DIP

1. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 1 | Page
Analysis of speech signal MLBP features
Prof. Ziad Alqadi, Dr. Mohammad S. Khrisat, Dr. Amjad Hindi,
Dr. Majed Omar Dwairi
Albalqa Applied University, Faculty of engineering technology, Jordan, Amman
Abstract: The digital audio signal is one of the most important types of data and the most used in
communication. It is used in many vital applications, the most important of which is digital protection systems.
And since the volume of the audio file is large, its use in carrying out the matching and verification process
requires a large amount of time, which leads to the low effectiveness of the security and protection system. So
we have to find a suitable way to represent the voice with a new number and a few values that can be used as a
sound features. In this paper we will discuss in details how to use MLBP method of features extraction, we will
show how this method is stable, flexible and efficient.
Keywords: Speech, features vector, LBP, MLBP, K_mean clustering, LPC, WPT, FIR, throughput, extraction
time, FS.
Introduction
Digital signals [1], [2] such digital audio signals (speech) [3], [4] and digital images (gray and color)
[5], [6], [7] are very important type of data because they are using in any vital applications such banking
systems, security systems and computer classification systems [4], [5]. Here in this paper we will in details
analyze the modified local binary pattern method of features extraction, which can be easily used in a human
speech classification system (HSCS) [6], [7].
The digital speech signal is a one column matrix ( mono signal), or two column matrix (stereo signal),
each column represents the samples amplitude [1], [8],[9], these samples values are obtained as a result of
converting analogue speech to digital as shown in figure 1 [10], [11], [12] by sampling (stage 1) and
quantization (stage 2).
Figure 1: Converting speech analogue signal to digital
Speech signal is an important digital data type due to the vital applications requiring this kind of data,
these applications such as security systems application [3], [4] require a high speed of implementation, but the
speech signals usually have a big size, and thus will negatively affects the system efficiency and here we will
seek a method to represent the speech by a small number of values to increase the process of speech
manipulation. Speech signal file size depends on the recording time and the sampling rate [7], [8].
The sampling frequency or sampling rate, fs, is the average number of samples obtained in one second (samples
per second), thus fs = 1/T. Table 1 shows some information about the speech signals which we will investigate
in this paper [7], [8], [9].

2. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 2 | Page
Table 1: Used speech signal files
Speech
#
Spoken words Fs Recording
time(seconds)
Size(samples) Size(bytes)
1 Aqaba is a beautiful city, it is
located on the red sea
44100 5.7832 255037 2040296
2 Stay home stay safe 44100 2.8451 125469 1003752
3 Albalqa applied university 44100 3.5109 154829 1238632
4 Amman is the capital city of Jordan 44100 4.1620 183544 1468352
5 How are you 44100 1.9204 84691 677528
6 My name is Ziad 44100 2.5021 110344 882752
7 Please open the door 44100 2.5362 111848 894784
8 Please shut down the computer 44100 3.3558 147990 1183920
9 Speech signal analysis 44100 2.9507 130127 1041016
10 Good by 44100 1.6909 74569 596552
Average 3.1257 137840 1102800
From table 1 we can see that the average number of samples is big, so the average file size is also big,
and this will lead to extra time to identify the speech [26], [27], so we can represent the speech file by a
histogram [12], [13],[14], [28] of 256 values and with size equal 2048 bytes for each speech file[5], [6], [9].
To reduce the classification time [14], [15], [16].we have to use signal features instead of using the
signal [13], the features vector for each speech signal will be unique, simple, fixed and will contain a small
number of values [17], [18], [19], [20].
Existing features extraction methods
Many methods were introduced to extract features for the digital speech signal, some of these methods
were based on calculating local binary pattern (LBP) operators [21], [22],[23], [24], the method here in this
paper to be analysed is modified LBP (MLBP) method, which will be introduced later in this section.
Some methods used the concepts of data clustering such as K_mean method of clustering (KMC)
[29],[30],[31],[32], in which we can use the centroids as a features, this method is not stable in generating the
features, the features can be changed from run to run and it requires a big amount of time.
