Wave File Features Extraction using Reduced LBP

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 8, No. 5, October 2018, pp. 2780~2787
ISSN: 2088-8708, DOI: 10.11591/ijece.v8i5.pp2780-2787  2780
Journal homepage: http://iaescore.com/journals/index.php/IJECE
Wave File Features Extraction using Reduced LBP
Aws Al-Qaisi1
, Saleh A. Khawatreh2
, Ahmad A. Sharadqah3
, Ziad A. Alqadi4
1
Department of Communication Engineering, Faculty of Engineering Technology, Al-Balqa Applied University,
Yordania
2
Department of Computer Engineering, Al-Ahliyya Amman University, Yordania
3,4
Computer Engineering Department, Faculty of Engineering Technology, at Al-Balqa Applied University, Yordania
Article Info ABSTRACT
Article history:
Received Nov 29, 2017
Revised Feb 6, 2018
Accepted Aug 21, 2018
In this work, we present a novel approach for extracting features of a digital
wave file. This approach will be presented, implemented and tested. A
signature or a key to any wave file will be created. This signature will be
reduced to minimize the efforts of digital signal processing applications.
Hence, the features array can be used as key to recover a wave file from a
database consisting of several wave files using reduced Local binary patterns
(RLBP). Experimental results are presented and show that The proposed
RLBP method is at least 3 times faster than CSLBP method, which mean that
the proposed method is more efficient.
Keyword:
CSLBP
DWF
FA
LBP
RLBP Copyright © 2018 Institute of Advanced Engineering and Science.
All rights reserved.
Corresponding Author:
Aws Al-Qaisi,
Department of Communication Engineering,
Faculty of Engineering Technology, Al-Balqa Applied University, Yordania.
Email: aws.alqaisi@bau.edu.jo
1. INTRODUCTION
The Digital wave file (DWF) represents the sampled and quantized sound wave which happens to be
above or below the equilibrium or ambient air pressure [1], [2]. The sampled values are organized in one
column (for mono DWF) or in two columns (for stereo DWF), so we can deal with DWF as a matrix which
consists of positive and negative values over the its entire range of samples [3], [4], [5]. Figure 1 shows some
samples of a DWF, while the Figure 2 shows the wave of a DWF.
Figure 1. Samples of DWF

Int J Elec & Comp Eng ISSN: 2088-8708 
Wave File Features Extraction using Reduced LBP (Aws Al-Qaisi)
2781
Figure 2. Samples of DWF
One of the most used applications involved with DWF processing is voice recognition. Hence, most
of these applications use the nature of DWF to extract voice feature by mean of calculating several
parameters like zero-crossing, dynamic range and peak value [6], [7], [15]. Calculating these parameters
require understanding the voice nature, and some time they do not give an acceptable recognition ratio if we
use them to recognize the voice [6], [7], [8].
To avoid the above mentioned weakness, we propose a method which can deal with DWF using two
dimensional matrix. This can be done by converting the one or two column voice array to two dimensional
matrix with multiple rows and columns. Hence, the DWF can be represented as a gray image, as shown in
Figure 3.
Figure 3. Part of DWF represented by matrix
In this paper, once the DWF is converted to a matrix, then the features of DWF can be obtained
using the neighborhood concepts, by considering the point in the center and the point neighbors as shown in
Figure 4. This technique is called local binary pattern (LBP) [9]. LBP computes binary numbers by labeling
the pixels of an image through thresholding the 3x3-neighborhood of each pixel with the center value. The
procedures of calculating these patterns are shown in Figures 5 and Figure 6 [10], [11],[14]. Figure 7 shows
an example of calculating LBP for a matrix.
0 1 2 3 4 5 6 7 8
x 10
4
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
Time
Amplitude

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 8, No. 5, October 2018 : 2780 - 2787
2782
Figure 4. Point neighborhood
Figure 5. Generating binary pattern
Figure 6. LBP calculation procedures

