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.
9
, No
.
5
,
Octo
ber
201
9
, pp.
4092
~4
098
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v9
i
5
.
pp4092
-
40
98
4092
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A new m
ethod fo
r voic
e s
i
gnal fe
atu
re
s
creation
Majed
O. Al
-
Dwairi
1
, Amj
ad
Y. Hen
di
2
,
Moham
ed
S.
So
li
ma
n
3
,
Z
iad
A
.A.
Alqadi
4
1
,2
D
epa
rtment
of
Com
m
unic
at
ion
Engi
ne
eri
ng
Tec
hnolog
y
,
Fa
cul
t
y
of
Eng
ineeri
ng
Te
chno
log
y
,
Al
-
B
al
qa
Applied Unive
rsit
y
,
Jor
dan
3
De
par
tment of
El
e
ct
ri
ca
l
Eng
in
ee
ring
,
Fa
cul
t
y
o
f
Engi
n
ee
ring
,
T
ai
f
Univ
ersity
,
T
ai
f, Kingdom
of S
audi
Arabi
a
3
Depa
rtment of
El
e
ct
ri
ca
l
Eng
in
ee
ring
,
Fa
cul
t
y
o
f
Ene
rg
y
Eng
ineeri
ng
,
As
wan
Univer
sit
y
,
As
wan,
Eg
y
p
t
4
Depa
rtment of
Com
pute
r
and
Network
Eng
ineer
ing
,
Fa
cul
t
y
of E
ngine
er
ing
T
ec
h
nolog
y
,
Al
-
B
al
qa
Applied
Univer
sit
y
,
Jor
dan
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
A
ug
7
, 2
01
8
Re
vised
A
pr
18
, 2
01
9
Accepte
d
Apr
25
, 201
9
Digit
al
audi
o
is
one
of
the
m
ost
important
t
y
pes
of
dat
a
a
t
pre
sen
t.
It
is
used
in
seve
ral
appl
i
ca
t
ions,
such
as
hum
an
knowledge
and
m
an
y
s
ec
uri
t
y
and
banki
ng
appl
i
ca
t
ions.
A
dig
it
a
l
v
oic
e
signal
is
us
ual
l
y
of
a
l
arg
e
size
wher
e
the
a
cousti
c
sig
nal
consists
of
a
set
of
va
lue
s
distri
bute
d
in
one
col
um
n
(one
cha
nne
l)
(m
ono
signal
)
or
distri
bute
d
in
t
wo
col
um
ns
(tw
o
cha
nnel
s)
(ster
eo
signa
l),
the
se
val
u
es
usually
are
the
result
s
of
sam
pli
ng
and
quantiz
at
ion
of
the
or
igi
n
al
ana
logu
e
voi
ce
s
igna
l
.
In
th
is
pa
per
we
wil
l
int
roduc
e
a
m
et
h
od
which
c
an
be
used
to
cr
ea
t
e
a
signat
ure
or
k
e
y
,
which
c
a
n
be
used
later
to
ide
ntif
y
or
re
c
ogniz
e
the
wav
e
fil
e
.
Th
e
propo
sed
m
et
hod
will
be
impleme
nte
d
and
te
st
ed
to
show
the
ac
c
ura
c
y
and
fl
exi
b
il
ity
of
thi
s
m
et
hod
.
Ke
yw
or
d
s
:
C
rest
f
act
or
D
ynam
ic
r
ang
e
F
eat
ur
es
W
ave
f
il
e
W
i
ndow
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ma
j
ed
O.
Al
-
D
wairi
,
Dep
a
rtm
ent
of
Com
m
un
ic
at
io
n
E
ngineeri
ng
Tech
no
l
og
y
,
Faculty
of E
ngineerin
g
T
ech
nolo
gy
,
Al
-
Ba
lqa
A
pp
l
ie
d
U
niv
e
rsity
,
P.O. Bo
x:
15008,
Amm
an
11
134
,
Jor
dan
.
Em
a
il
:
m
ajed
dw@
ba
u.
e
du.jo
1.
INTROD
U
CTION
A
ND R
E
LATE
D
W
O
RK
Digital
au
dio
i
s
one
of
the
m
os
t
im
po
rtant
ty
pes
of
data
at
pr
ese
nt.
It
is
use
d
in
se
ve
ral
app
li
cat
io
ns
,
su
c
h
as
hum
an
know
le
dg
e
a
nd
m
any
secur
it
y
and
ba
nkin
g
app
li
cat
io
ns
.
A
dig
it
a
l
vo
ic
e
s
ign
al
is
usual
ly
of
a
la
rg
e
siz
e
wher
e
the
acou
sti
c
s
ign
al
co
ns
ist
s
of
a
set
of
valu
es
distrib
uted
in
one
colum
n
(
on
e
c
hannel)
(
m
on
o
sign
al
)
or
dist
r
ibu
te
d
in
tw
o
colum
ns
(two
channels)
(ster
eo
sig
nal),
the
se
values
usua
ll
y
are
the
resu
lt
s
of
sam
p
li
ng
and
quantiz
at
io
n of
t
he ori
gin
al
a
na
logue
vo
ic
e
sig
nal [1,
2
]
.
Since
the
vol
um
e
of
t
he
a
ud
i
o
file
is
la
rg
e
,
[
3,
4]
it
is
di
ff
ic
ult
to
c
onduct
t
he
m
at
ching
of
two
voic
es
us
in
g
al
l
the
va
lues,
w
he
re
the
process o
f
m
atch
in
g
will
requ
ire
a
la
rg
e
am
ou
nt o
f
ti
m
e,
wh
ic
h
in
tur
n
le
ad
s
to
delay
in
the
process
of
sound
reco
gnit
io
n
[5
-
7].
