Audio watermarking scheme with dynamic adjustment in mute period

(1)

Audio watermarking scheme with dynamic adjustment in mute period

Tzong-Sun Wu

a

, Han-Yu Lin

b,⇑

, Wang-Chieh Hu

c

, Yih-Sen Chen

a

Department of Computer Science and Engineering, National Taiwan Ocean University, Keelung 202, Taiwan b

Department of Computer Science, National Chiao Tung University, Hsinchu 300, Taiwan c

Department of Informatics, Fo Guang University, Yilan 262, Taiwan

a r t i c l e

i n f o

Keywords: Audio ﬁle Information hiding Digital watermarking Mute period

a b s t r a c t

Watermark is some imperceptible information embedded into the work for later verification and thus it can be used to prove the originality or to protect the work from being illegally copied or modified. Kaab-neh and Youssef (2001) proposed a muteness-based watermarking method for audio file. Their method can successfully embed watermarking information into audio file such that the human ears cannot dis-tinguish the covered work from the original one. To improve the efficiency, we proposed a watermarking method for audio files in this paper. By adjusting the length of mute period dynamically, the proposed method can achieve the goal of watermark embedding with little variation of the original work. Experi-ment has shown that our method has the advantages of efficiency and fidelity as compared with theirs. Furthermore, it does not require the original audio file to derive the embedded watermark.

1. Introduction

With the continuous improvement of information technology (IT) and the increasing efﬁciency of computer facilities, more and more IT tools are available for human daily life. Take multimedia applications for examples. To draw a picture or to edit a song, we can use the computer software to assist us in some tedious works or precise tasks. Inevitably, IT also equips the adversary with copy or pirating tools. Furthermore, the popularity of Internet and increasing bandwidth accelerate such violations. As a result, it is very common to hear of the discussion about intellectual property protection against illegally coping.

Digital watermarking (Zeng, 1998), one kind of information hid-ing, is one of the alternatives used to protect digital rights of most multimedia files, such as image, audio, or video files. With this technique, some delicate information can be embedded into the file and the information can be extracted for certain usages, such as for ownership verification or access authorization. Some appli-cations of digital watermarking are listed below (Zeng, 1998):

1. Identification of ownership: The creator or owner could insert certain identification information relevant to himself/herself to media files so that it could be extracted by watermarking detector to demonstrate the ownership when questioned.

2. Fingerprinting: The digital watermark is like a ﬁngerprint. The identity of buyer and purchasing information can be packed into the embedding message. In case of a later dispute, this ﬁn-gerprint can be used to trace back the illegal distributor. 3. Conveying copyright information: The digital watermark could be

treated as the copyright information that is embedded into the media ﬁle. It conveys copyright information linking to some more detailed information about the copyrighted material such as the owner’s contact proﬁle, which helps with the determent of copyright infringement.

4. Copyright monitoring: With a copyright monitor, owners can search the Internet to look for the watermark of the copyright information and ﬁnd out unauthorized or inappropriate usages of his/her image without royalty payment.

5. Copy preventing: The digital watermark can also be used to pre-vent unauthorized copy or playback. Installed in the compliant machine, for example, the embedder can embed a watermark into a DVD indicating that the disc is allowed no copy or a single copy. Further, the built-in detector will not allow the machine to play the copy version of a DVD, which reduces the value of the illegal copies.

6. Embedding relevant information: The digital watermark can pro-vide some relevant information about the content. For example, it embeds some proﬁle information such as object shape, pro-duction date, etc. The information can be used to identify the content or some other usage.

7. Authentication: The digital watermark can be used as an authen-ticator. Once the watermarked ﬁle is modiﬁed, the extracted watermark would be also damaged. It might further give the 0957-4174/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.

doi:10.1016/j.eswa.2010.12.069

⇑ Corresponding author. Address: Department of Computer Science, National Chiao Tung University, No. 1001, Dasyue Rd., Hsinchu City 300, Taiwan, Republic of China. Tel.: +886 926 377200.

