Bao-Yuan Chen，Yi-Qiang Zhu，Lei-Lei Tian，Ying-Ying Li，Ya-Qiong Lan

(The Higher Educational Key Laboratory for Measuring＆Control Technology and Instrumentations of Heilongjiang Province，Harbin University of Science and Technology，Harbin 150080，China)

1 Introduction

In recent years，digital multimedia technology and the rapid development of Internet technology make the image video and audio transmit conveniently.However，Internet recklessly copy and the music product distribution make piracy works of authors and publishers suffer great damage［1］.As a result，the basic characteristics of digital watermark technology are robustness，invisible，undetectable and selfrestoration.Digital watermark can be divided into many types.According to the characteristics of the digital watermark technology，it can be divided into robust digital watermark and fragile watermark［2］.And it also can be divided into image watermark，audio watermark，video watermark，text watermark according to the load of media.Digital watermark is proposed to protect the copyright，however，with the development of digital watermarking technology［3］，this technology is used widely，such as radio monitoring，owners identify，ownership validation，operating tracking，content authentication，copy control and equipment control.Although the application of watermark technology is gaining more and more，but there are still many of the key technology to be difficult to make a breakthrough in the short term.Therefore，an algorithm of digital watermark technology based on wavelet transform is presented in this paper，and the use of this algorithm in audio watermark is studied.Digital watermark technology is a process that a specific meaning and easy extraction information is embedded to the original audio signal.According to the different aims for application，the copyright identifier，serial number of works，text(like an artist and name of songs)，even a short audio or a small image can be embedded.Watermark is combined with the original audio data tightly.Usually，it can not be heard and can resist some general audio signal processing and some malicious attacks of pirates.In this context，digital watermark technology arises at the historic moment in order to protect the copyright［4］.Digital watermarking technique is a new direction and new technology in the information security technology field.

An audio watermark technology based on wavelet transform is proposed，which embeds the specific meaningful information and is easy to extract.In the embedding process，it does not affect the original audio quality.

2 Audio Watermarking Technology Algorithm

Generally，there are two ways to implement audio watermarking technology algorithm:time-domain watermarking algorithm and transform domain watermarking algorithm.

2.1 Time-domain Watermarking Algorithm

2.1.1 Least significant bit method

The least significant bit watermarking algorithm is the way that embeds the secret information into carrier data，and it is a simplest method，and belongs to the time domain hidden algorithm.The basic thought of this algorithm is that the encrypted signal and carrier signal is deemed bittorrent sequence，when the encrypted information is hidden，the no-important bits of the whole speech signal values are replaced，so that the secret information is hidden into the speech signal［5］.

2.1.2 Echo hiding method

The principle of echo hiding method is:introduce the echo signal into the discrete signal，and code the watermark information by modifying the delay between echoes.When the watermark is extracted，autocorrelation function of each signal section signal cepstrum is calculated which will appear peak during the delay［6］.

2.1.3 Method by changing the signal amplitude［7］

Lie used frequency domain masking model，by enlarging or shrinking the three adjacent sample amplitude values，keeping their relative relationship of energy in order to embed watermark［8］，without original signal during extraction.It solved the synchronization problem by adding synchronization code words［9］.It can resist MP3 compression，lowpass filtering，amplitude normalization，shear and D/A，A/D conversion of the same sampling rate，but can not resist re-sampling，sample precision conversion，mono，multi-channel conversion and time stretching attacks and so on Ref.［10］.

2.2 Transform Domain Watermarking Algorithm

2.2.1 Fourier transform domain algorithms

Phase encodes use the characteristics of human auditory system which is not sensitive to the absolute phase of sound but sensitive to the relative phase of sound to embed the digital watermark.In phase encoding，firstly，the carrier signal is divided into a number of short sequences，and then the DFT transform is applied to them，absolute phase of the entire section signal are modified.At the same time，the relative phase remains unchanged，and then does the IDFT;finally，the signal which embeds the watermark is recovered［11］.Before the watermarking extraction，it must use synchronous technology and find the signal section，if the sequence＇s length is known，the receiver can calculate DFT and detect phase.This algorithm has robustness for re-sample of the carrier signal，but it is sensitive to many audio compression algorithms，because of encoding only in the first signal section，there is a low capacity for watermark.

2.2.2 Discrete cosine transform algorithm

Based on spread spectrum communication thoughts，the watermark information scatter is realized in the frequency domain.Spread spectrum watermarking technology can resist lossy compression while other data processing techniques have the disadvantages of signal distortion.However，in the process of watermark embedding produces additive noise.To overcome the side effect，audio masking technology must be used in the same time which can reduce the impact of watermark the lowest level.In addition，spread spectrum of watermarking extraction algorithm is more complex，and the algorithm requires higher synchronization，and has a poor robustness against the changes of the carrier audio signal.

2.2.3 Wavelet transform domain algorithms

Wavelet transform is a kind of time-frequency analysis tool，which can decompose signal to the timedomain and scale-domain，and different scales have different frequency ranges，for the audio signal which is a time-variant signal，so wavelet transform is a very appropriate tool［12］.