Table
1
s
hows
the
re
su
lt
s
of
voic
e
m
at
c
hing
with
it
sel
f,
an
d
her
e
we
ca
n
see
that
the
big
ge
r
wav
e
file
siz
e
will
increase
the
m
at
ching
tim
e,
and
the
process
of
m
atch
in
g
requires
a
big
a
m
ou
nt of tim
e
[
8
, 9]
.
To
dec
rease
th
e
reco
gnit
io
n
tim
e
[1
0],
we
ha
ve
to
seek
a
m
et
ho
d
bas
ed
on
feat
ur
es
e
xtracti
on,
this
m
et
ho
d
will
ge
ner
at
e
a
set
of
featur
e
s
f
or
an
y
wav
e
file
,
thi
s
set
m
us
t
be
a
un
iq
ue
a
nd
ca
n
be
us
ed
as
a
key
or
a
vo
ic
e
si
gn
at
u
re
t
o
retrie
ve
or
recog
nize
the
wa
ve
file
.
An
y
norm
al
iz
e
d
wa
ve
file
can
be
re
pr
ese
nt
ed
by
a
sinu
s
oid
al
sig
nal
as
show
n
in
F
ig
ur
e
1.
[
1,
3],
this
sig
nal
can
c
har
ac
te
rize
by
the
f
ollow
i
ng
pa
ra
m
et
ers:
a
m
plit
ud
e,
f
re
qu
e
ncy
an
d
ph
ase
sh
ifti
ng.
I
f
the
featur
es
a
re
base
d
on
th
ese
par
am
et
ers,
to
any
chang
es
on
these
par
am
et
e
rs
m
us
t n
ot aff
ect
the e
xtracted
vo
ic
e
featu
re
s.
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
meth
od
for v
oice sig
nal f
eatures
c
reat
ion
(M
ajed O.
Al
-
Dw
airi)
4093
Table
1.
Mat
c
hi
ng
ti
m
e fo
r
d
if
fer
e
nt w
a
ve fil
es
W
av
e f
ile
File size
(ele
m
en
ts
)
Nu
m
b
e
r
o
f
valu
es
Matchin
g
ti
m
e
(se
co
n
d
s)
W1
3
6
7
8
7
×2
7
3
5
7
4
0
.00
6
0
0
0
W2
3
9
7
3
0
×2
7
9
4
6
0
0
.00
8
0
0
0
W3
3
3
8
4
4
×2
6
7
6
8
8
0
.00
76
00
W4
1
7
6
5
8
×2
3
5
3
1
6
0
.00
5
0
0
0
W5
4
1
2
0
2
×2
8
2
4
0
4
0
.00
7
9
0
0
W6
3
6
7
8
7
×2
7
3
5
7
4
0
.00
6
0
0
0
W7
6
3
2
7
4
×2
1
2
6
5
4
8
0
.01
4
0
0
0
W8
4
8
0
4
9
×2
9
6
0
9
8
0
.01
0
0
0
0
W9
5
5
9
1
6
×2
1
1
1
8
3
2
0
.01
3
0
0
0
W
1
0
8
9
7
6
0
×2
1
7
9
5
2
0
0
.01
9
0
0
0
Av
erage
9
2
6
0
1
0
.00
9
7
Co
st o
f
1 v
alu
e
9
7
0
0
/
9
2
6
0
1
=0
.104
8
m
icros
econ
d
s
Figure
1. Sin
usoidal si
gn
al
Ma
ny
aut
hors
pro
posed
s
om
e
te
chn
iq
ues
of
vo
ic
e
fea
tures
e
xtracti
on
base
d
on
c
al
culat
ion
:
Crest
factor,
dynam
ic
ran
ge
,
sigm
a
(
m
ea
n
of
the
no
rm
al
iz
ed
data),
a
nd
M
u
(stan
da
rd
dev
ia
ti
on
of
the
norm
al
iz
ed
dat
a).
[
11, 1
2]. T
he
crest f
act
or
[4] i
s the r
at
io o
f
peak
value
to
RM
S v
al
u
e o
f
wav
e
f
or
m
as s
how
n
in
F
ig
ure
2.
T
hi
s r
at
io is als
o ca
ll
ed
to
pea
k
-
to
-
RM
S
rati
o.
Figure
2. Ca
lc
ulati
ng
c
rest
fa
ct
or
Dynam
ic
ran
ge
[4
-
6]
is
the
rati
o
between
t
he
la
rg
est
an
d
sm
a
ll
est
intensit
y
values
of
a
changeable
so
un
d
that
ca
n
be
reli
ably
tr
ansm
itted
or
r
epro
du
ce
d
by
a
par
ti
cula
r
sound
syst
em
,
m
easur
e
d
in
dec
ibels.
It’s
the
m
easur
e
m
ent
between
the
no
ise
fl
oor
and
the
m
axi
m
u
m
so
und
pr
essure
le
vel
an
d
w
hat
a
m
ic
ro
phone
can
captu
re.
In
[9
,
12
]
a
m
et
ho
d
was
pro
pos
ed
to
ge
ner
at
e
vo
ic
e
sig
nal
fe
at
ur
es
ba
se
on
the
above
-
m
entioned
par
am
et
ers,
an
y
changes
in
a
m
pl
it
ud
e,
fr
e
quency,
a
nd
pha
se
sh
i
ft
will
be
ref
le
ct
e
d
as
som
e
changes
i
n
vo
ic
e
sign
al
feat
ur
es
,
thu
s
will
le
ad
to
m
or
e
diff
ic
ul
ti
es
in
the
vo
i
ce
recogn
it
io
n
process
.