E-mail address:hanyu.cs94g@nctu.edu.tw(H.-Y. Lin).

Contents lists available atScienceDirect

Expert Systems with Applications

(2)

information of where or to what extent the file is modified. In some applications, we might also require that the watermark survive some specific processing such as compression. Partly because of the ease of presentation, quite a lot of research literatures focused on research of image watermarking techniques. There are relatively small amount of studies dedicated to the watermarking for audio and video. However, the protecting mea-sures for audio and video media are still in great need. Embedding messages in the least significant bits (LSB) might be one of the sim-ple ways to embed information. Since the LSB only has little effect on the presentation of the media, it is hard for human perception to tell the difference when the LSB is modified. Cvejic and Seppänen (2004) applied the LSB to design an audio steganography scheme for information hiding. Nevertheless, LSB is not so robust against active attacks. Uludag and Arslan (2001) proposed an audio water-marking scheme using DC level shifting to embed one bit informa-tion in the DC level of the sound frame. We shift the sound frame toward positive direction to embed 1 and negative direction 0. It only needs to compute the sampling mean of all audio frames of the file when decoding. If the mean is above/under some threshold value, 1/0 is embedded; otherwise, no information is embedded.

Megias, Herrera-Joancomarti, and Minguillon (2003) utilized the difference between the compression and decompression of MP3 to embed information (Megias et al., 2003). The watermark is in-serted into some pre-selected positions in the processed file. If 0 is embedded, the signal is reduced by n dB; if 1 is embedded, the signal is increased by n dB. By means of the mute period of the audio file,Kaabneh and Youssef (2001)proposed an audio water-marking scheme. The mute period is one characteristic of the sound, for example, it might be the interval between sentences or break between syllables. They embedded the watermarking information in the file by modulating the length of mute period. The information can be extracted by subtracting the mute period of the watermarked file from that of the original one.

To enhance the security, in this paper, we proposed an audio watermarking scheme based on the mute period which may contain some noises that are unperceivable by the human auditory system (HAS). Using mute periods as the positions for embedding the watermarking information can prevent an attack-er from easily detecting whattack-ere the watattack-ermark is. We slightly modify the manipulation process which results in a great improvement of performance as compared with the Kaabneh– Youssef scheme. For a better understanding of audio ﬁle’s water-marking technology, we ﬁrst give some requirements (Kaabneh & Youssef, 2001):

(i) The embedded watermarking information shall be unper-ceivable. It is usually a tradeoff between the ﬁdelity and the robustness. That is, more embedded watermarking infor-mation could result in more distortion and more robustness, and vice versa.

(ii) The watermarking system should be robust enough against general digital processing, such as loss compression, ﬁlter-ing, re-samplﬁlter-ing, digital/analog conversion, and noise inserting.

(iii) The system can meet the requirement of multiple water-marks. In other words, the original ﬁle can be embedded several different watermarking information such that each of the watermarks can be separately detected.

(iv) The attacker cannot derive the embedded watermarking information by means of analyzing many watermarked ﬁles in which the same information is embedded.

(v) It is infeasible for the attacker to remove the embedded watermarking information.

The human auditory system (HAS) has an audio perception fre-quency range from 20 Hz to 20 kHz (Kaabneh & Youssef, 2001; Xu, Zhu, & Feng, 2001). The watermarking information cannot be embedded outside this range, or it can be easily removed or ﬁltered out through the processing of audio signal. Therefore, it is a chal-lenging task to hide watermarking information in audio ﬁle; how-ever, it should be an important segment in future applications.

The rest of the paper is organized as follows. Section2is the description of some literature reviews. Section3 will introduce the proposed audio watermarking scheme. Some experiments and analyses according to the proposed method are given in Sec-tion4. Finally, we make a conclusion in Section5.