2783
Figure 7. LBP calculation example
Using the LBP method to create features for signal recognition leads to extra efforts because the size
of the generated features array is very big and it contains 256 entries (repetitions for the values from 0 to 255)
[10]. To minimize the number of entries in the features array we can use center symmetric LBP (CSLBP)
which can generate a features array with 16 entries [11], [12],[13]. The procedure of calculating CSLBP is
shown in Figure 8, while the Figure 9 shows an example of calculating CSLBP. It can be shown from Figure
9 that the CSLBP reduces the feature array entries to 16. Hence, the CSLBP minimize the requirements
needed to apply voice recognition process.
Figure 8. Procedures to calculate CSLBP

 ISSN: 2088-8708
2784
Figure 9. CSLBP calculation example
2. PROPOSED METHOD
The proposed method is based on CSLBP to create a DWF features array. The main contribution of
the proposed method lay in reducing the number of entries from 16 to 4 entries. The obtained DWF features
array by proposed method can be used for various applications such as:
a. Identifying the voice, because the feature array is a unique array for each DWF.
b. Using features array as a key to retrieve a wave file from a database consisting of several wave files.
c. Using features array as an input data set to any recognition or expert system.
d. Finding whether the DWF is disrupted or not, knowing the accurate features array of original DWF.
The proposed method using reduced LBP can be implemented by applying the following steps:
a. Get DWF.
b. Get the size of the DWF array (r: row, c: columns which is always 1 for mono files or 2 for stereo one).
c. Reshape the matrix of DWF to one column array(R).
d. Get M as a multiplication of r and c.
e. Find S as a floor of the square root of M.
f. Square S to get SS.
g. Get a number of values of R equal to SS and store them in RR array.
h. Reshape RR array to a square matrix (MR) (with rows=S, and columns=S).
i. Initialize features array to zero (FA=zeros (4, 1)).
j. For each point in MR calculate the reduced LBP applying the following steps as shown in Figure 10:
1) Find a threshold value using the following formula:
T=(I(i,j+1) + I(i+1, j)+I(i,j-1)+I(i-1,j)+I(i+1,j+1)+I(i+1,j-1)+ I(i-1, j-1)+I(i-1,j+1)-8*I(i,j))/9;
2) Find the following two binary digits
A0=((I(i,j+1) + I(i+1, j)-I(i,j-1)-I(i-1,j) > T ) );
B0=((I(i+1,j+1)+I(i+1,j-1) - I(i-1, j-1)-I(i-1,j+1) > T ) );
3) Get the index of FA as:
IFA=A0*2^0+B0*2^1
4) Increment FA(IFA=1) by 1
k. Save FA.
The following Matlab code was written to implement the proposed method:
[aa fs]=wavread('C:Userswin 7Desktopvoicehorse.wav');