He
re
we
m
us
t
no
ti
ce
that
any
cha
nge
in
the
voic
e
pa
ram
et
ers
m
us
t
no
t
c
ha
ng
e
t
he
voic
e
featu
res.
F
or
exam
ple,
le
t
us
ta
ke
the
sinu
s
o
idal
sig
na
ls l
i
ste
d
in
T
a
ble 2 as s
how
n i
n
F
i
gure
3
.
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4092
-
4098
4094
Table
2.
Or
i
gina
l si
gn
al
Y
a
nd
so
m
e v
ersi
on
s
w
it
h
c
ha
ng
e
s i
n
am
plit
ud
e,
f
r
equ
e
ncy a
nd phase s
hiftin
g
Sig
n
al
Featu
res(x=
-
4
×π
:0
.00
1
:4
×π
)
Crest f
acto
r
(dB
)
Dy
n
a
m
i
c r
an
g
e
(d
B)
Sig
m
a
Mu
=
5
sin
(
10
+
10
)
3
.01
0
3
8
2
.29
7
5
3
.53
5
6
-
2
.78
4
6
e
-
005
1
=
15
sin
(
10
+
10
)
3
.01
0
3
8
2
.29
7
5
1
0
.60
6
8
-
8
.35
3
9
e
-
005
2
=
5
sin
(
20
+
10
)
3
.01
0
3
9
0
.41
5
9
3
.53
5
6
-
2
.76
8
6
e
-
005
3
=
5
sin
(
10
+
40
)
3
.01
0
3
8
2
.29
7
5
3
.53
5
6
-
3
.04
5
3
e
-
006
Figure
3.
Cha
ngin
g
the
sig
nal
par
am
et
ers
We
can
see
fro
m
the
resu
lt
s
sh
ow
n
in
T
able
2,
that
so
m
e
par
am
et
ers
re
m
ain
the
sam
e,
an
d
the
oth
e
rs
change
d,
th
us
t
he
feat
ur
es
c
on
sist
ing
of
a
set
of
t
hese
par
a
m
et
ers
will
al
s
o
cha
nge.
If
w
e
record
t
he
previo
us
sinu
s
oid
al
sig
na
ls
and
create
a
wav
e
file
usi
ng
var
i
ou
s
s
a
m
pling
f
requ
encies,
the
n
a
pp
ly
in
g
the
pr
opos
e
d
m
et
ho
d
i
n [13] w
e
receive
the
r
es
ults sh
own
in
T
a
ble
s
3 t
hr
u 6
.
Table
3.
Wa
ve fil
es f
eat
ures
usi
ng
s
am
pling
fr
e
qu
e
ncy=1
00
0
Sig
n
al
Featu
res
Crest f
acto
r
(dB
)
Dy
n
a
m
i
c r
an
g
e
(d
B)
Sig
m
a
Mu
Y
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
2
.35
1
5
e
-
005
Y1
0
.12
4
8
4
5
8
.71
3
3
0
.98
5
7
5
-
2
.53
6
4
e
-
005
Y2
0
.38
6
9
9
7
6
.32
9
6
0
.95
6
4
4
-
2
.35
2
4
e
-
005
Y3
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
3
.04
5
3
e
-
006
Table
4.
Wa
ve fil
es f
eat
ures
usi
ng
s
am
pling
fr
e
qu
e
ncy=2
00
0
Sig
n
al
Featu
res
Crest f
acto
r(
d
B)
Dy
n
a
m
i
c r
an
g
e(dB
)
Sig
m
a
Mu
Y
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
2
.35
1
5
e
-
005
Y1
0
.12
4
8
4
5
8
.71
3
3
0
.98
5
7
5
-
2
.53
6
4
e
-
005
Y2
0
.38
6
9
9
7
6
.32
9
6
0
.95
6
4
4
-
2
.35
2
4
e
-
005
Y3
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
3
.04
5
3
e
-
006
Table
5.
Wa
ve fil
es f
eat
ures
usi
ng
s
am
pling
fr
e
qu
e
ncy=3
00
0
Sig
n
al
Featu
res
Crest
f
acto
r
(dB
)
Dy
n
a
m
i
c r
an
g
e
(d
B)
Sig
m
a
Mu
Y
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
2
.35
1
5
e
-
005
Y1
0
.12
4
8
4
5
8
.71
3
3
0
.98
5
7
5
-
2
.53
6
4
e
-
005
Y2
0
.38
6
9
9
7
6
.32
9
6
0
.95
6
4
4
-
2
.35
2
4
e
-
005
Y3
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
3
.04
5
3
e
-
006
0
1
2
3
x
1
0
4
-5
0
5
y
=
5
*
s
i
n
(
1
0
*
x
+
1
0
)
0
1
2
3
x
1
0
4
-
2
0
-
1
0
0
10
20
y
1
=
1
5
*
s
i
n
(
1
0
*
x
+
1
0
)
0
1
2
3
x
1
0
4
-5
0
5
y
2
=
5
*
s
i
n
(
2
0
*
x
+
1
0
)
0
1
2
3
x
1
0
4
-5
0
5
y
3
=
5
*
s
i
n
(
1
0
*
x
+
4
0
)
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
meth
od
for v
oice sig
nal f
eatures
c
reat
ion
(M
ajed O.
Al
-
Dw
airi)
4095
Table
6.
