Performance Comparison of Musical Instrument Family Classification Using Soft Set

Saima Anwar Lashari, Rosziati Ibrahim & Norhalina Senan
International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (3), Issue (4): 2012 100
Performance Comparison of Musical Instrument Family
Classification Using Soft Set
Saima Anwar Lashari hi100008@siswa.uthm.edu.my
Faculty of Computer Science & information technology
Universiti Tun Hussein Onn Malaysia, Malaysia`
Batu Pahat, 86400, Malaysia
Rosziati Ibrahim rosziati@uthm.edu.my
Universiti Tun Hussein Onn Malaysia, Malaysia
Norhalina Senan halina@uthm.edu.my
Universiti Tun Hussein Onn Malaysia, Malaysia
Abstract
Nowadays, it appears essential to design automatic and efficacious classification algorithm for the
musical instruments. Automatic classification of musical instruments is made by extracting
relevant features from the audio samples, afterward classification algorithm is used (using these
extracted features) to identify into which of a set of classes, the sound sample is possible to fit.
The aim of this paper is to demonstrate the viability of soft set for audio signal classification. A
dataset of 104 (single monophonic notes) pieces of Traditional Pakistani musical instruments
were designed. Feature extraction is done using two feature sets namely perception based and
mel-frequency cepstral coefficients (MFCCs). In a while, two different classification techniques
are applied for classification task, which are soft set (comparison table) and fuzzy soft set
(similarity measurement). Experimental results show that both classifiers can perform well on
numerical data. However, soft set achieved accuracy up to 94.26% with best generated dataset.
Consequently, these promising results provide new possibilities for soft set in classifying musical
instrument sounds. Based on the analysis of the results, this study offers a new view on
automatic instrument classification.
Keywords: Traditional Pakistani Musical Instruments Sounds, Classification, Soft Set, Fuzzy Soft
Set.
1. INTRODUCTION
Classification is one of the most important and significant machine learning areas, capable of
handling musical data effectively and efficiently, once the dataset is categorized into different
families. In general, the need for musical data analysis arises in different contexts, which leads to
many practical applications such as effective data arrangement, music transcription, indexing and
comprehensive digital libraries [1].
Generally, audio signal classification can be cast into musical genre, musical instruments,
composer, artist, music segmentation and so on. Thus, musical instruments classification is not
only subject matter for musicologists but also become an important research field for Music
Information Retrieval (MIR) [2]. Much of existing research focuses on the achievable classification
accuracy of different machine learning algorithms. The studies have shown that a number of
different algorithms are able to achieve high classification accuracy rate. Such as support vector
machine [3], [4], [5], artificial neural networks [6], [7] and rough set [8].

