate of the additional data may be small, a unit such as a word, a phrase, or a sentence may be used.

A zero cross point of the audio signal may be used as a boundary of a unit for inverting the polarity of the audio signal. When the polarity of the audio signal is inverted at the zero crossing point, the signal quality is slightly degraded.However, compared to the case where the polarity is inverted where the amplitude is large, there is no large noise, and the quality is not a problem. Can be used.

Further, as described above, the target audio signal may be a musical tone signal. In this case, the polarity inversion according to the additional data is performed with a boundary between a silent section of the music, a space between the music, and the like. What should I do?

The processing of embedding the additional data in the audio signal and reproducing the additional data from the audio signal based on the present invention can be realized by hardware such as a DSP (digital signal processor), but is similar to a personal computer. It can also be executed by software using a simple computer. Therefore, according to the present invention, it is possible to provide a program as described below, or a computer-readable recording medium storing the program.

(1) A program for causing a computer to execute a process of embedding binary additional data in an audio signal, or a computer-readable recording medium storing the program, wherein the audio signal is stored in the audio signal. A program for causing a computer to execute a process of inverting and outputting a polarity according to the additional data for each predetermined unit of a signal, or a computer-readable recording medium storing the program.

(2) A program for causing a computer to execute a process of embedding binary additional data in an audio signal, or a computer-readable recording medium storing the program, wherein the syllable boundary of the audio signal is A program for causing a computer to execute a process of detecting the audio signal and a process of inverting the polarity of the audio signal in accordance with the additional data for each syllable unit on the basis of the detected syllable boundary and outputting the audio signal, or a computer storing the program. A readable recording medium.

(3) In the above (2), the processing for detecting the syllable boundary includes processing for dividing an input audio signal into frames, processing for performing linear prediction analysis on the audio signal of each divided frame, and A process of obtaining a modified autocorrelation function by taking an autocorrelation of the residual signal in the linear prediction analysis; and a process of discriminating a voiced section and an unvoiced section from the modified autocorrelation function and low-frequency energy of each frame of the audio signal. And determining a syllable boundary from the energy of the frame determined to be a non-voiced section of the audio signal.

(4) A program for causing a computer to execute a process of reproducing the additional data from an audio signal in which the additional data is embedded by inverting the polarity in accordance with binary additional data for each predetermined unit. Alternatively, a computer-readable recording medium storing the program, the program causing a computer to execute processing of reproducing the additional data by determining the polarity of the audio signal for each predetermined unit, or A computer-readable recording medium stored.

(5) A program for causing a computer to execute a process of reproducing the additional data from an audio signal in which the additional data is embedded by inverting the polarity in accordance with binary additional data for each predetermined unit. Alternatively, a computer-readable recording medium storing the program, wherein the process of detecting a syllable boundary of the audio signal, and determining the polarity of the audio signal for each syllable unit based on the detected syllable boundary, A program for causing a computer to execute a process of reproducing additional data, or a computer-readable recording medium storing the program.

(6) In the above (5), the processing for detecting a syllable boundary includes processing for dividing an input audio signal into frames, processing for performing linear prediction analysis on the audio signals of each of the divided frames, and A process of obtaining a modified autocorrelation function by taking an autocorrelation of the residual signal in the linear prediction analysis; and a process of discriminating a voiced section and an unvoiced section from the modified autocorrelation function and low-frequency energy of each frame of the audio signal. And determining a syllable boundary from the energy of the frame determined to be the non-voiced section of the audio signal. The processing of reproducing the additional data is performed for each syllable sandwiched by the determined syllable boundary. The polarity of the audio signal is determined for each syllable unit by taking the majority decision of the polarity of the residual signal of the frame determined to be the voiced section. Having that process.

Further, according to the present invention, an apparatus having both functions of embedding the additional data in the audio signal and reproducing the additional data from the audio signal,
And a program or a recording medium.