Wa
ve fil
es f
eat
ures
usi
ng
s
am
pling
fr
e
q
ue
ncy=4
00
0
Sig
n
al
Featu
res
Crest f
acto
r(
d
B)
Dy
n
a
m
i
c r
an
g
e(dB
)
Sig
m
a
Mu
Y
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
2
.35
1
5
e
-
005
Y1
0
.12
4
8
4
5
8
.71
3
3
0
.98
5
7
5
-
2
.53
6
4
e
-
005
Y2
0
.38
6
9
9
7
6
.32
9
6
0
.95
6
4
4
-
2
.35
2
4
e
-
005
Y3
0
.38
6
9
9
6
8
.03
0
1
0
.95
6
4
4
-
3
.04
5
3
e
-
006
Fr
om
the r
e
su
lt
s sho
wn in t
hes
e table
s
we
ca
n raise
the
foll
owin
g fact
s:
-
Fo
r
the
sam
e file
the ex
t
racted
f
eat
ures
us
in
g vari
ou
s
sam
pli
ng freq
ue
ncies
rem
ai
n
the sam
e.
-
T
he
featu
res
of
the
wa
ve
file
are
change
d
wh
e
n
ad
j
us
ti
ng
the
si
nu
s
oi
dal
sign
al
pa
r
a
m
et
ers
(a
m
plit
ud
e,
fr
e
qu
e
ncy
an
d
ph
a
se
sh
ifti
ng
)
.
Table
7
s
how
s
the
featu
res
of
dif
fer
e
nt
wa
ve
file
s
with
dif
fer
e
nt
par
am
et
ers,
each
of them
is a uniq
ue se
t a
nd it
can be
use
d
to
ide
ntify t
he
s
pecific
wa
ve fil
e w
it
h
s
pe
ci
fic p
ar
am
et
er
s.
Table
7.
Wa
ve fil
es f
eat
ures
Wa
v
e f
ile
Sa
m
p
lin
g
f
requ
en
cy
Featu
res
Crest f
acto
r
Dy
n
a
m
i
c r
an
g
e
Sig
m
a
Mu
W1
4
4
1
0
0
1
4
.04
2
2
8
3
.93
4
6
0
.09
5
3
5
2
-
1
.24
4
1
e
-
005
W2
4
4
1
0
0
1
5
.96
5
4
8
2
.94
6
1
0
.06
8
1
7
-
5
.86
9
2
e
-
006
W3
4
4
1
0
0
1
5
.90
1
7
8
4
.52
9
4
0
.08
2
4
0
3
-
5
.75
9
2
e
-
006
W4
4
4
1
0
0
1
8
.05
4
7
7
6
.67
8
2
0
.02
6
0
5
3
-
3.
9937e
-
005
W5
4
4
1
0
0
1
8
.78
2
8
2
.72
4
3
0
.04
8
0
4
7
-
1
.89
0
1
e
-
005
W6
4
4
1
0
0
1
6
.82
0
2
7
5
.46
4
0
.02
6
1
0
6
-
9
.44
8
9
e
-
006
W7
4
4
1
0
0
1
6
.84
2
3
7
9
.28
0
5
0
.04
0
4
1
-
1
.08
9
5
e
-
005
W8
4
8
0
0
0
1
7
.50
7
3
7
9
.39
2
0
.03
7
9
1
3
-
1
.47
4
8
e
-
005
W9
4
4
1
0
0
1
6
.72
6
5
7
9
.13
5
9
0
.04
0
2
7
4
-
1
.90
3
1
e
-
005
W
1
0
4
4
1
0
0
1
9
.08
3
7
8
1
.89
1
5
0
.04
2
1
6
4
-
3
.30
4
e
-
005
In
[
14]
a
m
et
h
od
f
or
voic
e
fe
at
ur
e
extr
act
io
n
was
propose
d,
this
m
et
ho
d
is
based
local
b
ina
ry
patte
rn
to
fin
d
the
re
petit
ion
of
the
values
0,
1,
2
an
d
3.
This
m
et
ho
d
is
ve
ry
eff
ect
ive
in
creati
ng
a
wa
ve
file
sign
at
u
re,
bu
t
each
re
pleti
on
value
is
a
big
num
ber
and
it
will
be
increased
wh
e
n
the
wave
file
siz
e
increased.
The
ob
ta
ine
d
voic
e
featu
res
c
an
be
us
e
d
la
te
r
to
re
co
gniz
e
the
voic
e,
a
nd
the
voic
e
feat
ures
can b
e
pa
sse
d
to a
recog
nizer t
oo
l
capa
ble to
proc
ess an
y
a
ppli
cat
ion
relat
ed
to
voice p
r
ocessi
ng [1
5,
16]
.
Our
pa
per
will
fo
c
us
in
buil
ding
an
e
ff
eci
et
al
go
rithm
t
o
create
a
uniqu
e
featu
res
a
rr
ay
f
or
eac
h
wav
e
file
,
this
featur
es
ar
ray
can
be
us
e
d
as
a
sign
at
ur
e
to
recogn
iz
e
or
retrieve
a
w
ave
file
.
The
create
d
sign
at
ur
e
will
r
e
m
ai
n
the
sa
m
e
fo
r
a
wa
ve
file
with
def
e
ren
t
sa
m
pling
fr
e
quency,
this
wil
l
red
uce
the
m
e
m
or
y
sp
ace
rquires
for st
ori
ng
wave fil
es.
2.
THE
PROPO
SED
METHO
D
The
pro
po
se
d
i
n
this
pa
per
m
et
hod
us
es
a
n
al
gorithm
wh
ic
h
is
based
on
div
idi
ng
the
w
ave
file
int
o
windows
wit
h
fixe
d
num
ber
of
values,
t
his
al
gorithm
can
be
i
m
ple
m
ented
ap
plyi
ng
the
fo
ll
owin
g
seq
uen
c
e
of steps:
-
Get the
dig
it
al
wav
e
f
il
e.
