Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
, pp.
55
8
~
566
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v11
i
1
.
pp
558
-
566
558
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
A
n
ew
p
arallel
b
at
algori
thm f
or
mu
sica
l
note r
ecog
niti
on
An
s
am N
az
ar Youni
s, Faw
z
iya M
ahmo
od
Ra
m
o
Depa
rtment
o
f
C
om
pute
r
Scie
n
ce
,
Col
le
ge
of
Com
pute
r
Scie
n
ce
and
Math
ematic
s,
Univer
sit
y
of M
osul,
Ira
q
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n
18
, 2
0
20
Re
vised
Jun
13
,
20
20
Accepte
d
Aug
5
, 2
0
20
Mus
ic
is
a
unive
rsal
l
anguage
tha
t
do
es
not
req
uir
e
an
int
erp
r
et
e
r
,
where
feeli
ngs
a
nd
sensiti
vities
a
re
unit
ed
,
reg
ard
le
ss
of
the
diffe
r
ent
peoples
and
la
nguag
es,
The
proposed
s
y
stem
consists
of
two
m
ai
n
stage
s:
the
proc
ess
of
ext
racti
ng
important
prope
rties
using
the
li
near
discri
m
ina
ti
o
n
ana
l
y
sis
(LDA)
Thi
s
step
is
ca
r
rie
d
out
aft
er
t
he
ini
tial
treatment
proc
ess
using
var
ious
proc
edur
es
to
remove
m
usica
l
li
nes,
The
se
cond
stag
e
desc
ribe
s
th
e
r
e
cogni
ti
on
proc
ess
using
the
ba
t
al
gorit
hm
,
whic
h
is
one
of
the
m
etahe
urist
i
c
a
lgori
thms
after
m
odif
y
ing
th
e
ba
t
a
lgori
thm
to
ob
t
ai
n
bet
t
er
discr
imina
ti
ng
resul
ts.
The
proposed
s
y
st
e
m
was
supported
b
y
par
a
llel
implementa
t
ion
using
the
(de
veloped
bat
a
lgorit
hm
DB
A),
which
inc
rea
sed
the
spee
d
of
implementation
sig
nifi
c
ant
l
y
.
Th
e
m
et
hod
was
appl
ie
d
to
125
0
diffe
ren
t
images
of
m
usica
l
note
s.
The
propose
d
sy
st
em
was
implemente
d
using
MA
TL
AB
R2016a,
W
ork
was
done
on
a
W
indows10
Pr
oce
ss
or
OS
(Inte
l
®
Cor
e T
M i
5
-
7200U CPU
@ 2.
50GH
Z
2
.
70GH
Z)
compu
te
r.
Ke
yw
or
d
s
:
Ba
t
a
lgo
rit
hm
Linear
d
isc
rim
i
nate analy
sis
Me
ta
heu
risti
c
Musica
l
n
otes
Parall
el
p
r
oces
sing
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
An
sam
N
azar
You
nis,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce,
Un
i
ver
sit
y
of
Mosu
l
,
Mosu
l,
Iraq
.
Em
a
il
:
AN
SAM
201733@ya
hoo.
c
om
1.
INTROD
U
CTION
Music
is
a
kind
of
w
or
l
d
art
that
deals
with
the
com
po
si
ti
on
,
r
hythm
a
nd
distrib
utio
n
of
va
rio
us
m
el
od
ie
s.
Mu
sic
is
a
sci
ence
that
exp
lo
r
es
m
el
od
ie
s'
or
i
gin
s
an
d
pri
nciples
in
te
r
m
s
of
har
m
ony
and
disti
nction
a
la
ngua
ge
that
in
cl
ud
es
e
xpress
ive
and
c
omm
un
ic
at
ive
c
ompone
nts.
It
is
an
art
com
po
se
d
over
certai
n
pe
rio
ds
of
ti
m
e
with
so
un
ds
an
d
sil
ence
[
1].
As
m
us
ic
is
a
gen
eral
process
kn
own
to
th
os
e
inter
est
ed
in
the
m
us
ic
indu
stry,
the
c
om
pu
te
r
had
to
be
us
e
d
to
s
erv
e
the
m
us
ic
Seve
ral
c
ompu
te
r
sci
e
ntist
s
ha
ve
dev
el
op
e
d
pro
gr
am
s,
syst
e
m
s
and
al
gorit
hm
s
to
reco
gniz
e
and
disc
ov
e
r
va
rio
us
m
us
ic
al
no
te
s
an
d
app
li
cat
io
ns
for
m
us
ic
[2
]
.
The
patte
r
n
di
sti
nction
refe
r
s
to
the
funct
ion
of
putt
in
g
a
par
ti
cular
obj
ect
in
the
appr
opriat
e
disti
nction
bas
ed
on
that
ob
je
ct
'
s
par
a
m
et
ers,
Us
ually
done
autom
at
ic
a
lly
with
the
ai
d
of
a
com
pu
te
r
after
the
ob
j
ect
'
s
prop
e
rtie
s
hav
e
been
e
xtracted
.
T
he
te
rm
patte
rn
rec
ogniti
on
incl
udes
se
ve
ral
ot
her
im
p
or
ta
nt
te
rm
s
su
ch
as re
cogniti
on, d
es
cripti
on, classi
ficat
ion
, a
nd
groupin
g
of p
at
te
rn
s
. F
or exam
ple, a
patte
rn
m
ay
b
e
a finger
pri
nt i
m
age,
a
hand
w
ritt
en
w
ord, a
hum
an
face,
or
a sp
eec
h
si
gn
al
[3].
In
t
he
ye
ar
(20
11),
t
he
tw
o
re
searche
rs
(Jo
y
ce
Oo
i
B
oon
Ee
an
d
Ala
n
WC
Ta
n)
us
e
box
boundi
ng
and
te
m
plate
m
at
ching
[
4]
.
In
(
2014
)
the
tw
o
resea
rch
e
rs
(Yo
pp
y
Sazaki
a
nd
Ros
da
Ayu
ni
)
use
the
Mi
nim
u
m
Sp
a
nn
i
ng
Tree
al
gorithm
and
the
E
uclidean
distance
[5
]
.
In
(20
17)
the
res
earche
rs
(Jan
Haj
i
ˇc
and
Pa
vel
Peci
na)
s
ug
gest
ed
t
o
us
e
c
onvolut
ion
al
ne
ur
al
n
e
twork
CN
N
for
disco
ver
the
head
'
s
no
te
s
ap
plied
to
va
rio
us
pat
te
rn
s
of
ha
nd
wr
it
in
g
sty
le
s
[6
]
.
I
n
(20
18)
A
ns
am
Nizar
an
d
Fawzia
Ma
hm
ou
d
pr
opos
e
d
a
m
et
ho
d
for
r
ecognize
m
us
ic
al
no
te
s
u
sin
g
the
al
gorithm
(LDA
li
nea
r
d
i
scrim
inate
analy
sis)
an
d
an
e
quat
ion
of sim
il
arit
y sc
al
e ind
e
x
S
SIM
[
7].
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
par
allel
b
at
algorit
hm f
or
music
al
no
t
e reco
gnit
ion
(
Ansam
Naz
ar
Youn
is)
559
The
goal
of
t
he
resea
rch
is
to
dev
el
op
a
nd
cr
eat
e
an
i
ntell
igent
com
pu
te
r
s
yst
e
m
to
recog
nize
im
ages
of
m
us
ic
al
no
t
es
an
d
to
cl
assi
fy
their
form
i
n
order
t
o
help
sp
r
ead
m
us
ic
al
cultur
e
t
hro
ugh
dif
fer
e
nt
se
gm
ents
of
s
ociet
y
and
to
ensure
sim
pl
ic
it
y,
transp
are
ncy,
accu
r
acy
and
sp
ee
d
in
com
m
un
icati
ng
with
us
e
rs
at
the
sam
e
tim
e.
The
resea
rch
furthe
r
aim
s
t
o
te
ach
the
c
om
pu
te
r
and
a
pply
know
le
dge
to
it
in
the
fiel
d
of
m
us
ic
edu
cat
ion
a
nd
to
set
up
a
s
upportiv
e
syst
e
m
fo
r
m
us
ic
exp
erts
to
le
arn
m
us
ic
al
no
te
s
an
d
pro
vide
a syst
e
m
f
or th
os
e
wa
nting t
o l
earn
t
he
m
eaning
of m
us
ic
al
scale
sy
m
bo
ls.