Even though in the work of Ali & Smith [9], they conducted comprehensive study on eight different
classification algorithms. The author compared different eight classification algorithms with
hundred different datasets. The relative weighted performance measurers showed that there was
no single classifier to solve all classification problems with best performance over the different
experimental setups. This leads that no single method has been found to be superior over all
datasets. Nevertheless, rough set has been successfully applied for musical instruments sounds
data [10], [11], [12]. Therefore, it is interesting to see the application of soft set theory for
classification of musical instruments sounds. Moreover, to observe the viability, efficiency and
effectiveness of soft set, while performing this classification task.
Molodtsov [13] introduced the notions of a soft set as a collection of approximate description of an
object. The initial description of an object has as approximate nature, and there is no need of to
introduce the notions of exact solution. The applications of this theory boom in recent years and
are extended to data analysis [14], [15], decision-making [16], [17], [18] to texture classification
[19].This furnishes motivation from the advancement in soft set to apply classification algorithm
based on notions soft set theory (named as soft set classifier), which was proposed by [19] for
texture classification. This paper focuses to asses the performance of soft set classifier towards
classification of musical instruments. Moreover, the contemporary study expands the usual work
of using western musical instruments to the non-western musical instruments i.e. Traditional
Pakistani musical instruments.
The reminder of paper is organized as follows: Section 2 presents the dataset details. Feature
extraction producer is discussed in Section 3. Section 4 illustrates an overview of classification
algorithms followed by results and discussion in the Section 5. Finally the conclusions are made
in Section 6.
2. DATASET
Musical instruments have universal appeal, richness and soothing tones without language and
regional barriers. The history of Traditional Pakistani musical instruments (TMPI) can be gathered
from various sources such as literature (folk, music), visual representation (painting, sculptures,
and models). Interestingly, most of the Pakistani musical instruments remain still in use. Pakistani
music is represented by a wide variety of forms. It ranges from traditional styles such as qawwali,
sufism and ghazal to more recent shapes that is fusion of traditional music with western music.
Generally, these instruments can be cast into religious, geographic and tribal categories. For
instance: harmonium and tabla comes under the religious category and their purpose is for
spiritual uplifts. Likewise, under the umbrella of geographic category, instruments are: algoza,
dhol, dholok and ektara. Benjo, rubab, gheychak, santoor are under the tribal category. TPMI play
a significant role in Pakistani culture. They are used in wedding ceremonies, in traditional dances
such as bhangra, jhumar and khattak. The primary instrument that defines bhangra and jhumar is
the dhol. Whilst, the instruments for qawwali and sufism are harmonium and tabla [20].
Traditional Pakistani musical instruments play a significant role in Pakistani culture. In order to
perform classification of Traditional Pakistani musical instruments a dataset is designed. Table 1
shows the taxonomy of TMPI based on Hornbostel and Sachs system. Dataset consists of three
families which are string, woodwind, percussion and nine musical instruments. The dataset
consist of 104 pieces of Traditional Pakistani musical instruments from compact disks and via
internet. All pieces were solo music from three instrument families. The sampling rate was 44.1
kHz which was later resample to 22.1 kHz with mono .wav files. Table 1 presents the three
families with their description which is partially based on Sachs-Hornbostel system [21].

Family Instruments
String Benju, Ghaychak, Rubab, Sarood, Tumbi
Woodwind Bainsuri, Harmonium
Percussion Tabla, Dholok
TABLE 1: Description of musical instruments
3. FEATURE EXTRACTION
Feature extraction is the significant phase in musical instrument classification. Depending on the
characteristics of the problem domain; the crucial step in the classification is to identify the
relevant features. Though, ‘feature”, “attribute” or “variable” refers to the aspect of data.
Moreover, feature can be discrete, continuous or nominal. Therefore, usually during data
collection, features are specified or chosen. As, extracted features are an input to the classifier
(machine learning algorithm), therefore it is important to extract the right features which can help
classifier to produce encouraging results.
Many feature schemes have been proposed in the literature for audio classification. It is worth to
state that improving feature extraction process will be probably enhancing performance of the
classification algorithm. As mention earlier, diverse features have been proposed and identified
by different studies; each study either work on individual features or combination of features [7].
In general, designing a classification algorithm for musical instrument, feature extraction
techniques are mostly taken from speech and speaker recognition system as they have proved a
significant role to extract valuable information from the dataset [6], [7]. For this study, Mel–
Frequency Cepstral Coefficient (MFCCs) and perception based are adopted from the work of [6]
for extracting features. On the other hand, these features are reflecting different aspects of signal
spectrum, which are discussed in the following sub-section.
3.1 Perception–Based
The most popular features related to perception based are spectral centroid (brightness), spectral
flux and bandwidth (spectral range), which have popularity across the literature. These spectral
features are computed from the Fast Fourier Transform (FFT) of the segmented signals. Besides,
time domain zero crossing and root mean square are also included which reflects the temporal
properties of the signals.
3.2 Mel-Frequency Cepstral Coefficients
Mel-frequency cepstral coefficients are proven to be useful in an extensive range of classification
tasks such as speech classification [22], speaker identification [23] and musical genre
classification [24]. MFCCs proved to be a successful candidate for classification and recognition
For MFCCs, the steps derived in the work of [6] are adapted. The input signal is first divided into
frames. Afterwards, fast Fourier transform (FFT) is used to get the power spectrum of the each
frame. Finally, the Mel filter bank is generated to scale the frequencies logarithmically. To
conclude, the mean and standard deviations has been then calculated for each of the feature
vectors. Table 2 provides the depiction of features utilized