-
Re
sh
ape
the
w
ave
file
(
m
on
o
or stere
o vo
ic
e
)
int
o on
e
row.
-
In
it
ia
li
ze
a
4
el
e
m
ent
feature
s
victo
r
to
ze
r
os
(F
i
rst
el
em
ent
po
i
nts
to
the
re
petit
ion
of
zer
os
,
sec
on
d
el
e
m
ent
po
i
nts
to
the
re
pleti
on
of
ones,
t
hi
rd
el
em
ent
points
to
t
he
repl
et
ion
of
t
wos,
an
d
th
e
f
our
t
h
el
e
m
ent p
oi
nts
to the re
pleti
on of th
rees).
-
Divid
e
voice
r
ow of
values
in
to w
i
ndows
wi
th f
i
xe
d si
ze
of voice
v
al
ues.
-
Sele
ct
the w
i
ndow size
and
num
ber
of
wind
ow
s
-
Wh
il
e
no
t
of e
nd w
i
ndows
do
-
Fo
r
eac
h win
dow
f
i
nd the a
ve
rag
e
of the
f
i
r
st half a
v0, a
nd
the a
ver
a
ge o
f
the sec
ond hal
f
a
v1.
-
Com
par
e
av0
with
the
value
in
the
center
of
the
fir
st
half,
if
av
0
is
gr
e
at
er
or
eq
ual
to
the
center
va
lue
m
ake b0=1, els
e m
ake b
0=
0.
-
Com
par
e
av1
with
the
value
in
the
center
of
the
sec
ond
ha
lf,
if
av
1
is
gr
eat
er
or
e
qual
to
the
ce
nter
va
lue
m
ake a1
=
1,
el
se m
ake a1
=0
.
-
Convert t
he bi
nar
y
num
ber
a1
a0
to deci
m
al
d.
-
Add 1
t
o feat
ures ar
ray el
em
e
nt w
it
h i
ndex=
d
-
endw
hile
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4092
-
4098
4096
Figure
4
s
how
s
an
exam
ple
of
how
to
cal
culat
e
the
rep
et
i
ti
on
f
or
the
fir
st
window
with
siz
e
equ
al
te
n
val
ues
.
Figure
4.
Ca
lc
ulati
on
e
xam
ple
3.
IMPLEME
N
TATION
AN
D
E
X
PE
RI
M
ENTAL
RES
ULTS
3.1.
Fir
st
e
xp
eri
m
ent
A
sin
us
oi
dal
sign
al
was
ta
ke
n,
the
pr
opos
e
d
m
e
tho
d
was
i
m
ple
m
ented
to
get
the
feat
ur
es
of
thi
s
sign
al
,
a
nd
w
e
cha
ng
e
one
of
the
si
gn
al
pa
ram
et
ers
(a
m
plit
ud
e,
f
reque
ncy
an
d
ph
as
e
sh
ifti
ng)
,
f
or
each
change
d
sig
na
l
we
cal
culat
e
the
sign
al
fe
at
ur
es,
the
res
ults
of
this
exp
e
rim
ent
are
li
ste
d
in
T
able
8
.
Fr
om
T
able
8
we
can
see
t
ha
t
al
l
the
fo
ur
sign
al
s
have
the
sa
m
e
featur
e
arr
ay
,
this
m
ea
ns
that
cha
nging
t
he
sign
al
par
am
et
er
do
es
not a
ffec
t t
he
voic
e
fe
at
ur
es.
Table
8.
E
xper
i
m
ent 1
r
es
ults
Sig
n
al
Featu
res(x=
-
4
×π
:0
.00
1
:4
×π
)
=
5
sin
(
10
+
10
)
4
1255
1254
0
1
=
15
sin
(
10
+
10
)
4
1255
1254
0
2
=
5
sin
(
20
+
10
)
4
1255
1254
0
3
=
5
sin
(
10
+
40
)
4
1255
1254
0
3.2.
Second e
xp
eri
ment
In
this
ex
per
i
m
ent
we
reco
r
ded
the
pr
e
vious
f
our
sig
nals
as
a
wav
e
file
us
ing
diff
e
r
e
nt
sam
pling
fr
e
qu
e
ncies;
th
e res
ults o
f
this
experim
ent ar
e li
ste
d
in
T
abl
e
s
9 t
hru 1
2.
Table
9.
Feat
ur
es w
it
h sam
pli
ng freq
ue
ncy=100
0
W
av
e f
ile
Featu
res
Y
2136
172
192
13
Y1
2350
63
87
13
Y2
2094
177
218
24
Y3
2137
171
191
14
Table
10. F
eat
ur
es
w
it
h sam
pling
fr
e
qu
e
ncy
=200
0
W
av
e f
ile
Featu
res
Y
2136
172
192
13
Y1
2350
63
87
13
Y2
2094
177
218
24
Y3
2137
171
191
14
Table
11. F
eat
ur
es
w
it
h sam
pling
fr
e
qu
e
ncy=3
00
0
W
av
e f
ile
Featu
res
Y
2136
172
192
13
Y1
2350
63
87
13
Y2
2094
177
218
24
Y3
2137
171
191
14
Table
12. F
eat
ur
es
for
W
a
ve fil
e:
Y
with
dif
fer
e
nt
sam
pling
f
re
qu
encies
Sa
m
p
lin
g
f
requ
en
cy
Featu
res
1000
2136
172
192
13
2000
2136
172
192
13
3000
2136
172
192
13
4000
2136
172
192
13
Fr
om
the r
e
su
lt
s of e
xp
e
rim
ent 2
we
ca
n raise
the
fo
ll
owin
g fact
s:
-
Fixin
g
the
pa
r
a
m
et
ers
of
digi
ta
l
sign
al
an
d
recor
ding
it
with
var
io
us
s
a
m
pling
fr
e
quencies
will
ke
ep
th
e
featur
e
s
of
t
he
wav
e
sig
nal
without any c
ha
nge.