2. Literature review

In the section, we introduce the manipulation of digital water-marking and review the muteness-based method proposed by

Kaabneh and Youssef (2001). The digital watermarking is seen as a possible solution to the problem of copyright protection. In gen-eral, watermarking techniques can be classified into two broad cat-egories: spatial-domain-based and frequency-domain-based ones. The former watermarking techniques embed the watermark into the host image by directly modifying the pixel value of the host im-age without causing obvious change in appearance. By modulating the least-significant bit (LSB), the human visual organ is not sensi-tive to the difference because the change in intensity of certain pix-els is small. Several schemes were proposed by using statistic method, LSB (Chan & Cheng, 2004), and adjusting the value of blue color tone. The latter watermarking techniques embed the water-mark by modulating the magnitude of coefficients in a transform domain, such as discrete cosine transform (DCT) (Barni, Bartolini, Cappellini, & Piva, 1998), discrete Fourier transform (DFT), discrete wavelet transform (DWT) (Suhail, Obaidat, Ipson, & Sadoun, 2003; Tsai, Hu, & Chang, 2004; Víctor, Mariko, & Héctor, 2004), etc. Then, the transformed frequency coefficients are modified to embed the desired watermark. Finally, the inverse transform function is performed.

Kaabneh and Youssef (2001) proposed a watermarking scheme for embedding watermarks in audio file. They utilized the mute periods of audio files as the positions for information hiding. The characteristics of the mute period are (i) it is an inevitable segment of any audio file and cannot be omitted; (ii) it occurs randomly in an audio file; (iii) it represents a time span that will not be de-creased by compression. Usually, there must be some mute periods in audio files such as the break of sentences or phases in speech clips and the rest in music files. All these mute periods are consid-ered as the necessary parts of any song or speech and cannot be negligent or omitted. Note that, the mute period still may contain noises. However, such noises are imperceptible by the human auditory system (HAS). We briefly describe the Kaabneh–Youssef scheme as follows.

At the beginning, we ﬁrst search all mute periods in an audio ﬁle and record the length of the ith mute period as M(i). Then determine the variation amount d(i) of each M(i) according to the information of embedded watermark. Finally, compute the new mute period Mw(i) with the embedded watermark as Eq.(1)

MwðiÞ ¼ MðiÞ þ dðiÞ: ð1Þ

Without using the original audio file, their scheme only requires the length of each mute period of the original file to detect the embed-ded watermark. From the distribution copy of the audio file, we can extract all the lengths of all mute periods Mw(i)’s and calculate each

variation amount d(i) as Eq.(2).

(3)

It can be seen that, in the Kaabneh–Youssef scheme, the length of mute periods is increased with the amount of variation, which results in a size expending of the file. In addition, their scheme has to record all the lengths of mute periods of the original file for extracting the watermark. In the next section, we will propose an efficient audio watermarking scheme to eliminate these drawbacks.

3. Our proposed method

We use the dynamic adjustment in mute period (DAMP) to embed watermarking information into audio files. Since we modify the length of mute periods to encode messages and dynamically modulate each mute period, the length expansion of the audio file can be reduced to the minimum. Our approach needs three prede-fined parameters including the threshold amplitude of mute peri-ods (dbT), the threshold lasting time of the mute periods (LT)

with respect to the corresponding dbT, and the slice unit (Su) for

mute periods derived from the shortest mute period. With these three values, we can extract the watermark without the original audio ﬁle. The proposed scheme can be divided into two proce-dures: the watermark embedding and the extraction procedures. Details of them are given as follows.

3.1. Watermark embedding procedure

Fig. 1depicts the diagram of our watermark embedding proce-dure. First of all, we find out all mute periods of the audio file and record their lengths. Then, by seeking the minimum length of mute period and determining the number of slice n, we can compute the value of Su. Finally, we embed the message into the audio file to

complete the whole procedure. For facilitating the practical imple-mentation, we give the detailed pseudo code, DAMP-E, of the watermark embedding procedure in Appendix.