2.
COMM
ON N
OTATIO
N
The
m
us
ic
is
wr
it
te
n
in
le
tt
ers
that
are
dif
f
eren
t
f
ro
m
the
al
ph
a
bet,
Mu
sic
no
te
s
are
a
tim
e
-
bo
und
sy
m
bo
l,
de
pe
ndin
g
on
t
he
s
ound
of
that
le
tt
er,
the
dur
at
ion
of
each
le
tt
er
or
m
us
ic
al
sy
m
bo
l
on
the
sta
ff
[1,
8]
no
te
F
ig
ur
e
1
.
Figure
1. Staf
f l
ines
2.1.
The s
t
aff
Five
ho
rizo
ntal
li
nes,
inclu
din
g
fou
r
eq
ual
sp
aces,
a
re
paral
le
l
and
eq
ual
in
le
ng
t
h.
S
om
e
m
us
ic
al
no
te
s
a
re
po
sit
ion
e
d
on
or
at
a
distance
bet
ween
li
nes.
A
dd
it
io
nal
sho
rt
li
nes
m
ay
be
at
ta
ched
to
t
he
le
dg
e
r
li
nes
to
represe
nt a
ver
y
high
or v
e
ry lo
w
m
a
rk that i
s
no
t i
n st
an
dard c
ondi
ti
on
[8
]
.
2.2.
Mus
ic
al n
ot
es
It
is
c
om
po
sed
of
se
ven
sim
ple
m
us
ic
al
signs
of
diff
e
re
nt
s
ound
a
nd
the
e
igh
th
s
ound
is
a
re
petit
ion
of
t
he
first
no
t
e:
du
-
ri
-
mi
-
fa
-
la
-
si
the
n
du
ag
ai
n.
T
he
nam
e
of
t
he
m
us
ic
note
is
dete
rm
in
ed
by
the
key
us
e
d
at
the b
egi
nn
i
ng
of
the m
us
ic
sc
al
e, Th
e note fo
rm
i
s an
o
val h
ead th
at
can b
e com
plete
and e
m
pty, and
the h
ead
m
ay
be
at
ta
ched
to
the
ste
m
It
m
ay
a
lso
be
li
nk
e
d
to
on
e
or
m
or
e
flags
(f
la
g)
T
his
de
pe
nds
on
the
num
ber
of
strikes
an
d
the
le
ng
th
of
the
no
te
[
2].
Ma
ny
m
us
ic
al
no
te
s
aff
ect
the
m
us
ic
al
ton
e
in
on
e
way
or
an
oth
e
r
,
and
eac
h
ty
pe
of
these
sym
bols
is
wr
it
te
n
diff
e
ren
tl
y
from
the
oth
e
r
an
d
put
on
the
m
us
ic
al
scal
e
in
diff
ere
nt
po
sit
io
ns
a
nd
t
hese
sym
bo
ls
a
re:(cle
f,
Acci
de
ntal
sym
bo
ls,
Re
sts,
Tim
e
sy
m
bo
ls,
Breaks
,
Dynam
ic
s
sym
bo
ls
,
Augm
entat
ion
Do
t,
Ti
es,
A
rtic
ulati
on
,
Si
gn
s
Repeat
, E
co
no
m
ic
al
sy
m
bo
ls, K
ey
Si
gn
at
ur
e
, Tr
ia
d)
[
9].
3.
BAT AL
GO
R
ITHM B
A
The
Ba
ts
al
go
rithm
is
on
e
of
the
m
et
aheu
r
ist
ic
al
go
rithm
s
create
d
by
t
he
world
(
Ya
ng)
in
2010,
an
al
gorithm
i
ns
pi
red
by
nat
ur
e
[10].
Ba
se
d
on
the
previ
ou
s
fe
at
ures,
T
he
f
ollo
wing
three
basic
r
ule
s
were
dev
el
op
e
d
f
or
the
al
go
rithm
:
Ba
ts
us
e
distan
ce
sensi
ng
ec
holocat
io
n
a
nd
"
le
arn
"
t
he
diff
e
ren
ce
bet
ween
foo
d,
pr
ey
an
d
bac
kbones
,
In
posit
ion
Xi,
t
he
bat
flie
s
at
velocity
(V
i)
with
the
qmi
n
fr
e
qu
e
nc
y,
with
the
v
a
riable
wav
el
e
ng
t
h
λ
an
d
th
e
lo
ud
ness
L0
to
hunt
f
or
t
he
prey
.
It
ca
n
a
uto
m
atical
ly
adj
us
t
the
wa
ve
le
ngt
h
(or
f
re
qu
e
ncy)
from
it
s
e
m
i
tted
pu
lse
an
d
a
dju
st
t
he
pu
lse
em
issi
on
rate
(0,
1)
acco
r
din
g
to
it
s
pro
xim
ity
to
the
ta
r
get.
Althou
gh
t
he
loudn
e
ss
m
ay
diff
e
r
f
r
om
on
e
t
o
an
oth
e
r
in
m
any
resp
ect
s,
it
is
assum
ed
that
the
l
oudn
ess
is
us
ua
ll
y
lim
it
ed
an
d
th
at
L0
va
ries
f
ro
m
a
la
rg
e
fi
xed
val
ue
t
o
a
m
ini
m
u
m
con
sta
nt
value o
f
Lm
in
[11].
Ba
t Alg
or
it
hm
steps [1
2]
:
1)
Determ
ine the
ta
rg
et
fu
nctio
n so
t
hat it
re
pr
es
ents:
xd
x
X
X
f
T
,.
..,
1
),
(
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
55
8
-
56
6
560
2)
C
onfig
ur
e
the
bat po
pu
la
ti
on:
w
he
re
n
is t
he
m
axi
m
u
m
n
um
ber
o
f bats,
then dete
rm
ines the initi
al
v
el
oc
it
y fo
r
ea
ch
bat
(
Vi)
.
3)
Def
i
ne pu
lse
frequ
e
ncy
(q
i
)
a
t Xi locat
io
n.
4)
Config
ur
e
puls
e rates
(ri
)a
nd l
oudness
rate
(
Li).
5)
Re
peat the
f
ollow
i
ng steps
un
ti
l t
he
co
nd
it
io
n
is m
et
(
getti
ng the
m
axi
m
um
o
f
the
duplica
te
s).
Creat
e n
e
w
s
ol
utions
by adj
ust
ing
fr
e
qu
e
ncy
and up
dating s
peeds a
nd loca
ti
on
s s
olu
ti
ons
(
(
1
)
to
(
3
)
).
a)
Com
par
e the
pulse
fre
qu
e
ncy
value wit
h
a
r
a
ndom
value.
-
Sele
ct
on
e
of t
he best s
olu
ti
ons.
-
Creat
e a local
s
olu
ti
on a
bout t
he best s
olu
ti
on selec
te
d.
b)
Gen
e
rate a
ne
w
s
olu
ti
on t
hro
ugh
t
he ran
do
m
f
li
gh
t of
bats
.
c)
Com
par
e
the
l
oudness
value
with
t
he
r
an
dom
value
an
d
com
par
e
the
value
of
the
ne
w
sit
e
wit
h
the v
al
ue of
th
e o
ld
sit
e.
d)
Accept
ne
w
s
ol
ution
s
.
e)
In
c
rease t
he
(
ri
)pu
lse
value
a
nd d
ec
rease t
he
Li (lou
dn
e
ss
value
).
f)
Arrange t
he ba
t and fi
nd the
be
st x
*
c
urren
t.
6)
Pr
oc
essin
g
t
he results
of the at
ta
ched
t
reatm
e
nt and
presenta
ti
on
.
The
virt
ual
bat
m
ov
em
ent
is
by
si
m
ulati
ng
the
virt
ual
bat
m
ov
e
m
ent
natur
al
ly
.
W
e
ha
ve
three
basic
ru
le
s
w
her
e
Xi
an
d
Vi
are
de
te
rm
ined
in
d
-
dim
ension
al
a
nd
ho
w
to
up
da
te
it
.