Other methods were based on wavelet packet tree decomposition (WPD) [33], [34], [35], this method is
efficient, but the number of levels required for decomposition varies for speech to another, especially when the
speech file size is not fixed.
Other methods were based on using finite impulse response filter (FIR) coefficients as features [25],
[26] based on linear prediction coding, this method creates a stable features and it is efficient by providing a
small amount of features extraction time.
The speech files shown in table 1 were implemented and table 2 summarizes the total results of
implementation.
Table 2: Summery results of the used method of features extraction
Method Average extraction
time(second)
Throughput(samples
per second)
WPT 0.1466 931062
LPC 0.1052 1409429
KMC 10.92503 12494
The proposed MLBP method of speech features extraction can be implemented applying the following steps:
- Get the speech file.
- Reshape the speech file from one-column (mono speech) or two-column (stereo speech) matrix to one
row matrix.
- Initialize the features vector to zeros (4 elements vector).
- For each sample in the one row matrix apply the steps shown in table 3

3. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 3 | Page
Table 3: MLBP calculation
Samples …. S(i-2) S(i-1) S(i) S(i-1) S(i-2) …….
Values … -0.5 1 0 -0.8 1 …
<= <=
Binary 1 0
Decimal 2
So add 1 to the vector with index=2
Implementation and results discussion
Experiment 1: Varying sampling frequency for the same speech signal
Here we took the spoken words "ziad alqadi" and recorded the speeches using various sampling rate
(frequency), table 4 shows the results of implementation:
Table 4: Experiment 1 results
FS Features Samples ET(seconds)
8000 5312 539 435 12041 18331 0.0010
11025 7318 689 606 16645 25262 0.0020
12000 8079 646 521 18246 27496 0.0021
16000 10743 775 667 24473 36662 0.0023
22050 14819 1039 871 33792 50525 0.0026
24000 16387 819 678 37105 54993 0.0030
32000 20990 1830 1742 48758 73324 0.0034
44100 22272 9088 460 69215 101039 0.0040
48000 22732 11407 0 75832 109975 0.0060
Average 55290 0.0029
From table 4 we can see the following:
- Changing FS leads to changing the speech features.
- The method is efficient by providing an average extraction time 0.0029 seconds with a throughput of
19066000 samples per second.
- Number of samples increases when FS increases.
- Extraction time increases when FS increases (see figure 2)
Figure 2: Relationship between FS, number of samples and extraction time
Experiment 2: Varying sampling frequency for the same speech signal
In this experiment we took different spoken words by the same person fixing FS to 44100, the features of the
speech file were extracted, table 5 shows the results of this experiment.
0 2 4 6
x 10
4
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
x 10
-3
FS
Extractiontime(second)
0 2 4 6
x 10
4
1
2
3
4
5
6
7
8
9
10
11
x 10
4
FS
Samples

4. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 4 | Page
Table 5: Experiment 2 results
Spoken
word
Features Samples ET(second)
One 6216 2491 88 50419 59218 0.0020
Two 9642 3985 200 55410 69241 0.0030
Three 6619 2814 181 49930 59548 0.0023
Four 4217 1809 137 47030 53197 0.0017
Five 8549 3578 243 51395 63769 0.0027
Sex 5944 2917 446 50284 59595 0.0024
Seven 8904 3746 256 51928 64838 0.0029
Eight 5435 2240 121 48779 56579 0.0024
Nine 8677 3412 101 50653 62847 0.0025
Ten 8361 3494 181 50876 62916 0.0026
Average 61175 0.0024
From table 5 we can see the following:
- The method provides a high speed extraction process.
- The features for each spoken word are unique, thus we can identify the person and the spoken word.
- The results show that we can use this method to handle the pass-wording application system.