3.2. Watermark extraction procedure

The extraction procedure of embedded watermark is shown in

Fig. 2. With the threshold values of LTand dbT, we ﬁrst ﬁnd all mute

periods of the watermarked audio ﬁle. Then we utilize the critical value Suto extract the watermark from each mute period. The

de-tailed pseudo code, DAMP-R, of the watermark extraction proce-dure is given in Appendix.

4. Experiment designs and analyses

In the section, we will verify the feasibility of our proposed scheme and analyze the signiﬁcance of the three critical values

dbT, LTand Suin the watermark extraction procedure. We compare

the Kaabneh–Youssef scheme with ours in terms of the capacity of watermarking information, the length expansion and the informa-tion capacity required for examinainforma-tion. Based on the same infor-mation content and selection standard of mute periods, we design the following experiments and analyze the results.

In the first experiment, we select four different audio files, namely the WAVE files of chant, Erhu Fiddle, Chinese Zither (Guzheng) and Traditional Chinese music for experimenting. All these audio files are stereo with the sampling frequency of 44.1 kHz and the sampling size of 8 bits. Then, we use one para-graph of words as the watermarking information to embed it into the above audio files with the Kaabneh–Youssef scheme and ours, respectively.

In the second experiment, we choose two audio files: one is the SONGBIRD played by Kenny G. in his saxophone album ‘‘Greatest Hits’’ published in 1997, which is a stereo WAVE file with the sam-pling frequency of 44.1 kHz and the samsam-pling size of 16 bits; the other is the Declaration of Independence read by the President Kennedy, which is downloaded from the website of JFK Library and Museum (http://www.jfklibrary.org/audio/-jfk_declaration_ independence.rm). We then further convert this file into stereo WAVE format with the same sampling frequency and size. In this experiment, a picture is picked as the watermark. We illustrate the experiments in details as follows.

4.1. Analyses of the ﬁrst experiment

In the experiment, the text of ‘‘Fo Guang University’’ is used as the watermarking information. We ﬁrst transform the text into bit streams and then repeatedly insert them into all mute periods. The experiment results are demonstrated asTables 1–4. Some analyses are given below.

4.1.1. The capacity of watermarking information

As shown fromTables 1–4, it can be seen that under the same LT

value, the higher the value dbT, the more capacity of the

water-marking information it can embed. This is because increasing dbT

results more mute periods, so as to increase the capacity. Yet, if it dbTis too large, the quality of audio might be decreased. By

set-ting equivalent LT and dbT values, in this experiment, both the

Kaabneh–Youssef scheme and ours can obtain the same number of mute periods. Besides, we use the identical information content for embedding. Consequently, it comes out the equal capacity of watermarking information with two schemes.

4.1.2. The expansion of audio ﬁle

According toTables 1–4, one can see the difference of length expansion between the two schemes. With the proposed scheme, it is 0 ms for all audio ﬁles. With the Kaabneh–Youssef scheme, however, it is increased by 102 ms at most and 2 ms at least for dif-ferent audio ﬁles. Obviously, the proposed scheme can effectively reduce the length expansion compared with theirs. Further, con-sider the factor of the capacity of watermarking information. Since the proposed scheme dynamically modulate the mute period, the length expansion is slight or even none when the watermarking

Original audio file Find and record all mute period lengths Find the shortest mute period length Watermarked audio file Information hiding procedure Decide n and Sd Watermark

Fig. 1. Watermark embedding diagram.

Watermarked audio file

Find all mute periods

Extract the watermark with Su

(4)

information capacity is increased. On the contrary, seeing that the Kaabneh–Youssef scheme can not provide the function of dynamic adjustment, the length expansion will be increased proportional to the capacity of watermarking information.

4.1.3. The information capacity required for examination

The purpose of this analysis is to evaluate the information capacity required for examination, so as to calculate the necessary storage space. Assume that each vital data is stored with 2 bytes. FromTables 1–4, one can be seen that the proposed scheme only requires the space of 6 bytes on whatever conditions. Nevertheless, the Kaabneh–Youssef scheme needs more space for unﬁxed data capacity on different conditions. The reason of the above difference is that the proposed scheme only uses three critical values dbT, LT,

and Suto extract the watermark while their scheme has to keep

every mute period length of the original audio ﬁle for extracting.