The
ne
w
s
olu
ti
o
ns
Xi
t
an
d
velocit
ie
s V
it
a
t t
i
m
e
t are g
iv
en usin
g
t
he
f
ol
lowing e
qu
at
i
on
s
[
13]
:
q
q
q
q
i
)
(
m
i
n
m
a
x
m
i
n
(1)
q
X
X
V
V
i
t
i
t
i
t
i
)
(
*
1
(2)
V
X
X
t
i
t
i
t
i
(3)
β
ϵ
(
0,
1)
is
a
r
andom
vector
ta
ken
f
ro
m
a
re
gu
la
r
distrib
ution.
X*
re
pr
es
e
nts
the
best
c
ur
ren
t
ge
ner
al
locat
ion
(s
olu
ti
on)
wh
ic
h
is
locat
ed
a
fter
c
om
par
ing
al
l
th
e
so
luti
ons
am
ong
al
l
n
bats.
For
the
l
ocal
searc
h
segm
ent.
On
ce
a
so
luti
on
has
been
ide
ntifie
d
am
on
g
the
bes
t
current
s
olu
ti
on
s
,
a
new
so
l
ution
f
or
eac
h
bat
is
create
d
l
ocall
y usin
g ran
dom
walkin
g
a
s in
(
4)
:
t
RL
old
X
ne
w
X
(4)
R
ϵ
(
-
1,
1
)
i
s
a
r
a
n
d
o
m
n
u
m
b
e
r
,
L
t
=
(
L
i
t
)
i
s
t
h
e
m
e
a
n
o
f
l
o
u
d
n
e
s
s
o
f
a
l
l
B
a
t
i
n
t
h
i
s
t
i
m
e
s
t
e
p
[
1
3
]
.
T
h
e
L
i
l
o
u
d
n
e
s
s
a
n
d
t
h
e
p
u
l
s
e
e
m
i
s
s
i
o
n
r
a
t
e
m
u
s
t
b
e
u
p
d
a
t
e
d
w
i
t
h
r
e
p
e
t
i
t
i
o
n
,
a
c
c
o
r
d
i
n
g
t
o
t
h
e
f
o
l
l
o
w
i
n
g
e
q
u
a
t
i
o
n
s
:
t
i
L
t
i
L
1
(5)
t
i
r
t
i
r
)]
e
x
p
(
1
[
0
1
(6)
i
r
t
i
r
t
i
L
0
,
0
as
t
(7)
α
an
d γ a
re c
on
sta
nt v
al
ues.
F
or each:
0<α<1
and γ
>
0.
i
r
t
i
r
t
i
L
0
,
0
a
nd
t
(8)
The
sel
ect
ion
of
par
am
et
ers
requires
s
om
e
exp
e
rim
entat
io
n.
I
niti
al
ly
,
ea
ch
bat
m
us
t
hav
e
dif
fer
e
nt
values
for t
he
l
oudness
rate a
nd the
pulse
e
m
issi
on
r
at
e.
T
his ca
n be ac
hieved b
y
ra
ndom
d
ist
ribu
ti
on
[14].
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
par
allel
b
at
algorit
hm f
or
music
al
no
t
e reco
gnit
ion
(
Ansam
Naz
ar
Youn
is)
561
4.
FEATU
RE E
X
T
R
AC
TI
ON
Feat
ur
e
extract
ion
m
et
ho
ds
a
re
use
d
in
t
he
process
of
dim
ensio
nal
re
duc
ti
on
of
t
he
or
i
gin
al
high
-
dim
ension
al
da
ta
and
are
str
onge
r
than
t
he
m
et
ho
ds
of
c
ha
racter
sel
ect
io
n
[
7,
15]
.
T
he
process
of
e
xtr
act
ing
at
tribu
te
s
re
presents
the
pr
oc
ess
of
c
reati
ng
a
new
set
of
feat
ur
es
t
hat
are
m
or
e
i
m
po
rtant
by
m
ixing
the
ori
gi
nal
fea
tures
i
n
li
nea
r
or
nonli
near
w
ay
s
[16,
17]
.
L
inear
discrim
inate
analy
sis
(L
DA)
is
a
com
m
on
ly
su
pe
r
vised
m
o
dalit
y
(ie
ta
king
cl
assifi
cat
io
n
into
acc
ount
)
And
is
c
omm
on
ly
us
e
d
i
n
c
om
pu
te
r
visio
n,
patte
rn
recog
niti
on
,
m
achine
le
ar
ning
an
d
the
li
ke
,
al
so
know
n
as
Fisher'
s
lin
ear
discrim
in
at
or
.
T
o
the
Brit
ish
sta
ti
sti
cal
wo
rl
d
(Ronal
d
Aylm
er
Fisher),
w
ho
init
ia
ll
y
pr
opose
d
the
a
ppr
oach
a
nd
is
an
eff
ect
ive
m
et
h
od
of
discrim
inati
on
,
since
it
is
a
li
near
m
et
ho
d
of
li
near
da
ta
reducti
on
and
a
cl
assi
ficat
ion
m
et
ho
d
[18].
LDA
at
tem
pts
to
de
fine
a
ne
w
axis
t
o
re
du
ce
i
nter
-
cl
ass
var
ia
ti
on
du
e
to
sessio
n/c
hannel
ef
fects
and
t
o
m
axi
m
iz
e
diff
eren
ces
bet
we
en
cl
asses
[19
]
.
The
li
nea
r
discrim
inati
on
analy
sis
al
go
rithm
aims
at
fin
ding
vecto
rs
that
m
axim
iz
e
the
di
ff
e
ren
ce
bet
w
een
the
dif
fere
nt
cl
asses
(Bet
ween
-
Cl
ass
Scat
te
r
Ma
trix
CB
),
Re
du
ci
ng sam
ple v
a
riabil
it
y wit
hin
eac
h ca
te
go
ry
W
it
hin
-
Cl
ass Scatt
er
Ma
trix (
C
W)
[
18,
19
]
.
5.
PARALL
EL
PRO
CESSI
N
G
Parall
el
pr
oc
es
sing
m
eans
faster
program
execu
ti
on
by
sp
li
tt
ing
the
pro
gra
m
into
m
ulti
p
le
par
ts
an
d
perform
ing
at
the
sam
e
tim
e
as
each
par
t
r
uns
on
a
se
pa
rate
process
or
or
on
a
s
epa
rate
kernel.
T
he
pr
ogram
i
m
ple
m
ented
in
N
pr
ocess
ors
is
ap
pro
xim
a
te
ly
N
fa
ste
r
t
han
the
sam
e
pro
gr
am
on
a
sing
le
process
or
[
20
]
.
Its ch
a
racteri
sti
cs can
b
e
su
m
m
arized as
fo
ll
ow
s
[2
1]
:
-
Uses
m
or
e t
ha
n on
e
pr
ocessi
ng unit
(
CP
Us
)
.
-
D
ivide
the
pro
blem
into
sepa
rate pa
rts ca
n b
e so
l
ved sync
hro
nous
ly
.
-
Divid
e
each
pa
rt into
a se
ries
of
prom
pts.
-
The
i
ns
tr
uctions from
each pa
rt are
ca
rr
ie
d o
ut sim
ultaneousl
y on dif
fer
e
nt
pro
ces
sin
g un
i
ts.
-
All im
ple
m
entat
ion
is
unde
r
c
on
t
ro
l.
5.1.
Multic
ore p
r
oc
essors
This
ty
pe
of
proces
sors
is
al
so
cal
le
d
the
Chip
Mult
ipr
oc
esso
r,
com
bini
ng
tw
o
or
m
or
e
pro
cess
ors
(
c
a
l
l
e
d
C
o
r
e
s
)
o
n
o
n
e
p
i
e
c
e
o
f
s
i
l
i
c
o
n
(
c
a
l
l
e
d
t
h
e
m
o
l
d
)
.
E
a
c
h
k
e
r
n
e
l
c
o
n
s
i
s
t
s
o
f
a
l
l
c
o
m
p
o
n
e
n
t
s
o
f
a
n
i
n
d
e
p
e
n
d
e
n
t
p
r
o
c
e
s
s
o
r
,
s
u
c
h
a
s
R
e
g
i
s
t
e
r
s
,
A
L
U
,
C
o
n
t
r
o
l
U
n
i
t
,
A
s
w
e
l
l
a
s
d
a
t
a
s
t
o
r
a
g
e
(
L
1
d
a
t
a
c
a
c
h
e
s
)
.