[0060]

As described above, according to the present invention, embedding of additional data in an audio signal and reproduction of the additional data from the audio signal are less likely to be caused by the processing of the audio signal. It is possible to do.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an apparatus for embedding additional data in an audio signal according to an embodiment of the present invention.

FIG. 2 is a diagram showing waveforms of an audio signal before and after polarity inversion.

FIG. 3 is an exemplary block diagram showing the configuration of an apparatus for reproducing additional data from an audio signal according to the embodiment;

FIG. 4 is a block diagram showing a configuration of a boundary detection unit according to the embodiment;

FIG. 5 is a diagram showing an example of a vowel sound waveform, a residual signal, and a modified autocorrelation function of an audio signal.

FIG. 6 is a diagram showing an example of a vowel sound waveform, a voiced section, and a syllable boundary of an audio signal.

FIG. 7 is a block diagram showing a configuration of a polarity determination unit in the embodiment together with a boundary detection unit;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Audio signal input terminal 12 ... Additional data input terminal 13 ... Polarity inversion part 14 ... Boundary detection part 15 ... Data buffer 16 ... Switch 17 ... Audio signal output terminal 21 ... Audio signal input terminal 22 ... Polarity determination part 23 ... Boundary detection Unit 24 audio signal output terminal 25 additional data output terminal 31 frame division unit 32 windowing processing unit 33 linear prediction analysis unit 34 inverse filter 35 autocorrelation function calculation unit 36 voiced / non-voiced determination unit 37 ... Low-frequency energy calculator 38. In-frame energy calculator 39. Syllable boundary determiner 41. Polarity calculator.

Claims (6)

[Claims] An apparatus for embedding binary additional data in an audio signal, wherein the audio signal is inverted for each predetermined unit of the audio signal in accordance with the additional data and output. A device for embedding additional data in signals. 2. An apparatus for embedding binary additional data in an audio signal, comprising: means for detecting a syllable boundary of the audio signal; and detecting the audio signal based on the detected syllable boundary for each syllable unit. Means for embedding additional data in an audio signal with respect to the audio signal. 3. The means for detecting a syllable boundary includes: means for dividing an input audio signal into frames; means for performing linear prediction analysis on the audio signal of each of the divided frames; Means for obtaining a modified autocorrelation function by taking an autocorrelation of a signal; means for determining a voiced section and an unvoiced section from the modified autocorrelation function and low-frequency energy of each frame of the audio signal; 3. The apparatus for embedding additional data in an audio signal according to claim 2, further comprising means for determining a syllable boundary from the energy of the frame determined to be a voiced section. 4. An apparatus for reproducing said additional data from an audio signal in which said additional data is embedded by inverting the polarity in accordance with binary additional data for each predetermined unit, comprising: An apparatus for reproducing additional data from an audio signal, wherein the additional data is reproduced by making a determination every time. 5. An apparatus for reproducing said additional data from an audio signal in which said additional data is embedded by inverting the polarity according to binary additional data for each predetermined unit, wherein a syllable boundary of said audio signal is detected. And a means for reproducing the additional data by determining the polarity of the audio signal for each syllable unit based on the detected syllable boundary. 6. The means for detecting a syllable boundary includes: means for dividing an input audio signal into frames; means for performing linear prediction analysis on an audio signal of each divided frame; and a residual in the linear prediction analysis. Means for obtaining a modified autocorrelation function by taking an autocorrelation of a signal; means for determining a voiced section and an unvoiced section from the modified autocorrelation function and low-frequency energy of each frame of the audio signal; Means for determining a syllable boundary from the energy of the frame determined to be a voiced section, wherein the means for reproducing the additional data is determined as the voiced section for each syllable sandwiched by the determined syllable boundary. The polarity of the audio signal is determined for each syllable unit by taking a majority decision of the polarity of the residual signal of the frame. Additional data reproducing apparatus from the mounting of the audio signal.