To be precise, since the number of mute periods is equal to the watermarking information capacity in this experiment, (i.e., one Table 2

The 2nd test of Experiment 1.

Medium length (ms) Capacity (bit) Length expansion (ms) Overhead information (byte) SNR (dB) ASNR (dB)

DAMP KY DAMP KY DAMP KY DAMP KY

Chant 119000 5522 0 62 6 11048 64.72 45.37 90.74 73.97

Erhu Fiddle 240718 2566 0 29 6 5136 46.61 20.86 66.78 32.08

Chinese Zither 167416 9181 0 102 6 18366 61.69 40.63 80.77 53.47

Traditional Chinese music 252029 2437 0 28 6 4878 44.82 20.09 68.52 29.45

Remark: dbT= 8.96 dB, LT= 2.5 ms, Su= 0.13 ms and n = 20. Table 1

The 1st test of Experiment 1.

Chant 119000 3307 0 37 6 6618 64.85 45.57 91.37 74.7

Erhu Fiddle 240718 517 0 6 6 1038 53.32 21.05 74.44 35.19

Chinese Zither 167416 991 0 11 6 1986 63.52 40.74 79.63 55.07

Remark: dbT= 3.84 dB, LT= 2.5 ms, Su= 0.13 ms and n = 20.

Table 3

The 3rd test of Experiment 1.

Chant 119000 3307 0 37 6 6618 75.6 45.57 89.06 74.7

Erhu Fiddle 240718 517 0 6 6 1038 60.61 21.05 82.42 35.19

Chinese Zither 167416 991 0 11 6 1986 66.17 40.74 83.36 55.07

Table 4

The 4th test of Experiment 1.

Chant 119000 5522 0 62 6 11048 74.27 45.37 89.12 73.97

Erhu Fiddle 240718 2566 0 29 6 5136 56.86 20.86 79.01 32.08

Chinese Zither 167416 9181 0 102 6 18366 69.2 40.63 85.84 53.47

(5)

mute period embeds one bit,) we can express the information capacity required for examination with their scheme as Eq.(3).

Total bytes ¼ ðwatermarking information capacity 2Þ þ 4: ð3Þ

Note that the additional 4 bytes of the equation are used to ﬁnd dbT

and LTvalues.

4.1.4. The inﬂuence on audio quality and imperceptibility

If the watermarking information makes only a slight inﬂuence on audio quality, the imperceptibility is considered to be better; in other words, it is more difﬁcult to detect the position or even the existence of watermarking information (Megias et al., 2003). In the subsection, we use Signal-to-noise ratio (SNR) and Average SNR (ASNR) to evaluate the impact on audio quality with our pro-posed scheme. The SNR value is used to compare the difference be-tween the original signal and the processed one, i.e., the intensity of the noise it produced (Andres, 2002; Uludag & Arslan, 2001). Eq.(4)is the formula to compute SNR value, where N is the number of samples. SNR ¼ 10 log10 PN i¼1SðiÞ 2 PN i¼1ðSðiÞ S 0_ðiÞÞ2 ! ð4Þ

It is believed that the larger the value is, the less distortion the pro-cessed signal has. An audio file has less distortion also means that its quality is better. Generally speaking, an acceptable SNR value shall be greater than 35 dB (Andres, 2002). Namely, when the SNR value is greater than 35 dB, it is not easy to perceive the existence of watermarking information, i.e., it will be more difficult to tell the difference between signals. The ASNR value applies the pling number of SNR value to define a specific length as the sam-pling block, and then computes the SNR value of each block, respectively, to get its mean (Megias et al., 2003). Usually, under the 44.1 kHz sampling frequency, the length is defined to be 4 ms (Megias et al., 2003).