I
n
add
it
ion
t
o
m
ulti
ple
cor
es
, c
on
te
m
po
ra
r
y m
ulti
-
core chip
s also
inc
lud
e
L2
cache
and L
3
cac
he
i
n
s
om
e cases
[
22
]
.
6.
DESIG
N
TH
E PRO
POSE
D
S
YS
TE
M
The
pr
opos
e
d
PBM
RS
(
pa
rall
el
bat
m
us
ic
al
no
te
s
rec
ognit
ion
syst
em
)
sy
stem
was
desig
ned
t
o
buil
d
a
s
m
art
co
m
pute
r
pro
gr
am
to
recogn
iz
e
m
us
ic
al
no
te
s
an
d
t
o
disp
la
y
us
e
fu
l
in
form
at
io
n
ab
out
each
m
us
ic
a
l
scor
e
. T
he pr
opose
d sy
ste
m
g
oes
t
hroug
h
tw
o ph
a
ses: t
raini
ng and test
in
g.
6.1.
Design
of
th
e
tr
aining
phase
Buil
ding
datab
ase
of
im
ages:
In
t
his
sta
ge
,
r
ead
the
10
00
m
us
ic
al
no
te
s
im
ages
of
the
tr
ai
nin
g
phase
of 10
0 diff
e
rent
m
us
ic
labels.
Pr
e
processin
g
:
In
it
ia
l t
reatm
en
t i
nclud
e
s the
foll
ow
i
ng steps:
a.
Re
siz
e each im
age a
nd unify t
he
siz
e
of all
im
ages
b.
conve
rting i
m
a
ges
from
co
lor
i
m
ages (
RGB
) t
o
gray
c
olo
r
s (Gray)
c.
Convertin
g
im
ages
f
r
o
m
Gr
ay
color
t
o
Bi
na
ry
i
m
ages
(
bla
ck
an
d
wh
it
e)
,
in
orde
r
to
reduce
the
num
ber
of col
or
s
, it i
s
easy
to dist
inguish
the c
olor
of the m
ark
fro
m
the b
ac
kgr
ound c
olor.
d.
Delet
e
the
wh
it
e
fr
am
e
su
rroun
ding
the
m
us
ic
al
m
ark
an
d
the
ba
ckgr
ound,
the
reb
y
rem
ov
in
g
the im
po
rtant i
nfor
m
at
ion
to
focu
s
on
ly
on th
e sh
a
pe of
the
m
ark
.
e.
Re
siz
e the im
a
ge
a
fter the
pro
cess of
delet
in
g
the
fram
e.
f.
Feat
ur
e
e
xtrac
ti
on
:
fin
d
t
he
m
at
rix
of
wei
gh
ts
for
a
m
us
ic
al
m
ark
m
o
del
as
a
r
esult
of
t
he
li
nea
r
discrim
i
nation analy
sis al
gorithm
(
LDA)
.
Re
cogniti
on
usi
ng
t
he bat al
gorithm
:
a.
In
te
ll
igent
bat
al
gorithm
dev
el
op
e
d:
The
tra
diti
on
al
bats
al
gorithm
was
dev
el
oped
by
ad
ding
seve
ral
ste
ps
to
im
pr
ov
e
t
he
res
ults
an
d
th
ei
r
su
it
abili
ty
to
the
disc
rim
i
nating
pro
cess
.
The
pro
pose
d
al
gorithm
w
as
cal
le
d
the
de
ve
lop
e
d b
at
alg
ori
th
m
D
BA
w
hich
is c
onsist
s of the
foll
ow
i
ng steps (ste
ps a
dded
in b
old):
1)
Def
i
ne
the
tar
ge
t functi
on re
presente
d by the
m
us
ic
al
inp
ut.
2)
Param
et
ers
con
fi
gurati
on
:
G
iven
the
init
ia
l
values
an
d
con
sta
nt
values
of
the
pa
ram
et
ers
for
i
m
ple
m
enting
the
al
go
rithm
a
s foll
ow
s:
a)
Determ
ining
th
e num
ber
of
ba
ts wit
h
t
he num
ber
of
m
od
el
s of the
m
us
ic
al
n
otes
.
b)
Sets t
he veloci
t
ie
s (
V
i
) of t
he bats
by d
e
fining a m
at
rix
that
is eq
ual in s
iz
e
m
us
ic
al
n
otes.
c)
Determ
ine the
pu
lse
freq
ue
nc
y (Q
i
)
s
o
that t
he fre
qu
e
ncy
va
lue is
betwee
n Q
m
in
and
Q
m
a
x
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
55
8
-
56
6
562
d)
Determ
inati
on
of the
r
i
pu
lse
ra
te
s [
0
-
1].
e)
Sele
ct
a v
al
ue
f
or lo
udness
A
i
as a r
a
ndom
v
a
lue w
it
hi
n (0,
1), t
o be a c
onsta
nt v
al
ue fo
r
al
l cy
cl
es.
f)
Determ
ine the
nu
m
ber
of it
er
at
ion
s
by the
num
ber
of
bats.
3)
The
weig
hts
m
at
rix
f
or
t
he
ta
rg
et
f
unct
ion
an
d
f
or
the
c
urren
t
bat
f
or
each
cy
cl
e
are
init
ia
li
zed
as
a
new
ste
p
s
upporte
d
by
the
al
gorithm
and
re
pr
ese
nt
the
we
igh
ts
der
ive
d
f
ro
m
the
li
near
discrim
inati
on
analy
sis al
gorithm
(
LDA).
4)
Find
the
fitnes
s
functi
on
for
each
m
at
rix
of
b
at
wei
gh
ts
for
the
wei
gh
t
m
at
rix
of
the
ta
r
ge
t
functi
on
by
us
in
g
t
he
c
orre
la
ti
on
b
et
wee
n t
he
tw
o
m
at
ric
es r
e
pr
ese
nted
by the
f
ollow
i
ng
form
ula
[23]:
m
n
m
n
B
mn
B
W
mn
W
B
mn
B
m
W
n
mn
W
f
i
t
n
es
s
2
))
((
2
))
(
)
)(
(
(9)
W
:
represe
nts
the
weig
ht
m
at
rix
of
a
bat
w
ho
se
siz
e
is
m
*
n.
B:
represents
th
e
weig
ht
m
atr
ix
of
the tar
get
f
unct
ion
w
ho
s
e size
is al
so
m
*
n,
)
(
2
A
m
e
a
n
W
,
A
nd
)
(
2
B
m
e
a
n
B
C
r
e
a
t
e
n
e
w
s
o
l
u
t
i
o
n
s
b
y
a
d
j
u
s
t
i
n
g
f
r
e
q
u
e
n
c
y
a
n
d
u
p
d
a
t
i
n
g
s
p
e
e
d
s
a
n
d
p
o
s
i
t
i
o
n
s
b
y
t
h
e
f
o
l
l
o
w
i
n
g
e
q
u
a
t
i
o
n
s
:
Q
Q
Q
i
Q
)
m
i
n
m
ax
(
m
i
n
(10)
i
Q
X
t
i
X
t
i
V
t
i
V
)
*
(
1
(11)
t
i
V
t
i
X
t
i
X
(12)
As: Q
i
: The
ne
w
f
re
quency
of the
b
a
t i
.
X
i
t
-
1
: The
value
of t
he old
so
l
ution.
β
: A ran
dom
v
al
ue
c
onfine
d be
tween
durati
on
(0,
1).
V
i
t
: The am
ount of
new v
el
ocity
of the i i
n
the
c
urren
t
ste
p
t.
V
i
t
-
1
: The am
ount of
bat sp
ee
d
i i
n t
he
pre
vious st
ep
t
-
1.
X
i
t
: The s
olu
ti
on
wh
e
n
t
he
c
urre
nt step
t.
X
*
: R
epr
ese
nts th
e b
est
c
urre
nt bat
that re
pr
ese
nt
s the
best curr
ent sit
e.