2785
[n1 n2]=size(aa); a1=reshape(aa,n1*n2,1); b=n1*n2;
b1=floor(sqrt(b)); a2(:,1)=a1(1:b1*b1,1);
I=reshape(a2,b1,b1);
size(I)
h=zeros(4,1);
[y x]=size(I);
for i=2:y-1
for j=2:x-1
T=(I(i,j+1)+I(i+1,j)+I(i,j-1)+I(i-1,j)+I(i+1,j+1)+I(i+1,j-1)+I(i-1, j-1)+I(i-1,j+1)-8*I(i,j))/9;
% keeping I(j,i) as center we compute CSLBP
a=((I(i,j+1) + I(i+1, j)-I(i,j-1)-I(i-1,j) > T ) * 2^0 );
b=((I(i+1,j+1)+I(i+1,j-1) - I(i-1, j-1)-I(i-1,j+1) > T ) * 2^1 );
e=a+b;
h(e+1)=h(e+1) + 1;
end
end
h
One of the sample DWF used in the implementation process size=15547 * 2
then
fix (sqrt (15547 * 2))=176
Voice matrix size=176*176
3. RESULTS AND ANALYSIS
In this section, The propped method was implemented several times using various DWF with
different sizes, Table 1 shows some implementation samples:
Table 1. Implementation Samples
Wave file Size(byte) Features
1(bird.wav) 1254528 44770 32909 32445 45112
2(cow.wav) 495424 18148 11213 11963 19192
3(cow2.wav) 279904 11499 6316 6351 10059
4(dog.wav) 737104 28220 16817 16753 28811
5(dolphin.wav) 628704 23706 15046 14906 23626
6(donkey.wav) 1903904 73216 43555 43221 75233
7(duck.wav) 3288576 122618 81543 81334 122826
8(elephant.wav) 360272 14082 8094 7847 14077
9(horse.wav) 248752 8910 6299 6118 8949
The feature array is very sensitive to any changes in the DWF, even if we change one value in the
file as shown in Table 2, so this feature array can be used as an identifier or a primary key to retrieve the
wave file.
Table 2. RLPB Sensitivity
Wave file Changes Features
1
No change 44770 32909 32445 45112
Wav(220)=0.0038 changed to 0.5 44772 32906 32446 45112
2
No change 18148 11213 11963 19192
Wav(220)=-0.0095changed to 0.5 18147 11213 11964 19192
3
No change 11499 6316 6351 10059
Wav(220)=0.0078 changed to 0.5 11496 6317 6352 10060
4
No change 28220 16817 16753 28811
Wav(220)=-0.0019 changed to 0.5 28221 16817 16754 28809
5
No change 23706 15046 14906 23626
Wav(220)=0 changed to 0.5 23708 15045 14906 23625
6
No change 73216 43555 43221 75233
Wav(220)=0.0056 changed to 0.5 73215 43555 43222 75233
7
No change 122618 81543 81334 122826
Wav(220)=-0.0075 changed to 0.5 122620 81540 81335 122826
8
No change 14082 8094 7847 14077
Wav(220)=-0.0210 changed to 0.5 14079 8095 7848 14078
9
No change 8910 6299 6118 8949
Wav(220)=-0.0156 changed to 0.5 8911 6299 6117 8949

 ISSN: 2088-8708
2786
From the results of implementation, we can raise the following facts:
a. Each DWF has a unique features array.
b. FA consists of only 4 elements.
c. FA does not depend on DWF size and has always the size of 32 bytes.
d. FA can be used as a key or a signature to deal with digital wave files.
e. Using 4 element array reduces all the efforts concerning any application that deals with digital wave
files.
The proposed RLBP method was implemented several times using various wave files with different
sizes, the time to extract features was computed (T2), the same files were treated using CSLBP method and
the features extraction time was also computed (T1). The experimental results are shown in Table 3. From
Table 3, it can be seen that the speedup of the proposed method is at least 3 times faster, which mean that the
proposed method is more efficient compared with CSLBP method
Table 3. Experimental Comparison Results
Wave file size(byte)
CSLBP features
extraction time (Sec.) T1
Proposed RLBP features
extraction time (Sec.) T2
Speedup
2880000 1.953460 0.589395 3.3143
1903904 0.939946 0.290760 3.2327
1254528 0.418782 0.120215 3.4836
737104 0.338388 0.098481 3.4361
628704 0.100056 0.044843 2.2313
495424 0.030078 0.009357 3.2145
360272 0.030955 0.008587 3.6049
3288576 0.030710 0.008232 3.7306
279904 0.030672 0.007565 4.0545
248752 0.028529 0.006089 4.6853
4. CONCLUSION
A new method of digital wave file feature extraction was proposed, implemented and tested. The
proposed method has some advantages comparing with other existing method and it can suit any application
dealing with wave files. The extracted features were unique to a wave file and they can be used as an
excellent key or signature for a digital wave file.
REFERENCES
[1] Ashraf Abu-Ein, et al., " A Technique of Hiding Secrete Text in Wave File," International Journal of Computer
Applications, vol. 2, pp. 0975 – 8887, 2016.
[2] R. Arulmozhiyal and K. Baskaran, ", Experimental Investigation of Wave File Compression-Decompression,"
The International Journal of Computer Science and Information Security, vol.14, pp. 774, 2016.
[3] Jihad Nadir, et al., " A Technique to Encrypt-decrypt Stereo Wave File," International Journal of Computer and
Information Technology, Vol 5, pp. 465-470, 2016.
[4] Zhihua Cui, et al., " Training artificial neural networks using APPM", International Journal of Wireless and
Mobile Computing, Vol 5, pp. 168-174, 2012.
[5] Z. Yinhai, et al., " Jigsaw-based secure data transfer over computer networks," in Int. Conference on Information
Technology: Coding and Computing, 2004, pp. 2-6.
[6] Khaled Matrouk, et al., " Speech Fingerprint to Identify Isolated Word-Person," World Applied Sciences Journal,
vol. 31, pp. 1767-1771, 2014.
[7] Bin Wu, et al., Dynamic range estimation for nonlinear systems," in IEEE/ACM International Conference on
Computer-Aided Design, 2004.
[8] Timo O, et al., " Multi resolution gray-scale and rotation invariant texture classification with local binary patterns,"
IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 24, pp. 971 - 987, 2002.
[9] Marko H, et al., " Description of interest regions with local binary patterns," Pattern Recognition, vol. 42, pp425-
436, 2009.
[10] T. Ojala, et al., " A comparative study of texture measures with classification based on feature distributions,"
Pattern Recognition, vol. 29, pp. 51-59, 1996.
[11] K. J. Priya and R. S. Rajesh., " A Local Min-Max Binary Pattern Based Face Recognition Using Single Sample per
Class," International Journal of Advanced Science and Technology, vol. 36, pp. 41-50, 2011.
[12] Zhang J, et al., " Local features and kernels for classification of texture and object categories: A comprehensive
study," International Journal of Computer Vision, vol.73 , pp. 213-238, 2007.
[13] Sabina Yasmin, et al., " Performance Study of Soft Local Binary Pattern over Local Binary Pattern under Noisy
Images," International Journal of Electrical and Computer Engineering, vol.6 , pp. 1161- 1167, 2015.