-
Chan
ging a
ny
vo
ic
e
pa
ram
et
e
r
a
nd r
ec
ord
it
with a
ne
w
sa
m
pl
ing
fr
e
quen
cy
w
i
ll
ch
an
ge
the voice
featu
res.
3.3.
Ther
ed Ex
per
im
ent
Diff
e
re
nt
wa
ve
file
s
wer
e
t
aken
a
nd
trea
t
ed
usi
ng
the
pro
po
se
d
m
et
ho
d
,
T
a
ble
13
sh
ows
s
om
e
sa
m
ple
resu
lt
s
of
this
ex
per
im
ent
.
From
the
resu
lt
s
sh
ow
n
in
T
able
13
,
w
e
can
see
that
the
set
of
each
fe
at
ur
e
valu
es
is
a
uni
qu
e
set
,
t
hu
s
w
e
can
us
e
this
s
et
as
a
key
or
s
ign
at
ure
to
retr
ie
ve
or
recog
ni
ze
the
desire
d
wav
e
file
.
From
the
ob
ta
ine
d
e
xper
i
m
ental
r
esults,
w
e ca
n rase t
he
foll
ow
i
ng f
ac
ts:
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
meth
od
for v
oice sig
nal f
eatures
c
reat
ion
(M
ajed O.
Al
-
Dw
airi)
4097
-
For
a
wav
e
f
il
e
the
create
d
featur
es
ar
ray
is
a
un
iqu
e
ar
ray,
thu
s
we
can
us
e
this
ar
r
ay
as
a
sign
at
ur
e
to
recog
nize the
f
il
e.
-
Cha
ng
i
ng
t
he
sam
pling
fr
e
quency
do
e
s
no
t
aff
ect
the
f
ea
tures
a
rr
ay
values,
t
hu
s
t
her
e
is
no
nee
d
to
store
extra
c
op
ie
s
(
with
de
fer
e
nt
s
a
m
pling
f
re
qu
e
ncies)
of
a
file
,
and
t
his
will
m
ini
m
iz
e
the
stora
ge
siz
e
re
quire
d
to sto
re th
e
w
a
ve fil
e d
at
a
bas
e.
-
Feat
ur
e
s
arr
a
y
is
a
si
m
p
le
data
structur
e
w
hich
co
ntains
only
four
va
lues
,
this
will
si
m
p
li
fy
the
arch
it
ect
ur
e
of the
recog
niti
on to
ol s
uch as
artifia
ci
al
n
e
ural
n
et
w
ork.
Table
13. E
xpe
rim
ent 3
r
es
ults
W
av
e f
ile
Size
Featu
res
W1
3
6
7
8
7
×2
585
2989
3250
533
W2
3
9
7
3
0
×2
1170
5978
6500
1066
W3
3
3
8
4
4
×2
1755
8967
9750
1599
W4
1
7
6
5
8
×2
2340
1
1
9
5
6
1
3
0
0
0
2132
W5
4
1
2
0
2
×2
2925
1
4
9
4
5
1
6
2
5
0
2665
W6
3
6
7
8
7
×2
3510
1
7
9
3
4
1
9
5
0
0
3198
W7
6
3
2
7
4
×2
4095
2
0
9
2
3
2
2
7
5
0
3731
W8
4
8
0
4
9
×2
4680
2
3
9
1
2
2
6
0
0
0
4264
W9
5
5
9
1
6
×2
5265
2
6
9
0
1
2
9
2
5
0
4797
W
1
0
8
9
7
6
0
×2
5850
2
9
8
9
0
3
2
5
0
0
5330
4.
CONCL
US
I
O
N
A
sim
ple
and
hi
gh
ly
accu
rate m
et
ho
d
was
pr
opos
e
d
to
creat
e
a
wa
ve
file
fe
at
ur
es,
w
hich
c
an
be
us
ed
as
a
wa
ve
file
key
or
sig
natu
re.
T
he
propos
ed
m
et
ho
d
wa
s
i
m
ple
m
ented
and
te
ste
d
us
ing
va
rio
us
wa
ve
file
and
it
was
sho
wn
f
r
om
the
ob
ta
ined
e
xp
e
ri
m
ental
resu
lt
s
that:
a)
The
cr
eat
ed
arr
ay
of
featur
e
s
is
a
uniqu
e
f
or
each
wa
ve
file
,
thu
s
it
can
be
us
e
d
as
key
to
identify
the
wa
ve
file
;
b)
F
or
a
recorde
d
wa
ve
file
with
di
f
fer
e
nt
sam
pling
fr
e
quencies
,
the
f
eat
ur
e
a
rr
ay
do
e
s
no
t
cha
ng
e
,
rem
ai
ns
the
sa
ve
us
in
g
var
i
ous
sa
m
pl
in
g
fr
e
qu
e
ncies,
th
us
m
ake
the
propose
d
m
et
hod
m
or
e
fle
xib
l
e,
an
d
e
ff
ic
ie
nt
by
re
duci
ng
t
he
r
eq
uire
d
process
or
tim
e and
m
e
m
or
y s
pace
nee
de
d for the
pr
oc
ess of
vo
ic
e
r
e
cogniti
on
REFERE
NCE
S
[1]
M
.