5)
Ba
la
nces
the
pulse
rate
with
a
ra
ndom
valu
e
to
determ
ine
wh
et
her
a
s
olut
ion
ca
n
be
ide
ntifie
d
am
on
g
the b
e
st ne
w
s
ol
ution
s
and c
re
at
e a local s
olut
ion
a
rou
nd the
b
est
s
olu
ti
on
s
el
ect
ed.
6)
Gen
e
rate
a
ne
w
s
olu
ti
on
by
rand
om
fligh
t
of
t
he
bat
usi
ng
the
f
ollo
wing
e
qu
at
io
n.
T
he
am
ou
nt
of
the
ste
p
i
n
ra
nd
om
fligh
t
is
determ
ined
a
s
a
ra
ndom
va
lue
(
0.0
01).
The
fitness
functi
on
is
the
n
cal
culat
ed
f
or t
he new
s
olu
ti
on
for
the
tar
get fun
ct
io
n.
X
X
o
l
d
n
e
w
001
.
0
*
(13)
As: X
new
t
he va
lue of t
he best
new so
l
utio
n,
X
old
: B
est
v
al
ue
for
old s
olu
ti
on,
: ra
ndom
v
al
ue
[
-
1,
1].
7)
Ba
la
nce th
e fitness v
al
ue
of the n
e
w
so
l
utio
n
with the val
ue
o
f
fitnes
s for
the o
ld s
olu
ti
on and co
m
pare
the
loud
ness
with
the
ra
nd
om
value
to
determ
ine
if
t
he
cu
rr
e
nt
bat
is
cl
os
e
to
the
prey
bette
r
,
then
determ
ine the acce
ptance
of
new so
l
utio
ns
or not.
8)
Check
t
he
st
op
co
nd
it
io
n
(
nu
m
ber
of
bat)
?
I
f
the
c
onditi
on
is
m
et
,
it
will
c
on
ti
nue,
ot
herwise
it
will
be
ref
e
rr
e
d
to
step
5
.
9)
Arrange
bats a
nd f
i
nd the
bes
t X
*
c
urren
t,
by
f
in
ding the
larg
est
c
urre
nt f
i
tnes
s
functi
on.
10)
Find the
best s
olu
ti
on sit
e,
w
hich re
pr
ese
nts
the
best
bat sit
e.
11)
Disp
la
y re
su
lt
s
.
b.
Parall
el
bats
i
m
ple
m
entat
ion
al
gorithm
:
a
par
al
le
l
-
dev
el
oped
bats
al
gor
it
h
m
with
a
s
ing
le
m
ulti
-
cor
e
process
or
was
su
ggest
e
d.
Pa
r
al
le
l
pr
ogram
m
ing
m
od
el
us
ed
f
or
this
pur
po
s
e
is
SPMD
m
od
el
.
Each
cor
e
perform
s
on
e
pro
gr
am
in
the
process
or
at
on
e
tim
e
on
dif
fe
ren
t
dataset
s.
T
his
m
od
el
is
ba
sed
on
a
ty
pe
of
par
al
le
l
pro
gr
a
m
m
ing
m
od
e
l,
the
m
ast
er
-
sla
ve
Mo
del.
Parall
el
m
e
thod
us
ed
in
t
his
de
velo
ped
ba
t
a
lgorit
hm
w
as the alg
or
it
hm
-
le
vel.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
par
allel
b
at
algorit
hm f
or
music
al
no
t
e reco
gnit
ion
(
Ansam
Naz
ar
Youn
is)
563
6.2.
Design o
f the
t
est
p
ha
se
The
te
st
databa
se
was
create
d
from
a
colle
c
tio
n
of
diff
e
re
nt
m
us
ic
al
ta
g
i
m
ages.
Diff
e
re
nt
m
od
el
s
of
the
trai
ni
ng
m
od
el
s
are
create
d
f
r
om
each
ty
pe
of
m
us
ic
al
la
bel,
as
well
as
the
cr
eat
ion
o
f
m
od
el
s
that
in
cl
ud
e
the
m
us
ic
al
lin
es
a
nd
the
c
r
eat
ion
of
m
odel
s
that
are
st
r
ang
e
to
t
he
m
us
ic
la
bels.
I
n
the
f
old
e
r
(T
est
ing
)
500 (
300+
200) d
if
fer
e
nt im
age of 100
dif
fere
nt ty
pes of m
us
ic
al
labels,
a
s w
el
l as
50
oth
er
stra
ng
e
im
ages.
7.
EVAL
UA
TI
O
N
S
YS
TE
M
P
ERFO
RMA
N
CE
MEASU
R
ES
a.
Detect
ion
rate
DR: The
r
at
io
of the
recog
niz
ing
of m
us
ic
notes in
the syst
e
m
is correct.
b.
Tru
e
ne
gative
TN: T
he nu
m
ber
of
im
ages of
w
r
ong m
us
i
cal
n
otes
, a
nd r
ec
ognized
wr
ong
.
c.
False
posit
ive
r
at
e
FPR:
Th
e
num
ber
of
wro
ng m
us
ic
al
n
ote
s im
ages an
d
r
ecognize
d c
orr
ect
.
d.
Tru
e
posit
ive
r
at
e
TPR:
I
t
represents
the
nu
m
ber
o
f
c
orrec
t
m
us
ic
notes,
and is
prop
e
rly
cate
gorized
.
e.
False
ne
gative
FN
: It
r
e
pr
ese
nt
s the
nu
m
ber
of co
rr
ect
m
us
ic
notes, a
nd is
pro
per
ly
wr
ong
cat
eg
ori
zed.
f.
Sens
it
ivit
y:
Mea
su
re
s the
ab
il
it
y of
t
he discri
m
inati
ng
syst
e
m
to f
in
d
im
ages of wr
ong m
us
ic
al
notes.
g.
Sp
eci
fici
ty
: M
easur
e
s the
ab
i
li
ty
o
f
the
discr
i
m
inati
ng
syst
em
to f
in
d
im
ages of
c
orrect m
us
ic
al
notes.
h.
False
r
e
j
ect
io
n rati
o
FRR
: T
he
prop
or
ti
on
of
wrong
ly
tag
ge
d
m
us
ic
i
m
ages
(
no
t
f
ound).
i.
Wro
ng accepta
nce
rati
o WRR
: The
pro
portio
n of w
ron
gly acc
epted
m
us
ic
scor
e
s (Wr
on
g resu
lt
)
.
The follo
wing
equ
at
io
ns ca
n
be
ca
lc
ulate
d
f
ro
m
li
ne
rati
os
above
[
24
]
:
a.
Detect
ion
rate
DR=(
Nu
m
ber
of sam
ple co
rrec
tl
y detec
te
d)
/(
To
ta
l
nu
m
ber
of sam
ples)*
10
0.
b.
Sens
it
ivit
y=
TPR/
(TPR+FN
)
*100
%
.
c.
Sp
eci
fici
ty
=TN/(T
N+FP
R
)*100 %
.
d.
False
ne
gative
rate
(FN%
)=F
N/(TPR+
FN)=
100 %
-
Se
ns
it
iv
it
y
.
e.
False
posit
ive
r
at
e (
FPR%
)=F
PR/
(TN
+
FPR)
=100 %
-
S
pecif
ic
it
y
.
f.
Po
sit
ive
pr
e
dic
ti
ve
rate
(P
P%
)
=TPR/
(TPR+F
PR)*1
00 %
.
g.
Neg
at
ive
pre
dicti
ve
rate
(N
P
R%)=TN/
(TN+FN)*
100 %
.
h.
False
re
j
ect
io
n rate
(F
RR
)=(N
um
ber
of im
ages r
e
j
ect
ed
er
ror)
/(T
otal n
um
ber
of
im
ages)
*100%
.
i.
Wro
ng
acce
pta
nce
rate
(
WRR
)=(Nu
m
ber
of im
ages accepte
d
e
rror)/(T
otal
nu
m
ber
of im
a
ges)*
100%
.
8.
IMPLEME
N
TATION
OF
THE
PROPO
SED
ALGO
R
ITHM
Im
ple
m
entation
wer
e
c
onduct
ed
an
d
th
e
pr
op
os
e
d
PBM
RS
syst
e
m
was
te
ste
d
to
identify
the
di
ff
e
ren
t
m
us
ic
al
note
s
in
two
ways
(d
e
le
ti
ng
the
m
us
ic
al
li
nes
an
d
not
delet
in
g
the
m
us
ic
al
li
nes)
an
d
us
in
g
t
wo ty
pe
s
of im
ple
m
ent
at
ion
m
et
ho
ds
(seque
ntial
and
p
a
rall
el
).