Tables 1–4showed the results of the ﬁrst experiment, where DMAP and KY stand for the proposed and the Kaabneh–Youssef schemes, respectively. It is easy to see that our scheme has a better performance. Besides, one can see that with the same threshold values of dbTand LT, the SNR value is getting higher when the value

of Suis becoming smaller. This is because that when the Sdvalue is

becoming smaller, the scale for the adjustment of mute period length is also getting smaller; therefore, if the Sdvalue is too large,

it might make the SNR value out of the acceptable range. 4.2. Analyses of the second experiment

In the experiment, we use a monochrome BMP image (Fig. 3) of 196 dots in its width and height as the watermarking information to embed it into audio files. The experimental results are demon-strated asTable 5. Since an image file has a larger size and more variable content compared with text-based data, we observe that the length expansion of the Kaabneh–Youssef scheme is signifi-cantly increased while that of the proposed one still keeps the

same length. Consequently, we can argue that with the Kaabneh– Youssef scheme, the length expansion is proportionally increased with the capacity of watermarking information. However, our scheme dynamically adjusts the length of next mute period according to the variation of current one and hence can minimize the inﬂuence on the total audio length when the capacity of water-marking information is increased. As for the SNR values with our proposed scheme, they are still within an acceptable range. In addition, the ASNR values are even over 100 dB.

Clearly, it can be seen that the proposed method also has a bet-ter performance in the requirement of imperceptibility.

5. Conclusions

In this paper, we have presented a watermarking method for audio file. By adjusting the length of mute period dynamically, the proposed method can achieve the goal of watermark embed-ding with little variation of the original work. Experiments showed that our method has a better performance in term of efficiency and fidelity. To conclude the above descriptions and analyses, the mer-its of our proposed scheme are summarized as follows.

(a) The original audio ﬁle is not required for extracting the watermarking information.

(b) Each length of mute periods of the original audio ﬁle is also not necessary for extracting the watermarking information. (c) It only needs three critical values dbT, LT, and Suto extract the

watermarking information, which beneﬁts to the ﬁle man-agement and the communication bandwidth.

(d) The variation of length expansion is negligible. Table 5

The results of Experiment 2.

Saxophone 480000 65839 0 732 6 131682 39.1 6.78 113.89 7.34

Declaration of Independence 639840 48697 0 732 6 97398 94.85 36.19 100.88 60.96

Remark: dbT= 3.84 dB, LT= 1 ms, Su= 0.05 ms and n = 20.

Appendix A

Notations.

AO original audio ﬁle

AO0 _{audio ﬁle with watermarking information} ST sampling time, e.g., 0.023 ms

aoi vibration amplitude of ith sampling value of AO

ao0

i vibration amplitude of ith sampling value of AO0

x number of mute periods

dbT threshold value of vibration amplitude of mute

periods, e.g., 2 dB

LT threshold value of lasting time of mute periods, e.g.,

10 ms;

AMP() strength of vibration amplitude SP set of all mute periods of AO, i.e.,

SP ¼ fspiji 2 ½1; x; AMPðspiÞ 6 db and jspij P LTg SP0 _{set of all mute periods of AO}0_{, i.e.,}

SP0

¼ fsp0

iji 2 ½1; x; AMP sp0i

(6)

References

Andres, G. (2002). Measuring and evaluating digital watermarks in audio ﬁles. Master Thesis, Dept. of Communication, Culture and Technology, Georgetown University.

Barni, M., Bartolini, F., Cappellini, V., & Piva, A. (1998). A DCT-domain system for robust image watermarking. Signal Processing, 66, 357–372.

Chan, C. K., & Cheng, L. M. (2004). Hiding data in images by simple LSB substitution. Pattern Recognition, 37, 469–474.

Cvejic, N., & Seppänen, T. (2004). Increasing robustness of LSB audio steganography using a novel embedding method. In Proceedings of the international conference on information technology: coding and computing (ITCC ’04). Vol. 2 (pp. 533–537). Kaabneh, K. A., & Youssef, A. (2001). Muteness-Based audio watermarking technique. In Proceedings of the 21st international conference on distributed computing systems workshops (ICDCSW ’01) (pp. 379–783).