Guillem
ard
,
et
al.,
“
Signal
A
naly
s
is wi
th
Fra
m
e
The
or
y
and
Persistent
Hom
olog
y
,
”
Jun
2013.
[2]
Barke
r
J.
,
e
t
a
l.,
“
The
thi
rd
„
chim
e‟
spee
ch
sep
ara
t
ion
and
rec
o
g
nit
ion
cha
l
le
ng
e:
Dat
ase
t
,
t
ask
and
base
l
ine
s,
”
i
n
2015
IEEE
Wor
kshop
on
Au
tomati
c
Speech Re
c
ognit
ion
a
nd
Un
derstanding
(
ASR
U)
.
IEE
E
,
pp
.
504
-
511
,
2015
.
[3]
S.
K.
Dastoor,
“
Com
par
at
ive
Anal
y
s
is
of
Steganographic
Algor
it
hm
s
impact
ing
the
informat
ion
in
the
Spe
e
c
h
Signal
for
enhanci
ng
the
Mess
age
Se
cur
ity
in
next
Gene
r
at
io
n
Mobile
devic
es,
”
(
WICT2011)
,
2011
World
Congress
on
Inf
orm
ati
on
and
C
omm
unic
ati
on
T
ec
hnolog
ie
s.
IE
EE
,
pp.
279
-
284
,
2011
.
[4]
Bin,
et
al
.
,
“
D
y
n
amic
ran
g
e
estim
at
ion,
”
IEEE
T
rans
act
ions
on
Computer
-
Ai
ded
Design
of
In
te
g
rated
Circuits
a
nd
Syste
ms
,
pp
.
161
8
-
1636,
2006
.
[5]
B
.
W
u,
et
al.
,
“
An
ana
l
y
tical
appr
oac
h
for
d
ynamic
ran
g
e
est
imati
on,
”
in
AC
M/IE
E
E
41
Des
ign
Aut
omat
ion
Confe
renc
e
(
DAC
-
04)
,
San
Dieg
o,
Cal
if
.
,
Jun
20
04.
[6]
S
.
Mill
ers
and
D
.
Chi
lde
rs,
“
Prob
abi
lit
y
and
r
ando
m
proc
esses
,”
A
ca
demic
Press
,
pp
.
370
-
5
,
2012
.
[7]
F.
Gou
y
on,
e
t
al.,
“
Cla
ss
if
y
ing
p
erc
uss
ive
sounds
:
a
m
at
te
r
ofz
ero
-
cro
ss
ing
rat
e
in
Proce
edi
ngs
of
the
COS
T
G
-
6,
”
Confe
renc
e
on
Digit
al Audi
o
Effec
ts (
DAFX
-
00)
,
V
erona,
I
taly,
Dec
2000
.
[8]
D.
D.
Jaslen
e,
“
Feat
ure
Sel
ecti
on
and
Ex
traction
of
Audio,
”
I
nte
rnational
Jo
urnal
of
Innov
a
ti
v
e
R
ese
arch
i
n
Sci
en
ce,
Eng
ineering
and
Te
chn
ology
,
v
ol
/
issue:
5
(
3
)
,
Mar
2016
.
[9]
Sanji
vani
S
.
,
et
al.,
“
An
Over
vie
w
of
Te
chn
i
ca
l
Progr
ess
in
Speec
h
Rec
og
nit
ion,
”
In
te
rnat
ional
Journal
o
f
Adv
anc
ed Re
sea
rch
in
Comput
er
Scienc
e
and
Sof
tware
Eng
ine
eri
ng
,
vol
/i
ss
ue:
3
(
3
)
,
Mar
2013
.
[10]
M.
I.
Ouloul,
et
al
.
,
“
An
Eff
ic
i
ent
Fac
e
Rec
og
nit
ion
Us
ing
SIF
T
Descri
ptor
i
n
RGB
-
D
Im
ag
es
,
”
Int
ernati
on
al
Journal
of
Elec
t
rical
and
Computer
Eng
ine
ering
(
IJE
CE)
,
v
ol
/i
ss
ue:
5
(
6
)
,
pp
.
122
7
-
1233
,
De
c
201
5
.
[11]
S
.
Nara
ng
and
D
.
Gupta
,
“
Speec
h
Feat
ur
e
Ext
ra
ct
ion
Tech
nique
s:
A
Rev
i
ew,
”
I
JCSMC
,
v
ol
/i
ss
ue:
4
(
3
)
,
pp.
107
-
114
,
Ma
r
2015
.
[12]
R
y
ba
ch,
e
t
a
l.,
“
The
RW
TH
Aa
c
hen
Univer
si
t
y
Open
So
urc
e
Speec
h
Re
cogni
t
io
n
s
y
st
em,
”
Sep
2
009.
[13]
K
.
Matrouk
,
e
t
al.
,
“
Speec
h
F
inge
rprin
t
to
Id
ent
if
y
Isolated
W
ord
-
Person,
”
World
Appl
i
ed
Sci
en
ce
s
Journa
l
,
vol/
issue:
31(10)
,
pp
.
1767
-
1771
,
2014.
[14]
S
.
Khawat
reh
,
et
al
.
,
“
A
Novel
Methodol
og
y
to
Ext
ra
ct
Voic
e
Signal
Feat
ur
es,
”
I
nte
rnational
Jou
rnal
of
Computer
Appl
ic
a
ti
ons
,
vo
l
/i
ss
ue:
17
9
(
9
)
,
p
p.
0975
-
8887
,
Ja
n
2018.
[15]
E
.