8.1
.
Sy
s
tem
tra
ini
ng
and te
stin
g
st
ag
e
The
data
we
re
trai
ned
in
the
pr
op
os
e
d
PBM
RS
syst
e
m
consi
sti
ng
of
700
im
ages.
The
res
ults
of
the
trai
ning
wa
s
ob
ta
i
ned
acc
ordin
g
to
th
e
evaluati
on
m
eas
ur
es
,
w
her
e F
N,
F
PR,
FRR
a
nd
W
RR
eq
ual
to
0%
.
The
val
ue
of
D
R,
NP
R,
PP,
S
ensiti
vity
and
Sp
eci
fici
ty
is
10
0%
.
The
propose
d
PBM
RS
al
gorithm
was
te
ste
d
on a c
ollec
ti
on
of 30
0
im
ages o
f
m
us
ic
al
n
ot
e
s.
8.
2.
Recen
t
e
xp
eri
ences
200
ne
w
im
ag
es
wer
e
ad
de
d
to
the
i
m
ages
of
the
no
te
s,
t
he
fi
rst
sect
ion
con
ta
in
s
the
sta
ff
li
nes
an
d
ano
t
her
sect
i
on
is
fr
ee
f
r
om
t
he
li
nes,
a
nd
a
gro
up
of
e
xoti
c
i
m
ages
(50
pi
ct
ur
es)
,
w
hich
represe
nt
i
m
a
ges
of
s
o
m
e
s
h
a
p
e
s
a
n
d
s
y
m
b
o
l
s
,
a
r
e
a
d
d
e
d
t
o
t
h
e
m
u
s
i
c
a
l
n
o
t
e
s
.
I
m
a
g
e
t
h
e
t
e
s
t
r
e
s
u
l
t
s
w
e
r
e
a
s
s
h
o
w
n
i
n
t
h
e
T
a
b
l
e
1.
Table
1.
Res
ults o
f
the
syst
em
in
the
test
ing s
ta
ge
Ty
p
e of
cr
it
erion
Res
u
lt by
wa
y
of
deletin
g
staf
f
lines
Res
u
lt in a wa
y
tha
t the staf
f
lin
es are
no
t delete
d
FN
2
.67
%
1
3
.5%
FPR
0%
8%
DR
9
7
.34
%
8
6
.5%
NPR
100%
6
3
.01
4
%
PP
100%
9
7
.75
%
Sen
sitiv
ity
9
7
.34
%
8
6
.5%
Sp
ecif
icity
100%
92%
FRR
1%
1
1
.5%
W
RR
1
.67
%
2%
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
55
8
-
56
6
564
9.
CR
ITE
RI
A
F
OR
I
N
CR
E
A
SIN
G
S
YS
TE
M
S
PEE
D AND EFFI
CIE
N
CY
[
25,
26]
Spee
d
Up
=
The
e
xe
c
ut
ion
ti
me
of
seria
l
e
xe
c
ut
ion
(
on
e
c
ore
)
The
e
xe
c
u
tion
t
ime
on
n
c
o
res
(14)
Sp
ee
d up Ef
fic
ie
ncy
=(S
peed
Up)/n
(
15
)
n
:
represe
nts
the
nu
m
ber
of
cor
es
use
d
in
t
he
im
ple
m
enta
ti
on
,
t
he
syst
e
m
was
i
m
ple
m
ented
i
n
a
s
e
q
u
e
n
t
i
a
l
a
n
d
p
a
r
a
l
l
e
l
m
a
n
n
e
r
u
s
i
n
g
t
w
o
a
n
d
t
h
r
e
e
a
n
d
f
o
u
r
c
o
r
e
s
,
a
n
d
t
h
e
r
e
s
u
l
t
s
i
n
d
i
c
a
t
e
d
i
n
t
h
e
T
a
b
l
e
2
w
e
r
e
o
b
t
a
i
n
e
d
:
Table
2.
Res
ults o
f
se
rial
i
m
ple
m
entat
ion
a
nd p
a
rall
el
i
m
ple
m
entat
ion
Res
u
lt by
wa
y
of
deletin
g
staf
f
lines
Res
u
lt in a wa
y
tha
t the staf
f
lines
ar
e
n
o
t deleted
No
.
o
f
cores
Sp
eed Up
Sp
eed u
p
Ef
f
icien
c
y
Sp
eed Up
Sp
eed u
p
E
f
f
icien
c
y
serial
1
1
1
1
2
-
co
res
1
.61
6
8
8
0
.80
8
4
4
1
.56
1
5
0
.78
0
8
3
-
co
res
1
.95
3
7
6
0
.65
1
2
5
3
1
.82
2
0
.60
7
1
4
-
co
res
2
.37
1
0
.59
2
7
2
.01
9
8
0
.50
5
10.
DISCU
SSI
ON
It
was
no
te
d
in
Table
2
that
the
per
centa
ge
of
discrim
inati
on
reac
hed
95.5%
by
delet
ing
m
us
ic
a
l
li
nes,
wh
il
e
86
.5
%
by
not
del
et
ing
those
li
nes.
The
us
e
of
the
process
of
delet
ing
m
us
ic
li
nes
is
the
best
wa
y
to
ob
ta
in
cl
ear
er
im
ages
an
d
th
us
get
high
er
r
esults
of
di
scri
m
inati
o
n
a
nd
sta
nda
rd
s
a
t
bette
r
rates
.
It
was
ob
s
er
ved
in
T
able
2
that
the
us
e
of
th
ree
cor
es
an
d
f
our
co
res
achie
ves
a
higher
s
peed
tha
n
tw
o
cor
es,
al
tho
ug
h
the
use
of
th
ree
co
r
e
and
f
our
c
ore
achieves
a
hi
gh
e
r
acce
le
rati
on
factor
t
han
the
us
e
of
tw
o
cor
e
i
n
the
i
m
ple
m
entat
ion
.
H
ow
e
ve
r,
the
us
e
of
t
wo
co
re
is
the
m
os
t
eff
ic
ie
nt,
especial
ly
the
us
e
of
the
m
e
thod
of
delet
ing
t
he
li
nes
of
t
he
m
us
ic
al
scal
e
as
the
eff
ic
ie
ncy
of
the
sp
e
ed
of
the
syst
em
abo
ut
(
0.808
44)
be
caus
e
the
siz
e
of
dat
a
us
e
d
in
eac
h
nu
cl
e
us
in
t
he
i
m
ple
m
entat
io
n
of
the
us
e
of
two
c
or
e
is
la
rg
e
r
tha
n
the
s
iz
e
of
data
us
e
d
in
each
co
re
in
the
three
-
co
res
and
f
our
im
plem
entat
ion
.
A
com
par
iso
n
was
m
ade
between
the
res
ults
of
t
he
pr
opos
e
d
P
BM
RS
syst
e
m
and
t
he
res
ults
of
a
num
ber
of
researc
he
rs
in
the
s
am
e
fi
el
d
as
sh
ow
n
in
Ta
ble 3
.
Table
3.
A
c
om
par
ison
of r
e
su
lt
s w
it
h ot
he
r
al
gorithm
s
No
.
Res
earc
h
na
m
e
Alg
o
rith
m
Nu
m
b
e
r
o
f
Mus
ical Notes
T
est
m
o
d
els
n
u
m
b
er
Reco
g
n
itio
n
Rate
1
Mus
ic Sy
m
b
o
l Rec
o
g
n
itio
n
/2
0
1
1
Te
m
p
late
M
a
tch
in
g
7
585
9
4
.35
9
%
2
Mus
ical Note
Re
c
o
g
n
itio
n
Usin
g
Mini
m
u
m
Sp
an
n
in
g
T
ree
Alg
o
rith
m
/2
0
1
4
Mini
m
u
m
Sp
an
n
in
g
Tr
ee
&
Euclid
ean d
istan
ce
7
97
9
7
.45
%
3
Detectin
g
Note
h
e
ad
s in
Hand
written
Sco
res with Co
n
v
.