Experiment 3: The same words spoken by different persons
Here we took the spoken sentence "My name is ziad alqadi", table 6 shows the results of this experiment
Table 6: Experiment 3 results
Person Features FS Samples ET
1 6982 1631 1578 7277 11025 17472 0.0010
2 14068 84 100 14217 44100 28473 0.0017
3 23518 5115 3367 32313 11025 64317 0.0028
4 18220 5351 5241 18072 11025 46888 0.0021
5 30916 6815 7011 30980 11025 75726 0.0032
From table 6 we can see the following:
- For each speech the extracted features were unique.
- The extraction process was repeated several time and the features remain the same.
- The method can be considered stable and efficient
The speeches shown in table 1 were treated using MLBP method and the average extraction time was equal
0.0069 seconds, table 7 shows the summer results comparing with other methods results:
Table 7: Speed up of MLBP method
Method Average extraction
time(second)
Throughput(samples
per second)
Speed up of MLBP
WPT 0.1466 931062 21.2
LPC 0.1052 1409429 15.2
KMC 10.92503 12494 1583.3
MLBP 0.0069 19782000 1
From table 7 we can see that MLBP method is the most efficient method by providing minimum
extraction time and maximum throughput.
Conclusion
MLBP method of speech features extraction was introduced, tested and implemented; the experimental
results showed that the introduced method is very efficient comparing with other methods, the extracted features
for each speech file were stable and unique, the features can be easily used to identify the person and to identify
the spoken words by a certain person.

5. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 5 | Page
References
[1]. Ziad Alqadi, Bilal Zahran, Jihad Nader, Estimation and Tuning of FIR Lowpass Digital Filter
Parameters, International Journal of Advanced Research in Computer Science and Software
Engineering, vol. 7, issue 2, pp. 18-23, 2017.
[2]. Haitham Alasha'ary, Abdullah Al-Hasanat, Khaled Matrouk, Ziad Al-Qadi, Hasan Al-Shalabi, A Novel
Digital Filter for Enhancing Dark Gray Images, European Journal of Scientific Research , pp. 99-106,
2014.
[3]. Majed O Al-Dwairi, Ziad A Alqadi, Amjad A Abujazar, Rushdi Abu Zneit, Optimized true-color image
processing, World Applied Sciences Journal, vol. 8, issue 10, pp. 1175-1182, 2010.
[4]. Jamil Al Azzeh, Hussein Alhatamleh, Ziad A Alqadi, Mohammad Khalil Abuzalata, Creating a Color
Map to be used to Convert a Gray Image to Color Image, International Journal of Computer
Applications, vol. 153, issue 2, pp. 31-34, 2016.
[5]. Musbah J. Aqel, Ziad A. Alqadi, Ibraheim M. El Emary, Analysis of Stream Cipher Security
Algorithm, Journal of Information and Computing Science, vol. 2, issue 4, pp. 288-298, 2007.
[6]. Jamil Al-Azzeh, Ziad Alqadi, Mohammed Abuzalata, Performance Analysis of Artificial Neural
Networks used for Color Image Recognition and Retrieving, International Journal of Computer Science
and Mobile Computing, vol. 8, issue 2, pp. 20 – 33, 2019.
[7]. Akram A. Moustafa and Ziad A. Alqadi, Color Image Reconstruction Using A New R'G'I Model,
Journal of Computer Science, vol. 5, issue 4, pp. 250-254, 2009.
[8]. Dr. Ziad Alqadi, Akram Mustafa, Majed Alduari, Rushdi Abu Zneit, True color image enhancement
using morphological operations, International review on computer and software, vol. 4, issue 5, pp.
557-562, 2009.
[9]. Mohammed Ashraf Al Zudool, Saleh Khawatreh, Ziad A. Alqadi, Efficient Methods used to Extract
Color Image Features, IJCSMC, vol. 6, issue 12, pp. 7-14, 2017.
[10]. Ayman Al-Rawashdeh, Ziad Al-Qadi, Using wave equation to extract digital signal features,
Engineering, Technology & Applied Science Research, vol. 8, issue 4, pp. 1356-1359, 2018.