No. FEATURES
1 Zero Crossings
2-3 Zero Crossing Rate(Mean and Standard deviation)
4-5 Root Mean Square(Mean and Standard deviation)
6-7 Spectral Centriod(Mean and Standard deviation)
8-9 Bandwidth(Mean and Standard deviation)
10-11 Spectral Flux(Mean and Standard deviation)
12-37 MFCC (Mean and Standard deviation of the first 13 values)
TABLE 2: Features Descriptions
Prior to the classification stage, vector normalization was done to make sure that data must lie
between the ranges of [ ]1,0 .
4. CLASSIFICATION ALGORITHMS
As it was mentioned in Section 1, automatic classification of musical audio data can be performed
in many ways. However, in the described research, two algorithms have been used. The reason
for considering soft set is to see the viability of this mathematical tool, even though rough set has
been successfully applied to audio signal classification. Moreover, soft set theory is
straightforward, simple and it allows reducing important problems to well-known Boolean ones
using model assumptions. On the other hand, Maji et al [16] proposed an algorithm for decision
making problems. This algorithm has similarity with soft set classifier proposed by Mushrif et al,
[19] I in particular for texture classification.
While in case of fuzzy soft set, this algorithm has similarity with the work of Mushrif et al, [19],
where building the model is same (training set), however for the classification phase which
involves construction of comparison table is replaced by the similarity measurement function.
Next subsection provides an overview of soft set and fuzzy soft set.
4.1 Soft Set
In 1999, Molodtsov [13] introduced the notion of a soft set as collections of approximate
descriptions of an object. This initial description of the object has an approximate nature, and
hence there is no need to introduce the notions of exact solution. The absence of restrictions on
the approximate description in soft set makes this theory suitable and easily applicable in real
world problems. Soft set theory is a newly emerging mathematical tool to deal with uncertain
problems. The applications of this theory boom in recent years and are extended to data analysis
[14], [15], decision-making [16], [17], [18], classification [19].
Maji, Roy & Biswas [16] presented some new definitions on soft set in decision making problems.
Afterwards, Mushrif [19] offered a novel method for classification of natural texture using the
notions of soft set theory. All features from the natural textures were consists of real numbers. For
the dataset, 25 texture classes with 14 textures features were used. Out of 49 images, 14 images
were randomly selected for training and 35 remaining for testing. The proposed method
successfully classifies natural textures. In this section, soft set classifier has been explained in
details.
The soft set classifier learns by calculating the average value of each attributes (features) from
the entire objects. In other words, an object in the universe represents data which is derived
from the same class label. Afterwards, to classify the test data, first to construct a comparison
table as designed in the case of decision making problem.

For calculating cluster center vector, the following expression is used [19].
∑
=
=
N
i
wiE
N
wE
1
1 ( )1
where wE is mean of feature vectors in the same class label. Moreover, DW × is table in which
element of each table is wdg , where Ww ,..,2,1= and Dd ,..,2,1= . In this way, a row wdg is a
cluster center vector for every class w having D features.
Similarly, for calculating element wdp , the following expression is used [19].
)max(
1
wdg
Efdwdg
wdp
−
−=
( )2
where Ww ,..,2,1= and Dd ,..2,1= .
The comparison table is constructed (for more details please refer to [19]). Afterwards, the
score S is computed. The score vector S is a vector containing largest element in S . The score
vector expression is as follows [19].
( )



=
= SW
w
w
1
maxarg ( )3
4.2 Fuzzy Soft Set
In 1965, Zadeh [25] introduced the concept of fuzzy set, where he stated that fuzzy set is a kind
of soft set. Let A be a fuzzy set and Aµ be a member function where Aµ be a mapping of
U into[ ]1,0 . Later, fuzzy set has been studied by Roy et al. [27] where this general concept has
been combined with soft set, named as fuzzy soft set. The results from fuzzy soft set further
expand the scope of applications of soft set theory.
Interestingly, Maji et al [16] proposed an algorithm for decision making problems. This algorithm
has similarity with soft set classifier proposed by Mushrif et al [19] I in particular for texture
Soft Set Classifier (SSC)
Training phase
1. Given N samples obtained from the data class w .
2. Calculate the cluster center vector NiwiE ,..,2,1, = .
3. Obtain soft set model ( )EF, which is DW × .
4. Repeat the process for all W classes.
Classification phase
1. Obtain the unknown class data.
2. Obtain a soft set ( )AF, which have elements .wdp
3. Compute comparison table of soft set ( )AF,
4. Compute the score vector S .
5. Assign the unknown data to class w if Sw max=