Megias, D., Herrera-Joancomarti, J., & Minguillon, J. (2003). A robust audio watermarking scheme based on MPEG 1 layer 3 compression. Communications and Multimedia Security, (LNCS), 2828, 226–238.

Suhail, M. A., Obaidat, M. S., Ipson, S. S., & Sadoun, B. (2003). A comparative study of digital watermarking in JPEG and JPEG 2000 environments. Information Sciences, 151, 93–105.

Tsai, P., Hu, Y. C., & Chang, C. C. (2004). A color image watermarking scheme based on color quantization. Signal Processing, 84, 95–106.

Uludag, U., & Arslan, L. M. (2001). Audio watermarking using DC level shifting, Electrical & Electronics Engineering Department, Bogazici University, Turkey, Advanced Topics in Speech Processing Project Report (pp. 1–6).

Víctor, V. H. G., Mariko, N. M., & Héctor, M. P. M. (2004). Analysis of a wavelet-based watermarking algorithm. In Proceedings of the 14th international conference on electronics, communications and computers (CONIELECOMP’04) (pp. 283–287). Xu, C. S., Zhu, Y., & Feng, D. D. (2001). Digital audio watermarking based-on

multiple-bit hopping and Human Auditory System. In Proceedings of the 9th ACM international conference on multimedia (pp. 568–571).

Zeng, W. (1998). Digital watermarking and data hiding: technologies and application. In Proceedings of the international conference on information system, analysis and synthesis. Vol. 3 (pp. 223–229).

SPmin the minimum mute period of SP, i.e., SPmin= min (SP)

n the number of slice, e.g., 10

Su information examination value satisfying that

Su= SPmin/n

HD watermarking information, e.g.,

HD ¼ 10011001110110 ¼ fb0;b1;b2;b3; . . . ;bng

¼ fbiji 2 ½0; n and bi¼ ½0; 1g; HD0 _{extracted watermarking information}

N bit-length of HD

SP0 _{set of all mute periods with the watermarking} information

DA cumulative variations of mute period lengths

Algorithm DAMP-E (AO, dbT, LT, n, HD)

Initialization: x 0, DA 0, L 0, i 0, SP 0 /⁄ L: Length of mute period, i: for index use ⁄/ Do Until EOF (AO)

(a) i i + 1 (b) if aoi6dbTthen (b-1) L L + ST (c) else (c-1) if L P LTthen (c-1-1) x x + 1 (c-1-2) spx L (c-1-3) Add spxto SP (c-2) L 0 END DO SPmin min (SP) Su SPmin/n For i 1 to x (a) if DA < 0 then (a-1)

a

spi LT (a-2) if

a

+ DA P 0 then (a-2-1) spi spi+ DA (a-2-2) DA = 0 (a-3) else (a-3-1) spi spi+

a

(a-3-2) DA

a

+ DA (b) else (b-1) spi spi+ DA (b-2) DA = 0 (c) A bspi/Suc

(d) A A + ((A + HD(i 1)mod N) mod 2)

(e) sp0 i A Su (f) if sp0 i<LTthen (f-1) sp0 i sp0iþ 2Su (g) DA DA þ spi sp0i (h) spi sp0i Next i

Algorithm DAMP-R (AO0_{, db}

T, LT, Su)

Initialization: x 0, L 0, i 0, SP0₀ /⁄ L: Length of mute period, i: for index use ⁄/ Do Until EOF (AO0₎

(a) i i + 1 (b) if ao0 i6dbT then (b-1) L L + ST (c) else (c-1) if L P LTthen (c-1-1) x x + 1 (c-1-2) sp0 x L (c-1-3) Add sp0 xto SP0 (c-2) L 0 END DO For i 1 to x (a) C spi/Su (b) D C mod 2 (c) HD0i D Next i