Benmalek,
et
al
.
,
“
Voic
e
As
s
essm
ent
s
for
De
te
c
ti
ng
Patients
with
Parkinson’s
Disea
ses
in
D
iffe
ren
t
Stage
s
,
”
Inte
rnational
Jo
urnal
of El
e
ct
ri
c
al
and
Comput
er
Engi
n
ee
ring
, v
ol
/
issue:
8
(
6
)
,
pp.
4265
-
4271
,
De
c
2018
.
[16]
F
.
P
.
George
,
e
t
al
.
,
“
Rec
ogn
it
i
on
of
emotional
stat
es
using
EEG
signal
s
base
d
on
ti
m
e
-
fre
qu
e
nc
y
anal
y
s
is
an
d
SV
M
cl
assifie
r
,”
Int
ernati
onal
Journal
of
Elec
t
rical
and
Comp
ute
r
Engi
nee
rin
g
,
v
ol
/i
ss
ue:
9
(
2
)
,
pp.
1012
-
102
0
,
Apr
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.
9
, N
o.
5
,
Oct
ober
20
19
:
4092
-
4098
4098
BIOGR
AP
H
I
ES
OF
A
UTH
ORS
Maj
ed
O.
D
w
a
ir
i
,
i
n
comm
unic
ation
s
y
stems
,
was
born
on
10
of
dic
ember
1968
in
Jordan.
He
rec
e
ive
d
h
is
dip
loma
degr
e
e
in
1994
and
PhD
degr
ee
from
Ukrai
ne
st
ate
Aca
d
em
y
in
1998in
t
he
fie
ld
o
f
m
ult
i
ch
anne
l
comm
unic
at
io
n
.
An
associ
at
e
p
rofe
ss
or
in
the
d
epa
rtment
of
comm
unic
at
i
on
Engi
ne
eri
ng
Tec
hnolog
y
,
fac
u
lty
of
Engi
nee
ring
Te
chno
log
y
/Al
-
Bal
qa
Appli
ed
Univer
sit
y
Am
m
an
-
Jordan.
His
res
ea
rch
intere
sts
i
ncl
ude
optical
c
om
m
unic
at
ion
n
et
works
,
dig
it
a
l
comm
unic
at
ion
s,
signal
and
imag
e
pross
ec
ing, Ant
enna
d
esign, a
nd
m
ic
rostrip
p
atch
antenna
s.
Amjad
Y.
He
ndi
,
i
n
Radi
o
and
T
Vs
y
st
ems
.
He
rec
ei
v
ed
his di
plo
m
a
degr
ee
in
19
94
and
PhD
degr
ee
from
Ukrai
ne
stat
e
Ac
ade
m
y
in
1998in
the
f
ie
l
d
of
Radi
o
&
T
v
s
y
stems
.
His
rese
arc
h
intere
s
t
s
inc
lud
e
digi
t
al
comm
unic
at
io
ns,
signal
and
image
pross
ec
ing,
Antenna
d
esign,
opti
m
izat
ion
te
chn
ique
s in
an
t
enna
d
esign and ante
nn
a
m
ea
sure
m
ent
techniqu
es
and
m
ic
rostrip
p
at
ch
antenna
s.
Mohame
d
S.
Soliman
a
n
assista
nt
profe
ss
or
in
t
he
depa
r
tment
o
f
E
lectr
i
cal
Enginee
ring
,
Facu
lty
of
Ene
rg
y
Engi
ne
e
ring,
As
wan
Un
ive
rsit
y
,
Eg
y
p
t.
Curre
ntly
,
he
is
with
th
e
dep
artm
ent
of
Elec
tri
c
al
Engi
ne
eri
ng,
Fa
cul
t
y
of
Engi
n
e
eri
ng
T
ai
f
Univ
ersity
,
Saudi
Ar
abi
a
.
His
rese
ar
ch
intere
sts
in
clude
wire
le
ss
comm
unic
a
ti
ons,
pha
sed
and
ti
m
ed
arr
a
y
signa
l
proc
essing,
U
W
B
m
ic
rostrip
pat
ch
ant
enn
as,
d
ie
l
ectric
Resonant
ant
enn
as,
num
e
ric
a
l
m
e
thods
i
n
elec
tromagn
e
ti
cs,
opti
m
ization
te
chn
ique
s
in
an
te
nna
d
esign
an
d
ant
enn
a
m
ea
su
rement
techniqu
es.
Dr.
Solim
an
is
a
m
ember
of
t
he
IEE
E
-
AP
Soci
ety
,
KA
US
T
cha
p
t
er,
Saud
i
Arab
ia.
Z
iad
AL
-
Qadi,
in
computer
eng
ine
er
ing,
was
born
on
09
Marc
h
1955
in
Jor
dan.
He
rec
ei
v
ed
hi
s
dipl
om
a
degr
ee
i
n
1980
and
PhD
degr
ee
from
Ukrai
ne
in
1986
in
t
he
fie
ld
of
Com
pute
r
Engi
n
ee
rin
g.
Curre
ntly
he
is
a
Profess
or
at
the
Com
pute
r
Engi
ne
eri
ng
De
par
tment,
Facu
l
t
y
of
Engi
n
ee
ri
ng
Te
chno
log
y
,
Al
-
Bal
qa
Appl
ie
d
Univer
sit
y
,
Jordan
.
His
m
ai
n
int
er
est
include
signal
pro
ce
ss
i
ng,
pat
t
ern
r
ec
ogni
tion,
a
lgori
thms
,
m
odel
li
ng
and
si
m
ula
ti
ons.
Evaluation Warning : The document was created with Spire.PDF for Python.