classification. In addition, the concept of measuring similarity between two soft set has been
studied by [28], [29] and [30].
4.2.1 Similarity Between Two Soft Sets
Measuring similarity between two entities is a key step for several data mining tasks such as
classification and clustering. The similarity measure calculates the extent to which different image
or signal patterns are alike [30]. Later, Majumdar & Samanta [28] defined and studied the
generalised fuzzy soft sets where the degree is attached with the parameterization of fuzzy sets
while defining a fuzzy soft set and provides similarity measured of soft set and fuzzy soft set. Next
sub-section provides the preliminaries for general fuzzy soft sets
a) Preliminaries
In this sub-section, similarity between the two general fuzzy soft sets are explained by Majumdar
& Samanta [28]. Let { }nxxxU ,....,2,1= be the universal set of elements and { }meeeE ,....,, 21=
be the universal set of parameters. Let δρ GandF be two general fuzzy soft sets over the
parameterized family universe ( )EU, . So
( ) ( )( ){ }miieieFF ,...,2,1,, == ρρ and ( ) ( )( ){ }miieieGG ,...,2,1,, == δδ .
Therefore, ( ){ } ( ){ }miemiieFF ,..,2,1,iGG
~
and,..,2,1,
~
==== are two families of fuzzy soft sets. As
a result, similarity between GF
~
and
~
is determined and it is denoted by ( )GFM
~
,
~
. Next the similarity
between the two fuzzy soft sets ρ and δ is found and is denoted by ( )δρ,m . Later, the similarity
between the two fuzzy soft sets ρF and δG is denoted as ( ) ( )GFMGFS
~
,
~
, =δρ and 1== δρ .
Now the formula for similarity measure is
( ) ( )
( )∑
=
+
∑ = −
−==
n
j
ijGijF
n
j ijGijF
GFiMGFS
1
~~
1
~~
1
~
,
~
, δρ
( )4
where F and G are two families of fuzzy soft sets.
5. RESULTS AND DISCUSSION
Sample data from compact disks and via internet was set up for the different experiments. Seven
datasets were utilized. Table 3 sums up the characteristics of different experimental sets carried
out during dataset formations. In the first experimental setup, the parameter audio length
(duration) will be evaluated to the performance of classification algorithm. Since, there is lack of
uniform approach for audio length of sounds. Therefore, this study investigates three audio
lengths which are 10 seconds, 20 seconds and 30 seconds. Likewise, this is done in order to
generate diverse datasets and to examine whether this identified parameter plays major role in
determining the classification results [6].
Afterwards, the second set focused on finding the optimal frame size (window analyzes). For
feature extraction it is assumed that only short-term audio streams are present. For classifying
longer audio streams, segmentation is necessary. Each of the audio samples is divided into
frames [31]. In addition, the reason to look into the frame size 256 samples and 1024 samples,
these two frame sizes have been commonly used, when it comes to expands the usual work of
using western musical instruments to non-western musical instruments such as in the work of [6],
[32]. Moreover, for analyzing features of audio and speech algorithms, they are computed on
frame basis. By doing so, the amount of data to be processed may possibly reduce.

