You are on page 1of 5

Research for Synchronic Audio Information Hiding Approach

Based On DWT Domain

Wang Junjie Mo Qian*


(Network Center, Beijing Technology and (Network Center, Beijing Technology and
Business University,100048,China, Business University,100048,China,
amerwang@tom.com) moqian@th.btbu.edu.cn)

Mei Dongxia Yao Jun


(Beijing Municipal Supply and (Network Center, Beijing Technology and
Marketing Institute, 100076,China, Business University,100048,China,
frenchamei@163.com) yaojun@btbu.edu.cn)

tions [1-2].
Abstract Based on the method of information hiding, a Actually, cryptography hides the meaning of
novel algorithm for security speech communication is information and makes the information seem as
designed in this paper,and the synchronization code can stochastic codes.When the adversaries obtain the
be used to search the embedded location. Firstly, the encryption information, they will know the existence of
DWT is performed on each segment of the original the secret information by the sense organ, and then the
carrier audio; Secondly, the embedding bits are risk of being attacked increases[3]. But information
constructed by the synchronization code and secret hiding technology hides the existence of the secret
audio ,and the secret bits are formed by chaotic information, thus reduces the possibility of being
encrypting; Thirdly, the secret bits are embedded into the attacked. So,it is attracting people's interests and widely
low frequency point of Discrete Wavelet Transform
used in many fields,such as military security,secret
(DWT) by quantization;Lastly,the IDWT are performed
communication and so on.The information hiding
on each segment and the setgo audio are constructed.The
algorithms can be classified into two categories:
original carrier audio is not required in the secret audio
algorithms in time domain and algorithms in transform
recovery. Experimental results show that the stego audio
domain[4].
has transparent feature, and the quality of the recovered
In this paper, based on chaotic sequence, an audio
audio is satisfying. The algorithm is strongly robust to
information hiding algorithm in wavelet domain is
many attacks, such as resampling, cropping and so on..
presented. A synchronic signal is applied in the algorithm
Keywords Information Hiding; Quantization; Chaotic
to search the embedding location of the secret audio.
Encrypting; Synchronization Code;DWT
Frame synchronization is used to search synchronic
Ⅰ. Introduction signal.
Information hiding is a new subject with profound
historical origins. It is associated with lots of subjects Ⅱ Principle of Embedding Secret Information
such as information theory, cryptology, sensation science
By Quantizing
and etc. The Information hiding technology is mainly
Quantizing ,whose rule is that coefficient to be
about how to hide the secret information into the
embedded is quantized to the center of the nearest
cover-media to carry on the secret communication.
quantization interval away from the coefficient to be
Carrier signal shields the actual information of the
embedded . Supposing g is the coefficient to be
communica-
quantized, g is the quantized coefficient, w is the
* Corresponding author.
* This paper is supported by Beijing Science New NovelFund(2006B10)

978-1-4244-4589-9/09/$25.00 ©2009 IEEE

Authorized licensed use limited to: HOLON ACADEMIC INSTITUTE OF TECHNOLOGY. Downloaded on February 24,2010 at 04:22:41 EST from IEEE Xplore. Restrictions apply.
secret bi- nary bit, the embedding rule is described in audio information can be embedded into the original
figure 1 and the quantized result can be described as carrier audio by five steps. The construction of
the following formulas: embedding information is shown in Fig. 3, and the
process of secret audio information embedded into
original audio is shown in Fig. 2.

Original Audio Secret Audio Synchronization


Information Code

Fig 1. Embedding Rule By Quantizing Segmenting


Encoding
(1) if g  0, w  1

 1 DWT Chaotic Encrypting


2k   2  if n  2k

2k   1 
 2
 (1) Quantizing Embedding
 1
g   if n  (2k  1)and   
 2
IDWT
 1
2(k  1)  2 

 if n  (2k  1)and   1  Setgo Audio
 2 Fig 2. Process of Secret Audio Embedded into Original Audio

(2) if g  0, w  0 Step 1: the origin digital audio is segmented and then


the DWT are performed on each segment;
 1
(2k 1) 2 , if n  2k 1 Step 2: the embedding information is constru-
 cted by the synchronization code and secret audio
 1 1

g  2k , if n  2k &   (2)
information.
 2 2
 1 1 Step 3:logistic chaotic sequence is used to encrypt
(2k 1) 2 , if n  2k &  2  the embedding bits;