[11]. AlQaisi Aws, AlTarawneh Mokhled, A Alqadi Ziad, A Sharadqah Ahmad, Analysis of Color Image
Features Extraction using Texture Methods, TELKOMNIKA, vol. 17, issue 3, 2018.
[12]. Dr Ziad A AlQadi, Hussein M Elsayyed, Window Averaging Method to Create a Feature Victor for
RGB Color Image, International Journal of Computer Science and Mobile Computing, vol. 6, issue 2,
pp. 60-66, 2017.
[13]. Dr. Ghazi. M. Qaryouti, Prof. Ziad A.A. Alqadi, Prof. Mohammed K. Abu Zalata, A Novel Method for
Color Image Recognition, IJCSMC, Vol. 5, Issue. 11, pp.57 – 64, 2016.
[14]. Jihad Nader Ismail Shayeb, Ziad Alqadi, Analysis of digital voice features extraction methods,
International Journal of Educational Research and Development, vol 4, issue 1, pp. 49-55, 2019.
[15]. Ahmad Sharadqh Naseem Asad, Ismail Shayeb, Qazem Jaber, Belal Ayyoub, Ziad Alqadi, Creating a
Stable and Fixed Features Array for Digital Color Image, IJCSMC, vol. 8, issue 8, pp. 50-56, 2019.
[16]. Majed O. Al-Dwairi, Amjad Y. Hendi, Mohamed S. Soliman, Ziad A.A. Alqadi, A new method for
voice signal features creation, International Journal of Electrical and Computer Engineering (IJECE),
vol. 9, issue 5, pp. 4092-4098, 2019.
[17]. Dr. Amjad Hindi Dr. Majed Omar Dwairi Prof. Ziad Alqadi, PROCEDURES FOR SPEECH
RECOGNITION USING LPC AND ANN, International Journal of Engineering Technology Research
& Management, vol. 4, issue 2, pp. 48-55, 2020.
[18]. Yousf Eltous Ziad A. AlQadi, Ghazi M. Qaryouti, Mohammad Abuzalata, ANALYSIS OF DIGITAL
SIGNAL FEATURES EXTRACTION BASED ON KMEANS CLUSTERING, International Journal
of Engineering Technology Research & Management, vol. 4, issue 1, pp. 66-75, 2020.
[19]. Prof. Yousif Eltous, Dr. Ghazi M. Qaryouti, Prof. Mohammad Abuzalata, Prof. Ziad Alqadi,
Evaluation of Fuzzy and C_mean Clustering Methods used to Generate Voiceprint, IJCSMC, vol. 9,
issue 1, pp. 75 -83, 2020.
[20]. Ahmad Sharadqh Naseem Asad, Ismail Shayeb, Qazem Jaber, Belal Ayyoub, Ziad Alqadi, Creating a
Stable and Fixed Features Array for Digital Color Image, IJCSMC, vol. 8, issue 8, pp. 50-56, 2019.
[21]. Ismail Shayeb, Ziad Alqadi, Jihad Nader, Analysis of digital voice features extraction methods,
International Journal of Educational Research and Development, vol. 1, issue 4, pp. 49-55, 2019.
[22]. Aws Al-Qaisi, Saleh A Khawatreh, Ahmad A Sharadqah, Ziad A Alqadi, Wave File Features
Extraction Using Reduced LBP, International Journal of Electrical and Computer Engineering, vol. 8,
issue 5, pp. 2780, 2018.

6. International Journal of Latest Research in Engineering and Technology (IJLRET)
ISSN: 2454-5031
www.ijlret.com || Volume 06 - Issue 04 || April 2020 || PP. 01-06
www.ijlret.com 6 | Page
[23]. Ziad Alqad, Prof. Yousf Eltous Dr. Ghazi M. Qaryouti, Prof. Mohammad Abuzalata, Analysis of
Digital Signal Features Extraction Based on LBP Operator, International Journal of Advanced
Research in Computer and Communication Engineering, vol. 9, issue 1, pp. 1-7, 2020.