The remaining of the datasets dealt with finding the best starting point of audio files by providing
the overall better classification results in terms of accuracy rate. The reason for taking this
parameter into consideration is to look into the problem of locating the beginning of the sound.
The purpose is observed and to identify whether noise is there at the beginning of the sound or
not [8]. Therefore; the starting points of each dataset were altered to 0 second, 0.05 seconds and
0.3 seconds. The small size of the starting points are considered since, few of the original sounds
in the datasets have shorter length (less then 1 second).
NO. PARAMETERS DATASET VALUES
1 Audio Length
(Duration)
L1 0.1 to 10 Seconds
2 Frame Size
(Segmentation)
FS1 256 Samples
FS2 1024 Samples
3 Starting Point Of
Audio File
SP1 0 Second
SP2 0.05 Second
SP3 0.3 Second
TABLE 3: Experimental setup designed for study
The experiments were performed on two different models: comparison table and similarity
measure of two soft sets. For the comparisons of both classification algorithms, performance
measurement was classification accuracy. Table 4 shows the classification accuracy. As
mentioned earlier, the dataset consists of three families with nine musical instruments. Feature
extraction in these experiments are implemented in an overlapped analysis window of 256
samples at 22050Hz, calculated the means and standard deviations of all 37 attributes. After
feature extraction, we utilize soft set classifier and fuzzy soft set classifier on those extracted
features.
Comparison Table
(SSC)
Similarity Measure
(FSSC)
Data
Distribution
60:40 70:30 60:40 70:30
DatasetL1 77.91 86.84 83.97 84.20
DatasetL2 94.26 92.27 82.44 81.87
DatasetL3 83.45 85.98 82.15 81.90
DatasetFS2 74.29 69.83 82.03 82.58
DatasetSP1 73.75 81.22 82.47 82.41
DatasetSP2 81.19 88.23 82.63 82.05
DatasetSP3 81.13 88.34 81.80 81.50
TABLE 4: Percentage of correct classification for all datasets
Figure 1 presents the performance comparison of classification accuracy of different datasets
with SSC and FSSC. In case of SSC, the classification accuracy gradually increases when the
modified starting points are used, instead of zero. The highest achievement occurred in the
Dataset L2 with accuracy 94.26%, while, the lowest achievement 73.75% occurred in the
DatasetSP1. In case of FSSC, the highest classification hit rate was 83.97%, while the lowest
achievement was 81.80%.

Though, the results for FSSC are about 1.64% better on mean average with data partition 60:40.
Some what the results from SSC are little weaker then of that sophisticated method (FSSC).
However, the simplicity of SSC makes it appealing and can be pretty helpful in classifying musical
instruments sounds.
Data Distribution(60:40)
60%
65%
70%
75%
80%
85%
90%
95%
SSC
Classifier
PercentageofCorrectClassification
Dataset L1
Dataset L2
Dataset L3
DatasetFS2
Dataset SP1
Dataset SP2
Dataset SP3
Data Distribution (60:40)
60%
65%
70%
75%
80%
85%
90%
95%
FSSC
Classifier
PercentageofCorrectClassification
Dataset L1
Dataset L2
Dataset L3
DatasetFS2
Dataset SP1
Dataset SP2
Dataset SP3
(a) (b)
FIGURE 1: Classification performance of both classifiers (a) and (b)
with data distribution 60:40
While, Figure 2 shows the graphical comparison with data fraction 70:30 for both classifiers. It
can be seen that classification range is 69.83% to 92.27% for SSC. Whilst, for FSSC the range
from 81.50% to 84.20%. Interestingly, the results for SSC are about 2.32% better on mean
average. And the hypothesis that larger the training set make sure classifier to gain knowledge of
data, this graph plots the evidence of that myth.
60%
65%
70%
75%
80%
85%
90%
95%
SSC
Classifier
PercentageofCorrect
Classification
Dataset L1
Dataset L2
Dataset L3
DatasetFS2
Dataset SP1
Dataset SP2
Dataset SP3
60%
65%
70%
75%
80%
85%
90%
95%
FSSC
Classifier
PercentageofCorrect
Classification
Dataset L1
Dataset L2
Dataset L3
DatasetFS2
Dataset SP1
Dataset SP2
Dataset SP3
(a) (b)
FIGURE 2: Classification performance of both classifiers (a) and (b)
with data distribution 70:30
6. CONCLUSIONS
Classification is a key intervention for initiating process of labeling monophonic sounds. Most of
the studies conducted to find differences and similarities with features schemes, afterwards
evaluate and compare with various classifiers. At the same time as, the right extracted features
can simplify the design of a classification algorithm, whereas lousy features can hardly be
compensated with any of the classification algorithm. Therefore, it is likely to state that
appropriate parameterization of audio sounds allows efficacious classification of musical
instruments. Additionally, the choice of classification algorithm makes a quite sophisticated
approach towards classification task. However, despite the massive research has been carried