Step 4:the secret bits are embedded into the low
(3) if g<0, w  1 frequency point of Discrete Wavelet Transform (DWT)
by quantizing;
 1
(2k 1) 2  if n (2k 1) Step 5: IDWT is performed on each segment and the
 setgo audio is constructed.
 1 1 (3)
g  2k  if n 2k &  
 2 2
 1 1
(2k 1) 2 ,ifn 2k &  2 

(4) if g<0, w  0
 1
2k 2 if n 2k
 (4)
 1 1
g 2k  if n (2k 1)&  
 2 2
 1 1
2(k 1) 2 if n (2k 1)&  2

Ⅲ Audio Embedding Algorithm

In our audio embedding algorithm, the secret

Authorized licensed use limited to: HOLON ACADEMIC INSTITUTE OF TECHNOLOGY. Downloaded on February 24,2010 at 04:22:41 EST from IEEE Xplore. Restrictions apply.
Segment k-1 Segment k Segment k+1

Synchronization Secret Audio Synchronization Secret Audio Synchronization Secret Audio


… Code Information Code Information Code Information

Fig 3. Construction of Embedding Information

Relative Parameters:
Synchronization code is m sequence and its cycle is 63
Ⅳ Audio Extracting Algorithm and the judge threshold is 0.85;the initial value x0 and
parameter u of chaotic sequence is 0.25 and 3.94
The flow of extracting the secret audio informa-
respectively;. the synchronization code embedding
tion is shown in Figure 4. Without the original audio, the
location is 9th and the Secret Audio embedding location
algorithm can extract the secret audio information from
is 11th; Quantization stepΔis 0.2.
the distorted setgo audio.
Evaluation Standard:
Setgo Audio
As defined in [5], signal to noise ratio for the
watermark embedded in time-domain is computed using
Searching Synch- the relation:
ronization Code
1 N 1
1 ci'  ci
BER 
N
 0 ci'  ci
(5)
i 0 
Segmenting

 N

DWT   x i2  (6)
SNR  10 log 10  N i 1

  ( x i'  x i ) 2 
Quantizing  i 1 

Extraction In formula (5), represents recovery bit and


c i'
c i stands for bits of secret sequence. In formula (6), xi
Decoding Chaotic Decrypting represents the sample of input audio sequence and
x i' stands for the sample of modified audio.
5.1 Basic Test
In the condition of no attack, the waveform of original
Secret Audio Information
carrier audio and stego audio are shew in Figure 5 and
Fig 4. Process of Secret Audio Extraction Figure 6 respectively. In figure 5,(a) and (b) is the
waveform of the orignianl carrier audio and stego audio
When extracting the secret audio information , the
respectively;in figure 6,(a) and (b) is the waveform of
synchronization is then established. According to the
secret audio and recovery audio.
positions of the synchronization codes, the stego audio
signal is split into blocks. Each block are DWT
1
transformed. By quantizing extraction and chaotic
decrypting, the secret audio information 0
is extracted.
-1
Ⅴ Experimental Results and analysis 5
0 1 2 3 4 5 6 7 8 9 10 ×10
(a)
Experimental Conditions:
Carrier audio is a 16-bit signed mono audio Carrier Audio

signal sampled at 44.1 KHz with the length of 47


seconds in WAVE format.
Secret audio is a 16-bit signed voice signal sampled at
44.1 KHz with the length of 2.03 seconds.

Authorized licensed use limited to: HOLON ACADEMIC INSTITUTE OF TECHNOLOGY. Downloaded on February 24,2010 at 04:22:41 EST from IEEE Xplore. Restrictions apply.
Table 1 Changing the embedded location

1 Location of Embedded SNR of BER of

Syn Code Location of Stego Recovery Audio


0
Secret Audio Audio

-1 8 9 19.927 0.0082
5
0 1 2 3 4 5 6 7 8 9 10 ×10 8 10 24.325 0.0101

8 11 26.792 0.0122
(b) Stego Audio
8 12 27.719 0.0183
Fig 5. Carrier audio and Stego audio
9 10 24.418 0.0101