Generally speaking，digital audio watermarking should have the following basic features:imperceptibility，robustness，security，certainty and the restoration.A complete watermarking system should include three parts:the formation of watermark，embedding and extracting process，detection［13－14］.

Wavelet transform algorithm is employed to realize the research of the digital audio watermark.

3 Principle of Wavelet Transform

In recent years，attentions have been paid to the method of multi-resolution analysis of wavelet transform.One of its applications is data compression，and like other transformation，wavelet transform can be used in the original data，then coded the data that already transformed，and got effective compression，because the wavelet transform decomposes the signal to the time domain and scale domain，and different scales corresponding to different frequency range，therefore，towards the audio signal that always change in frequency，wavelet transform is a kind of very appropriate tools.An audio watermark embedding method based on wavelet transform is shown in Figs.1 and 2.

Fig.2 Watermark extracting

Wavelet transform decomposes the signal into a group orthogonal basis through translations and pressure expansion［15］，ifφ(t)∈L2(k)andφ(t)Fourier transformation as formula one［14］:

ψ(ω)meet the constraints:

So letφ(t)wavelet or mother wavelet［16］，translate and stretch mother wavelet and get formula three，which is a wavelet sequence，eachψa，b(t)is called a wavelet function;the“a”reflects a specific function scale expansion situation;variable“b”indicates that it goes along the x axis translation position［10］.

For arbitrary function f(t)∈L2(R)，continuous wavelet transform is defined as:

The inverter change is defined as:

4 Embedding and Extracting Process of Audio Watermarking System

4.1 Embedding Process

1)Separate the audio signal into frames，sections，and divide the original audio signal A into some frames that contain identical sample point.And each frame contains the same sample point section.

2)Implement Haar discrete wavelet transform to every frame of the audio signalrespectively，and get fine componentsand approximate components:

Do not process the approximate component，but embed watermark into the fine component.

Among them:S is the number for each section fine component.

4)When the watermark signal for"1":if＜，there is no need to change，else reduce the，until≤;if≥，it says that the energy has a downward trend，according to the lag masking properties of HAS，when reduce，the power signal in front ofwill mask the effect to reduce.

5)When the watermark signal for"0":if≥，don’t change it，else reduce，until≥;if，it says that the energy has an increase trend，according to the masking properties of HAS，when reduce，the power signal in front ofwill mask the effect to reduce.

6)Reconstruct audio signal:discrete wavelet inverter is applied to，and the audio signalare reconstructed:

So，finally get the audio signal A'with embedded watermark is reconstructed.

4.2 Identify the Headings

1)Divide A'into frames and sections according to the way when embedding.

2)For every frame of the audio signalbefore two sections，implement Haar discrete wavelet transform respectively，and get fine componentand approximate component:

3)Calculate and compare the energyof the first two fine components of every frame，then extracted watermark sequence，the method is as follows:ifwatermark extraction for"1";ifwatermark extraction for"0".

Finally，convert W'(i)to detect the watermark position in the binary watermark image，and ascend dimension to recover the watermark image.

5 MATLAB Simulation Experiment

The algorithm adopts the speech signal of 44.1 kHz，16 bit，18 s and the watermark is a 32×32 binary image.The audio signal is decomposed with Haar wavelet base and 3 level wavelet decomposition，then embed the watermark into the audio signal，the PSNR=84 dB.The original signal is shown in Fig.3，while the original image is shows in Fig.4，and the watermark signal is shown in Fig.5.

Fig.3 Original waveform

In order to detect the robustness of the algorithm，the following treatment should be done to the audio signal with MATLAB and other software，such as resampling，low-pass filtering and adding noise，shearing，compressing，and the specific practices processing are as follows:

Fig.4 Original image

Fig.5 Watermark waveform

1)Re-sampling:Re-sampled the watermark signal with the 48 kHz and got the waveform as shown in Fig.6，then after inverse wavelet transform the watermark is extracted as shown in Fig.7.

Fig.6 Watermark signal after re-sampled

Fig.7 Watermark extraction from re-sample signal

2)Low pass filtering:Filter the watermark signal with Butterworth low-pass filter of N=20，wp=0.1，ws=0.9，rp=0.6，rs=0.9.The filtered waveform is shown in Fig.8，and the watermark extracted from signal using the low pass filter is shown in Fig.9.

Fig.8 Watermark signal by low pass filter

Fig.9 Watermark from signal that by low pass filter

3)Adding noise:Add Gaussian white noise to the watermark signal with the Y=AWGN(X，SNR)function，watermark signal that added Gaussian noise is shown in Fig.10，and watermark extraction from signal that added Gaussian noise is shown in Fig.11.

Fig.10 Watermark signal that added Gaussian noise

Fig.11 Watermark from signal that added Gaussian noise

4)Shearing:Use a kind of audio editor software to seperate the watermark audio signal into some sections or parts before extracting the watermark.After that the shear wave is achieved as shown in Fig.12 and watermark as shown in Fig.13.