out on this filed, studies have mainly dealt with western musical instruments and few works can
be found on non-western musical instruments. However, individual musical instrument has
different behavior and so does Pakistani musical instruments. Therefore, this study incorporates
soft set theory for classification of Traditional Pakistani musical instrument and investigates the
impact of three factors, which are length of audio file, frame size and starting point towards
classification performance of soft set classifier.
A small–scale database of instrumental music was built, in line with popular taxonomy known as
Sachs & Hornbostel, categorizing instruments into three families: string, woodwind and
percussion. With experiments on 37 features and two classifiers, all music clips were
automatically classified by instrument families. In the light of obtained results from the both
classifiers, provides a relatively new picture about instrument classification. The experiments
testified that FSSC suits well for automatic instruments classification, while applied SSC best to it.
In addition, based on the obtained results, this study offers a new view on automatic instrument
classification. Thus, the soft set can be considered to be employed in musical instrument
classification problem. Nevertheless, it is experimentally demonstrated that this classification
algorithm yields better accuracy when compared with fuzzy soft set. These results have further
expanded the scope of soft set for real world problems. Future work will consider the selection of
the most relevant features in discriminating between different instruments and extension of the
present feature set.
ACKNOWLEDGMENT
The authors wish to thank Ministry of Higher Education (MOHE) and Universiti Tun Hussein Onn
Malaysia for supporting this research under the Fundamental Research Grant Scheme (FRGS).
REFERENCES
[1] Kotsiantis, S.B. Supervised machine learning: A review of classification techniques. Informatica 31,
pages 249-268. (2007)
[2] McKay, C. & Fujinaga, I. Automatic music classification and the importance of instrument
identification. In: Proceedings of the Conference on Interdisciplinary Musicology (CIM05). Montreal,
Canada (2005).
[3] Mierswa, I., & Morik, K. Automatic Feature Extraction for Classifying Audio Data. Journal of Machine
Learning: Volume 58 Issue 2-3 (2005).
[4] Essid, S., Richard, G., & David, B. Musical Instruments Recognition by pair wise classification
strategies. Audio, Speech and Language processing, IEEE Transaction on Speech and Audio
Processing, 14(2), 1401-1412 (2005).
[5] Liu, J. & Xie, L. Comparison of Performance in Automatic Classification between Chinese and
Western Musical Instruments. In: Proceeding of WASE International Conference on Information
Engineering. Beidaihe,Hebei, (2010)
[6] Senan, N., Herawan, T., Mokji, M.M., Nawi, N.M., & Ibrahim, R. The Ideal Data Representation for
Feature Extraction of Traditional Malay Musical Instrument Sounds Classification. Advanced
intelligent computing theories and applications: Lecture Notes in Computer Science, 2010, Volume
6215/2010,pages 345-353
[7] Ding, Q., Zhang, N. Classification of Recorded Musical Instruments Sounds Based on Neural
Networks. In: IEEE Symposium on Computational Intelligence in Image and Signal Processing. Pp
157-162. Honolulu, HI (2007).
[8] Wiezorkowska, A. Towards Musical Data Classification via Wavelet Analysis. In Proceeding of the
12th International Symposium on Foundations of Intelligent Systems, Springer–Verlag London Uk.
(2000)