[24]. Ziad Alqadi, Aws Al-Qaisi, Adnan Manasreh, Ahmad Sharadqeh, Digital Color Image Classification
Based on Modified Local Binary Pattern Using Neural Network, IRECAP, vol. 9, issue 6, pp. 403-408,
2019.
[25]. Dr. Amjad Hindi Prof. Yousif Eltous Prof. Mohammad Abuzalata Prof. Ziad Alqadi Dr. Ghazi M.
Qaryouti, USING FIR FILTER COEFFICIENTS TO CREATE COLOR IMAGE FEATURES,
International Journal of Engineering Technology Research & Management, vol. 4, issue 2, pp. 6-14,
2020.
[26]. Ziad Alqadi, Bilal Zahran, Jihad Nader, Estimation and Tuning of FIR Lowpass Digital Filter
Parameters, International Journal of Advanced Research in Computer Science and Software
Engineering, vol. 7, issue 2, pp. 18-23, 2017.
[27]. Prof. Yousif Eltous Dr. Amjad Hindi, Prof. Ziad Alqadi, Dr. Ghazi M. Qaryouti, Prof. Mohammad
Abuzalata, Using FIR Coefficients to Form a Voiceprint, International Journal of Innovative Research
in Electrical, Electronics, Instrumentation and Control Engineering, vol. 8, issue 1, pp. 1-6, 2020.
[28]. Dr. Amjad Hindi Dr. Majed Omar Dwairi Prof. Ziad Alqadi, PROCEDURES FOR SPEECH
RECOGNITION USING LPC AND ANN, International Journal of Engineering Technology Research
& Management, vol. 4, issue 2, pp. 48-55, 2020.
[29]. Yousf Eltous Ziad A. AlQadi, Ghazi M. Qaryouti, Mohammad Abuzalata, ANALYSIS OF DIGITAL
SIGNAL FEATURES EXTRACTION BASED ON KMEANS CLUSTERING, International Journal
of Engineering Technology Research & Management, vol. 4, issue 1, pp. 66-75, 2020.
[30]. Prof. Yousif Eltous, Dr. Ghazi M. Qaryouti, Prof. Mohammad Abuzalata, Prof. Ziad Alqadi,
Evaluation of Fuzzy and C_mean Clustering Methods used to Generate Voiceprint, IJCSMC, vol. 9,
issue 1, pp. 75 -83, 2020.
[31]. Ahmad Sharadqh Naseem Asad, Ismail Shayeb, Qazem Jaber, Belal Ayyoub, Ziad Alqadi, Creating a
Stable and Fixed Features Array for Digital Color Image, IJCSMC, vol. 8, issue 8, pp. 50-56, 2019.
[32]. Ahmad Sharadqh Jamil Al-Azzeh, Rashad Rasras , Ziad Alqadi , Belal Ayyoub, Adaptation of matlab
K-means clustering function to create Color Image Features, International Journal of Research in
Advanced Engineering and Technology, vol. 5, issue 2, pp. 10-18, 2019.
[33]. Dr. Ghazi M. Qaryouti, Prof. Mohammad Abuzalata, Prof. Yousf Eltous, Prof. Ziad Alqadi,
Comparative Study of Voice Signal Features Extraction Methods, IOSR Journal of Computer
Engineering (IOSR-JCE), vol. 22, issue 1, pp. 58-66, 2020.
[34]. Amjad Y. Hindi, Majed O. Dwairi, Ziad A. AlQadi, Analysis of Digital Signals using Wavelet Packet
Tree, IJCSMC, vol. 9, issue 2, pp. 96-103, 2020.
[35]. Amjad Y. Hindi, Majed O. Dwairi, Ziad A. AlQadi, Creating Human Speech Identifier using WPT,
International Journal of Computer Science and Mobile Computing, vol. 9, issue 2, pp. 117 – 123, 2020.