9 11 26.957 0.0122
1
11 12 27.981 0.0183

11 12 28.263 0.0346
0
12 14 29.341 0.0659

-1 13 14 28.343 0.0658
0 2000 4000 6000 8000 10000 12000
13 15 28.112 0.1214
(a) Secret Audio
14 16 33.066 0.252

5.3 Robustness Test


1
In order to test the robustness of the synchroni-
0 zation codes against some attachs,we have done the
following experiment.
-1 ① Resampling
0 2000 4000 6000 8000 10000 12000 The audio signal is first down sampled at 22.05 KHz,
(b) Recovery Audio and then up-sampled at 44.1 KHz.
Fig 6. Secret Audio and Recovery Audio ② Noise Interference
The interference from white noise to the stego audio is
Experimental results show that the stego audio has verified.Set cut-off frequency is 2kHz,and
transparent feature, and the quality of the recovered SNR is 0 dB. Experimental result shows we can extract
audio message is satisfying, there is no obvious the secrect audio correctly in the circumstances of strong
difference between the original carrier audio and the noise.
recovered audio message in the hearing. ③ MP3 Compression
5.2 Test of Changing the Embedded Location The audio signals are compressed into MP3 format by
In order to test the influence on the algorithm MPEG-1 Layer-3 (LAME 3.92) encoder and decoded.
of changing the embedded location, of embedding ④ Requalization
secret audio and synchronization code were changed Requantizing the original 16-bit audio signal down to
respectively.Experimental Result shows that with the 8-bit samples and back to 8-bit again.
location becoming higher, the stego audio is becoming ⑤ LP filtering
more and more difficultly to distinguish from the original Using a second-order Butterworth filter with cut-off
public carrier audio. The experimental result is showed frequency 10 KHz.
in the Table 1. The results after the described attacks are given in
Listening tests for distinguishing the original public Table 2. Results prove that the recovery audio can be
carrier audio from the stego audio have been done,but distinguished for resampling and requalization attack and
the dissimilarities between the two signals were the stego audio has good perceptually transparent.
negligible.

Authorized licensed use limited to: HOLON ACADEMIC INSTITUTE OF TECHNOLOGY. Downloaded on February 24,2010 at 04:22:41 EST from IEEE Xplore. Restrictions apply.
Table 2 The Experiment Result of Some Attacks

Attacks SNR BER

Resample 24.4078 0.0591

Noise Interference 24.4009 0.0601

MP3 Compression 23.964 0.0671

Requalization 23.5879 0.1007

LP filtering 11.2301 0.3456

We also tested robustness of the synchronization codes


against cropping. Experimental results shows that the
algorithm is

also robust cropping.

Ⅵ Conclusion

In this paper, we proposed a blind audio water-


marking algorithm in the frequency domain that is proper
for covert communication. Experimental results show
that embedding the carefully designed synchronization
codes together with the secret voice to resist some
attacks such as resampling,noise interference, MP3
compression,
cropping and so on..The security of algorithm is good
and the capacity is large. It is fit for covert
communication, it proposes a new measure for covert
communication.

REFERENCES

[1] Wei Li, Xiangyang Xue, and Peizhong Lu. Localized Audio
Watermarking Technique Robust Against Time-Scale
Modification.IEEE Transactions on multimedia,VOL.8, No. 1,
FEBRUARY 2006:p 60-69.
[2] C. W. Tang and H. M. Hang, “A feature-based robust digital image
watermarking algorithm,” IEEE Trans. Signal Process., vol. 51,
no. 4, Apr. 2003, p.950–959.
[3] Li Li, Zong Fu-Jian. An Anti-cutting Watermarking
Algorithm.NCIG2005, p. 149-152.
[4] M. P. Kennedy,G. Kolumban.Digital communication using
chaos,Signal Process., vol. 80, 2000,p. 1307–1320.
[5] Jiang Ji-tao, Zhou Xue-qin, Liu Xiao-hong. An improved
algorithm based on LSB in digital image hidden. Journal of
Shangdong University of Technology(Sci & Tech), Vol.20 No.3
May 2006 :p. 66-68.

Authorized licensed use limited to: HOLON ACADEMIC INSTITUTE OF TECHNOLOGY. Downloaded on February 24,2010 at 04:22:41 EST from IEEE Xplore. Restrictions apply.

You might also like