[9] Ali, S. & Smith, K.A. On learning algorithms selection for classification. Applied Soft Computing 6,
pages119-138 (2006).
[10] Wieczorkowska, A. Rough Sets as a tool for audio signal classification. Lecture Notes in Computer
Science, Volume 1609/1999, 367-375, (1999) DOI: 10.1007/BFb0095123.
[11] Wieczorkowska, A. & CzyŜewski, A. Rough Set Based Automatic Classification of Musical Instrument
Sounds. In: International Workshop on Rough Sets in Knowledge Discovery and Soft Computing. pp.
298-309, (2003).
[12] Senan, N., Ibrahim, R, Nawi, N. M., Yanto, I. T. R.& Herawan, T. Feature Selection for Traditional
Malay musical instruments sounds classification using rough set. .Journal of Computing Volume 3,
Issue 2, (2011).
[13] Molodtsov, D. Soft set theory –first results. Computer and mathematics with applications. Pp19-31.
(1999).
[14] Zou, Y.& Xiao, Z. Data analysis approaches of soft sets under incomplete information. Knowledge-
Based System 21, 2128-2137 (2008).
[15] Herawan, T. Deris, M. M. A Direct Proof of Every Rough Set is a Soft Set. In Third Asia International
Conference on Modeling & Simulation, AMS '09 pages119–124, Bali (2009).
[16] Maji, P. K., Roy, A. R., Biswas, R. An application of soft sets in decision making problem. Computers
and mathematics with applications. pp 1077-1083. (2002).
[17] Roy, A. R. & Maji, P. A fuzzy soft set theoretic approach to decision making problems.Journal of
Computational and Applied Mathematics. Volume 203(2), pp.540-542 (2007).
[18] Jiang, Y., Tang, Y. & Chen, Q. An adjustable approach to intuitionistic fuzzy soft sets based decision
making. Applied Mathematics Modeling. Volume 35 (2), pp. 824-836 (2011).
[19] Mushrif, M., M, Sengupta, S., Ray, A.K. Texture Classification Using a Novel Soft Set Theory Based
Classification Algorithm. In: LNCS, vol.3851, pp.246-254.Springer, Heidelber (2006).
[20] Ziagham, N. (2003). Forms of Pakistan Music. Retrieved March, 9, 2011 from
http://www.maighmalhaar.com/IntroductionPage1.html
[21] Marshall, P. (2011). Sachs Hornbostel System of instrument classification. Retrieved December 8,
2010 from website http://www.drumdojo.com/sachshornbostel.htm
[22] Davis, S.B. and Mermelstein, P. Comparison of parametric representations for monosyllabic word
recognition in continuously spoken sentences. IEEE Transaction on Acoustics, Speech and Signal
Processing, Volume number 28, issue 4:357-366 (1980).
[23] Sarimollaoglu, M., Dagtas, S., Iqbal, K., Bayrak, C. A text–independent speaker identification system
using probabilistic neural networks. International conference on computing, communications and
control technologies (CCCT), Austin, USA, pages 407-411(2004).
[24] Tzanetakis, G., Cook, P. Musical genre classification of audio signals. IEEE Transaction on Speech
and Audio Processing: Vol. 10, No. 5 (2002).
[25] Zadeh, L. Fuzzy sets. Inform. Control. Volume number 8: 338-353. (1965).
[26] Maji, P. K., Biswas, R. & Roy, A. Fuzzy soft sets. Journal of Fuzzy Mathematics 9(3), pp.589-602
(2001).
[27] Roy, A. R. & Maji, P. A fuzzy soft set theoretic approach to decision making problems. Journal of
Computational and Applied Mathematics. Volume 203(2), pp.540-542 (2007).
[28] Majumdar, P. & Samantra,S.K.(2010). Generalized fuzzy soft set. Journal of Computational and
Applied Mathematics. Pp. 1279-1286.

[29] Kharal, A. Distance and similarity measurement for soft set. New Math.. & Nat. Com.(NMNC). Volume
(6), pp. 321-334 (2010).
[30] Handaga, B. & Deris M. M. Similarity approach on fuzzy soft set based numerical data classification.
Communications in Computer and Information Science. Volume 180(6), pp. 575-589 (2011).
[31] Kumari, R. S. S., Sugumar, D. & Sadasivam,V. Audio signal classification based on optimal wavelet
and support vector machine. Proceeding of International conference on computational intelligence
and multimedia applications. Volume 2, pp: 544 – 548. (2007)., ISBN:0-7695-3050-8
[32] Gunasekaran, S. & Revathy, K. Fractal dimension analysis of audio signals for Indian musical
instrument recognition. International Conference on Audio, Language and Image Processing,
Shanghai, ISBN: 978-1-4244-1723-0, pp. 257-261 (2008).

Performance Comparison of Musical Instrument Family Classification Using Soft Set

More Related Content

Similar to Performance Comparison of Musical Instrument Family Classification Using Soft Set (20)

More from Waqas Tariq (20)

Recently uploaded (20)

Performance Comparison of Musical Instrument Family Classification Using Soft Set