Nets an
d
Bo
u
n
d
i
n
g
Bo
x
Re
g
ressio
n
/2
0
1
7
CNN Co
n
v
o
lu
tio
n
al
Neu
ral
N
etwo
rk
7
140
96%
4
Distin
g
u
ish
M
u
sical Sy
m
b
o
l Pr
in
ted
u
sin
g
the Linear
/
Discri
m
in
an
t
An
aly
sis
L
DA
an
d
Si
m
ila
rit
y
Scale/2
0
1
8
Linear
Discri
m
in
a
n
t
An
aly
sis
L
DA
&
Si
m
ila
rity Scale
S
SIM
15
180
8
9
.5%
5
A New
Par
allel Ba
t Algo
rith
m
For
Mus
ical Note
Re
c
o
g
n
itio
n
/2
0
2
0
Bat Alg
o
rith
m
s
(Sequ
en
tial execu
tio
n
,
P
arallel
execu
tio
n
)
100
1250
9
7
.34
%
11.
CONCL
US
I
O
N
The
e
xtracti
on
of
pro
per
ti
es
usi
ng
L
D
A
al
go
rithm
h
el
ped
r
e
du
ce t
he
re
ser
voir ca
pacit
y of
the im
ages
and
re
du
ce
t
he
i
m
ple
m
entation
ti
m
e
of
the
syst
e
m
.
Af
te
r
the
siz
e
of
t
he
i
m
ages
use
d
a
bout
10
00*
625
,
it
beca
m
e
a
m
at
rix
of
cha
rac
te
risti
cs
wit
h
two
dim
ensions
about
62
5*
625.
Als
o
ch
oo
se
a
s
m
al
l
co
ns
ta
nt
rand
om
v
al
ue
rep
rese
nted by the v
al
ue (0
.00
1)
wh
ic
h
is use
d
to d
et
erm
ine the a
m
ount o
f st
ep
(f
li
ght dist
ance)
of
t
he
bat.
It
is
an
im
po
rta
nt
ste
p
ad
de
d
to
the
al
go
rithm
becau
se
t
he
ste
p
am
ou
nt
in
t
he
bat
is
us
ua
ll
y
la
rge
because
it
is
wait
ing
for
ech
o
befor
e
decidin
g
the
ty
pe
of
ta
rg
et
an
d
dete
r
m
ining
it
s
destinat
ion
.
Deter
m
ining
the
am
ou
nt
of
stock
ste
p
re
duces
the
unne
ce
ss
ary
la
rg
e
dim
ensio
ns
that
do
no
t
fit
their
use
.
So
t
he
pe
rce
ntag
e
of
discrim
inatio
n
in
the
m
ann
e
r
of
dele
ti
ng
the
li
ne
s
was
e
qu
al
t
o
(
96.
6%)
a
nd
the
per
c
ent
age
of
discrim
inati
on
by w
ay
of
non
-
delet
ion
of li
ne
s equal t
o (
93%).
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A n
ew
par
allel
b
at
algorit
hm f
or
music
al
no
t
e reco
gnit
ion
(
Ansam
Naz
ar
Youn
is)
565
As
a
fu
t
ur
e
works:
A
dd
i
ng
the
m
et
ho
d
for
p
ri
nciple
com
po
ne
nts
analy
sis
(P
C
A
),
the
n
us
in
g
the
li
near
discrim
inati
on
a
na
ly
sis
m
et
ho
d
t
o
ob
ta
i
n
str
on
ger
cha
racteri
s
ti
cs
of
the
im
ages,
an
d
t
hus
ob
ta
in
bette
r
discrim
i
nation,
T
he
use
of
a
great
er
num
ber
of
i
m
age
m
od
el
s
of
m
us
ic
al
no
t
es,
whi
c
h
stre
ng
t
hens
the
process
of
disti
nguish
i
ng
an
d
be
ne
fiti
ng
from
the
a
dv
a
ntage
s
of
par
al
le
l
i
m
plem
entat
ion
in
a
m
or
e
eff
ic
ie
nt
m
anner,
s
uch
as
a
dd
i
ng
ha
ndwr
i
tt
en
m
us
ic
al
sign
s
,
an
d
anal
yz
e
the
m
us
ic
al
no
te
s
place
d
on
the staf
f by det
erm
ining
their
po
sit
i
on
on the
staff
li
nes
a
nd
wh
at
t
on
e
they
pro
du
ce
b
ase
d on the Cl
ef
.
ACKN
OWLE
DGE
MENTS
The
a
uthors
are
ve
ry
grat
efu
l
to
t
he
unive
rsity
of
Mosu
l
/C
ollege
of
c
om
pu
te
r
Scie
nce
a
nd
Ma
them
a
ti
cs f
or their
pro
vide
d faci
li
ti
es, which
helpe
d
t
o im
pr
ove the
qu
al
it
y
of
this
work.
REFERE
NCE
S
[1]
B
.
H
a
n
s
e
n
,
e
t
a
l
.
,
“
I
n
t
r
o
d
u
c
t
i
o
n
t
o
M
u
s
i
c
A
p
p
r
e
c
i
a
t
i
o
n
,
”
e
P
r
e
s
s
C
o
u
r
s
e
M
a
t
e
r
i
a
l
s
,
A
m
e
r
i
c
a
n
P
u
b
l
i
c
U
n
i
v
e
r
s
i
t
y
S
y
s
t
e
m
,
2
0
1
4
.
[2]
S.
J.
Mohammed,
“
Standa
rd
pri
nte
d
m
usica
l
not
e
rec
o
gni
ti
on
ba
sed
on
neur
al
ne
twork
,
”
J
ournal
of
U
niv
ersity
o
f
A
nbar f
or
P
ure
S
ci
en
ce
,
v
ol
.
6
,
no
.
2
,
pp
.
20
-
25
,
2012.
[3]
J.
Tohka,
“
SGN
-
2506
:
Introdu
ct
i
on
to
Pa
tt
ern
Re
cogni
ti
on
,
”
T
ampere
Univ
ersity
of
Technol
og
y
,
20
06
.
[4]
J.
O.
B.
Ee
and
A.
W
.
C.
Ta
n
,
“
Mus
ic
Sy
m
bol
Rec
ognition
,
”
S
ymposium
on
Information
&
Co
m
pute
r
Sci
en
ce
s
(
ICS
2011)
,
pp.
1
-
4,
2011
.
[5]
Y.
Saza
ki
,
et
a
l.
,
“
Mus
ic
al
not
e
rec
ogni
ti
on
usin
g
M
ini
m
um
Spa
nning
Tree
Alg
orit
hm
,
”
2014
8
th
Inte
rnat
ional
Confe
renc
e
on
Telecomm
unic
ation Sy
stems S
erv
i
ce
s and Appl
ic
at
ions (
TSSA)
,
Kuta
,
p
p
.
1
-
5,
2
014.
[6]
J.
Haji
ˇc
and
P.
Peci
na
,
“
Dete
cting
Notehea
ds
in
Handwrit
te
n
Score
s
wit
h
ConvNets
and
Bounding
Box
Regre
ss
ion
,
”
ar
Xi
v:
1708.
01806
v1
,
2017
.
[7]
A.
N.
Younis
an
d
F.
M.
R
emo
,
“
Distingui
sh
Mus
ic
a
l
S
y
m
bol
Prin
te
d
using
the
Linea
r
Discr
imina
nt
An
aly
sis
LD
A
and
Sim
il
ar
ity
S
c
ale
,
”
Int
ernati
o
nal
Journal
of
C
omputer
Applica
ti
ons
,
v
ol
.
179
,
n
o.
47
,
pp
.
20
-
24
,
2018.
[8]
C
.
Schm
idt
-
Jone
s,
“
Rea
ding
Mus
ic
:
Com
m
on
Notat
ion
,
”
CONNEXIONS
,
Ri
ce Unive
rsit
y
,
2008
.
[9]
J.
Harnum
,
“
Bais
c
m
usic
th
eor
y
:
how t
o
re
ad
,
wri
te
,
and
under
st
an
d
written
m
usic
,”
Sol
-
Ut
Press
,
2
001.
[10]
X.
-
S.
Yang,
“
A
New
Meta
heur
isti
c
Bat
-
Inspire
d
Algorit
hm
,
”
i
n:
Nature
Inspired
Cooperati
ve
Strate
gie
s
for
Optimizati
on
(
N
ISCO
2010)
(
Eds.