Fig.12 Shear watermark audio signal

Fig.13 Watermark after shear

5)Compressing:Also，choose a kind audio editor software to compress the whole watermark audio signal，compressed 1 dB，and the compression ratio is 89.13%，then got the compression signal wave as shown in Fig.14 and watermark as shown in Fig.15.

Fig.14 Compression watermark audio signal

Fig.15 Watermark after compression

In the process of MATLAB simulation，the normalized coefficient NC is used to assess the correlation of watermarks.For the audio signal with embedded watermark:Table 1 shows the NC values，such as，untreated original image and extracted watermark，after re-sampled of the original image and extracted watermark，after added white noise of the original image and extracted watermark，after low pass filtering，heared and compression of the original image and extracted watermark.

Table 1 NC value

In the MATLAB simulation process，use the normalization function:NC=nc(ImageA，ImageB)，in this function，ImageA is the original watermark image;ImageB is the extracted watermark;both ImageA and ImageB are binary image，and with this function we can get NC value.The value is used to test the relevancy of the watermarks which extract the original image.Also，there is a function mentioned in the first paragraph:x=psnr(a，b)，the function in order to calculate the Peak Signal to Noise Ratio.In this function，“a”is the original audio signal;“b”is the embedded audio signal，and by calling the functions we can easily get the similarity between the two signals.

Through the results in Figs.7，9，11，13，15 and Table 1，it is found that the algorithm has strong ability against the attack.

6 Conclusions

In this paper，the human auditory characteristics and discrete wavelet analysis algorithm are used to realize the image watermark embedded in audio signal.There is certain reliability in the algorithm，and a specific characteristic.The watermark extraction belongs to the blind watermark extraction，and don＇t need original carrier audio signals.The simulation results show that there is certain transparency for the algorithm，a good robustness for the resampling and low pass filtering attack.Therefore，it is able to embed and extract the specific meaning information easily when thetechnology of audio watermark based on wavelet transforms is used in the audio information security field，and the copyright protection will be realized.

［1］Mao Chenxi，Zhou Wensong，Ou Jinping.Damage assessment based on wavelet energy transfer function spectrum.Journal of Harbin Institute of Technology，2011，43(8):130－134.

［2］Zhang Yifan，Jiang Tianfa.Research on time spread echo hiding for audio watermarking.Computer Engineering and Applications，2008，44(31):119－120.

［3］Gui Guofu，Jiang Lingge.HVS-based color image watermarking scheme.Journal of Harbin Institute of Technlogy，2008，40(11):1805－1807.

［4］Ma Shanshan，Qian Jiansheng.A survey of digital audio watermarking techniques.Communications Technology，2008，41(11):135－138.

［5］Niu Xinxin，Yang Yixian.Based on wavelet transform digital watermarking hidden and the detection algorithm.Chinese Journal Computer，2000，23(1):21－27.

［6］Liu Bingquan，Xu Shuai，Li Xiangqian.Boundary detection of special English audio sentence based on dualthreshold.Journal of Harbin Institute of Technology，2010，42(2):259－263.

［7］Chen Licong，Chen Liping.Audio watermarking algorithm based on time domain synchronization and blockembedded.Technology of Modern Electronic，2008，11(21):86－89.

［8］Bao Yongqiang，Deng Aidong.The speech information hiding model of transparency in fractional fourier domain based on the wind interval of hopping.Signal Processing，2010，10(26):1544－1548.

［9］Liu Haiyan，Zheng Xuefeng，Cheng Renfeng，et al.Analysis of audio watermarking energy and reliability based on psychoacoustic model.Computer Engineering，2008，34(4):34－36.

［10］Cvejic N.Audio Watermarking using m-sequences and temporal masking.Proceedings of the IEEE Workshop on the Applications of Signal Processing to Audio and Acoustics.Piscataway:IEEE，2001.227－230.

［11］Zhong Weizhi，Guo Qing，Gu Xuemai.Optimal WPT selection algorithm in adaptive WPM interference rejection system.Journal of Harbin Institute of Technology，2010，42(5):727－731.

［12］Liu Runtao，Sun Zhongxi.Gray-level digital watermarking algorithm based on singular value decomposition in wavelet domain.Journal of Harbin Institute of Technology，2009，41(11):193－196.

［13］Kong Huafeng.Blind audio watermark algorithm based on chaotic spread spectrum and difference of energy coefficients.Computer Engineering，2010，36(5):131－133.

［14］Xu Dawen，Wang Rangding，Bao Jilong.Adaptive audio watermarking algorithm based on auditory model.Computer Engineering and Applications，2006，42(31):68－70.

［15］He Jianzhao.Based on Wavelet Transform，the Digital Audio Watermarking Algorithm Research and Implementation.Harbin:Harbin University of Science and Technology，2008.

［16］Huang Xionghua，Zhou Wei.Adaptive robust audio blind watermarking algorithm in DCT domain.Journal of Guilin Institute of Electronic Technology，2009，29(1):41－44.