2787
[14] Taha H. Rassem, et al., "Face Recognition Using Completed Local Ternary Pattern (CLTP) Texture Descriptors,"
International Journal of Electrical and Computer Engineering, vol.7 , pp. 1594-1601, 2017.
[15] Fooad Jalili, et al., " Speech Recognition Using Combined Fuzzy and Ant Colony Algorithm," International
Journal of Electrical and Computer Engineering, vol.7 , pp. 2205-2210, 2016.
BIOGRAPHIES OF AUTHORS
Dr Aws Al-case is an associated professor in the Communication Engineering Department,
Faculty of Engineering and Technology, Al-Balqa’ Applied University, Jordan. Tudor was
received his PhD and MSc in communication and signal processing from Newcastle university in
2006 and 2010 respectively. Dr. Aws research interest includes Digital signal processing,
seismic signal processing, Wireless communication, Digital communication and wireless sensor
network. He served as reviewer in many international journals where he has published more than
16 scientific papers in the field of communication and signal processing.
Dr Saleh A.Khawatreh is Assistant Professor in Computer Engineering Departement ,Faculty of
Engineering,Al-Ahliyya Amman University. Dr Saleh A.Khawatreh research interest includes
Computer Networks, computer Application and signal processing.
Dr. Ahmed A.M Sharadqh received his PhD Degree in Computer, computing system and
networks from National Technical of Ukraine "Kyiv Polytechnic Institute Ukraine in 2007. Since
2009, Dr. Ahmed sharadqh has been an Associate professor in the Computer Engineering
Department, Faculty of Engineering Technology, at Al-Balqa Applied University. His research
interests include Performance of network, Quality services, security network, image processing,
digital systems design, operating system, and Microprocessors.
Dr Ziad A.A.Alqadi has been professor in the Computer Engineering Department, Faculty of
Engineering Technology, at Al-Balqa Applied University. He is the Head of Computer
engineering department. His research interests include Signal processing, parallel processing,
image processing.

Wave File Features Extraction using Reduced LBP

More Related Content

What's hot (19)

Similar to Wave File Features Extraction using Reduced LBP (20)

More from IJECEIAES (20)

Recently uploaded (20)

Wave File Features Extraction using Reduced LBP