J.
R
.
Gonz
ale
z
et
al.
)
,
Studi
es
in
Computati
on
al
Inte
l
li
gen
ce,
Springer
Be
rlin
,
284,
Springer
,
65
-
74
,
2010
.
[11]
El
-
Ghaz
al
i
T.
,
“
Meta
heur
isti
cs
f
r
om
Design
t
o
Im
ple
m
ent
a
ti
on,
”
Wil
e
y
,
2009.
[12]
Y.
Zhou,
et
a
l.
,
“
A
Novel
H
y
bri
d
Bat
Algorit
hm
for
the
Multi
level
Thre
shold
in
g
Medic
al
Im
ag
e
Segm
ent
at
ion
,
”
Journal
of
Me
d
i
cal
Imaging
and
Healt
h
In
formati
cs
,
v
ol
.
5
,
no
.
8
,
pp.
1
742
-
1746
,
2015.
[13]
X.
-
S.
Yang
and
X.
He
,
“
Ba
t
Alg
orit
hm
:
Liter
at
ur
e
Review
and
Applicati
ons,
”
Int
ernati
onal
J
ournal
of
Bio
-
Inspire
d
Computati
on
,
v
o
l.
5
,
n
o
.
3
,
p
p
.
14
1
-
149
,
2013
.
[14]
A.
M.
Ta
h
a,
et
al
.
,
“
Naive
B
a
yes
-
Guided
Bat
Algorit
hm
for
Fea
tur
e
Sel
ec
t
ion
,
”
The
S
ci
en
ti
f
ic
World
Journal
,
v
ol
. 2
013
,
pp
.
1
-
9,
2013
.
[15]
I.
Gu
y
on
and
A.
E
li
ss
ee
ff
,
“
An
Introd
uc
ti
on
to
Vari
abl
e
and
Feat
ur
e
Sele
c
ti
on
,
”
Journal
of
Mac
hine
Learning
Re
search
,
vol
.
3
,
pp.
1157
-
1182,
2003.
[16]
S.
Cat
en
i,
et
a
l.
,
“
Vari
abl
e
S
el
e
ct
ion
and
Feat
ur
e
Ext
r
actio
n
t
hrough
Arti
fic
i
al
Int
el
l
ige
n
ce
T
ec
hniqu
es,
”
Mult
iv
aria
te A
n
aly
sis in
Manag
eme
nt, E
ng
ine
er
ing
and
the Sc
i
e
nce
s
,
InTech,
20
13.
[17]
A.
El
e
y
a
n
and
H.
Dem
ire
l,
“
PC
A
and
LDA
base
d
neur
al
n
et
w
o
rks
for
Hum
an
F
ac
e
R
ec
ogni
ti
o
n
,
”
in
K
.
Del
ac
an
d
M
.
Grgic
(ed
s)
,
“
Face
Re
cogni
t
i
on,
”
I
-
Tech Edu
cat
ion
and
Pub
lis
hing
,
Vienn
a, A
ustria
,
pp.
93
-
10
6
,
2007
.
[18]
C.
Bishop
,
“
Pattern
Re
cogni
t
ion and
Mac
h
ine L
e
arn
ing
,”
Spring
e
r Sc
ie
n
ce
+Busin
ess Me
dia
,
L
LC,
2006.
[19]
S.
Singh
,
“
Ba
yesia
n
dist
ance
m
et
ric
l
ea
rn
ing
and
it
s
appl
i
c
at
ion
in
aut
om
a
ti
c
spe
ake
r
r
ecogniti
on
s
y
s
te
m
s,
”
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
(
IJE
CE)
,
v
ol
.
9
,
n
o
.
4,
pp.
2960
-
296
7
,
2019
.
[20]
B.
Basil
,
“
New
Plat
form
for
Mobile
Cloud
Com
puti
ng
,”
M.Sc.
Th
esis
,
U
niv
ersity
of
Mos
ul/
Com
pute
r
an
d
M
at
hematics Co
ll
eg
e,
2017.
[21]
B.
Barn
e
y
,
“
Intr
oduct
ion
to
Par
a
ll
el Com
puti
ng
,”
La
wren
ce Liver
m
ore
Nati
ona
l L
abor
at
or
y
,
2010
.
[22]
W
.
Stal
li
ngs,
“
Com
pute
r
Organi
za
t
ion
a
nd
Archi
te
c
ture
Designin
g
f
or
Perform
an
ce
,”
Ei
ghth
Edi
tion
,
P
rentice
Hal
l
Upper Saddl
e
Rive
r
,
2010
.
[23]
R.
C.
Gonza
lez
a
nd
R.
E.
W
oods,
“
Digit
al Im
age
Proce
ss
ing
Thi
rd
Edition
,”
Pe
ars
on
Prentice Hall
,
2008
.
[24]
A.
Naz
ar,
“
Deve
lopi
ng
and
Im
pl
ementa
t
ion
of
Meta
h
eur
isti
c
Alg
orit
hm
Sequenti
al
l
y
and
Pa
rallel
for
Mus
ic
al Not
es
Re
cogni
t
ion
,”
M.
The
sis,
Dep
art
m
ent
o
f
Com
pute
r
Sci
ence,
Coll
ege
o
f
Com
pute
r
a
nd
Math
emati
c
al
Sc
ie
n
c
es
,
Univer
sit
y
o
f
Mos
ul,
2018.
[25]
P.
Pa
che
co
,
“
An Int
roduc
ti
on
to P
ara
llel
Progr
amm
ing
,”
E
lsevie
r
Inc
,
2011
.
[26]
P.
Salz
a
and
F.
Ferruc
ci,
“
Speed
up
gene
tic
a
lgor
it
hm
s
in
the
cl
o
ud
using
software
containe
rs
,”
F
uture
Gene
ratio
n
Computer
System
s
,
vol.
92
,
pp.
276
-
289,
2019
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
55
8
-
56
6
566
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
An
sam
Na
z
ar
You
nis
She
has
bee
n
an
assistan
t
li
t
era
tur
e
at
d
e
par
tment
of
computer
scie
n
ce
s
,
col
l
ege
of
computer
sc
ie
nc
es
and
m
at
hemat
ics
,
the
Univer
sit
y
of
Mos
ul,
Ir
aq
sinc
e
2018
,
Gradua
te
d
f
rom
the
Com
pute
r
Sc
ie
nc
e
and
Ma
them
at
ic
s
Coll
age
a
t
the
Univ
ersity
of
Mos
ul,
Ira
q
i
n
2005,
and
worked
as
a
prog
ramm
er
in
the
s
ame
col
l
age
un
t
il
2013
when
s
he
al
so
sta
rte
d
stud
y
ing
M
aste
r
s
of
Scie
nc
e
in
the
sam
e
col
l
a
ge,
th
en
she
f
in
ished
MS
C.
De
gre
e
at
2018
.
Gene
ral
expe
rt
ise
is
computer
scie
nc
e,
and
spe
cialty
is
in
the
area
of
art
ifici
a
l
intell
ig
ence
and
image
proc
essing.
She
is
thi
s
r
ese
arc
h
'
s
co
-
au
t
hor.
She
h
as
a
rese
ar
ch
g
ate
ac
coun
t
unde
r
the
n
ame
Ans
am Naz
a
r. Email i
s:
an
y
m
a8@uom
osul.
edu.iq
Fa
w
z
iy
a
Mahm
ood
Ramo
Obt
ai
ned
BA
degr
e
e
in
Com
pute
r
Scie
nc
e
in
1992
,
the
n
obt
ai
n
ed
MA
degr
ee
in
Com
pute
r
Archi
te
c
ture
in
2001
and
PhD
in
A
rti
fi
ci
a
l
Inte
l
li
ge
nce
in
2007
,
o
bta
in
ed
As
sistant
Profess
or
in
2013
,
intere
sted
b
y
rese
arc
h
in
computer
scie
n
c
e
and
ar
ti
fi
ci
a
l
int
ellige
n
ce
and
m
ac
hine
learni
n
g.
ema
il
:
fm
rm
b
7@
y
ahoo
.
com
Evaluation Warning : The document was created with Spire.PDF for Python.