T
E
L
KO
M
NIK
A
, V
ol
.
17
,
No.
4,
A
ug
us
t
20
1
9,
p
p.
1
91
4
~
1
92
2
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No: 2
1/E/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
4
.
12615
◼
19
14
Rec
ei
v
ed
J
ul
y
1
,
20
1
8
;
Rev
i
s
ed
F
e
bruar
y
23
,
20
1
9
;
A
c
c
ep
ted
Ma
r
c
h 2
7
, 2
01
9
FPG
A
-
ba
sed i
m
plem
e
nt
a
tio
n o
f
speec
h reco
gn
i
tio
n
fo
r
robo
c
ar co
nt
rol
usin
g MFC
C
Ba
y
u
aji
Ku
r
n
i
adh
ani
*
1
,
S
u
g
o
n
d
o
Hadi
y
o
so
2
,
S
u
ci
Au
lia
3
, Rita
M
agd
ale
n
a
4
1
,4
Sc
h
o
o
l
o
f
El
e
c
tr
i
c
a
l
En
g
i
n
e
e
ri
n
g
,
T
e
l
k
o
m
Un
i
v
e
rs
i
ty
,
T
e
ru
s
a
n
Bu
a
h
Ba
tu
,
Ba
n
d
u
n
g
,
In
d
o
n
e
s
i
a
2
,3
Sc
h
o
o
l
o
f
Ap
p
l
i
e
d
Sc
i
e
n
c
e
,
T
e
l
k
o
m
Un
i
v
e
rs
i
ty
,
T
e
ru
s
a
n
B
u
a
h
Ba
t
u
,
Ba
n
d
u
n
g
,
In
d
o
n
e
s
i
a
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r,
e
-
m
a
i
l
:
k
u
rn
i
a
d
h
a
n
i
.b
a
y
u
a
j
i
@gm
a
i
l
.c
o
m
1
,
s
u
g
o
n
d
o
@t
e
l
k
o
m
u
n
i
v
e
rs
i
t
y
.a
c
.i
d
2
,
s
u
c
i
a
u
l
i
a
@te
l
k
o
m
u
n
i
v
e
rs
i
ty
.a
c
.i
d
3
,
r
i
ta
m
a
g
d
a
l
e
n
a
@te
l
k
o
m
u
n
i
v
e
rs
i
ty
.a
c
.
i
d
4
Ab
strac
t
Th
i
s
r
e
s
e
a
rc
h
p
ro
p
o
s
e
s
a
s
i
m
u
l
a
ti
o
n
o
f
th
e
l
o
g
i
c
s
e
r
i
e
s
o
f
s
p
e
e
c
h
r
e
c
o
g
n
i
t
i
o
n
o
n
th
e
M
FCC
(M
e
l
Fre
q
u
e
n
c
y
Sp
re
a
d
Sp
e
c
t
ru
m
)
b
a
s
e
d
FPGA
a
n
d
E
u
c
l
i
d
e
a
n
Di
s
ta
n
c
e
t
o
c
o
n
tro
l
th
e
r
o
b
o
ti
c
c
a
r
m
o
t
i
o
n
.
Th
e
s
p
e
e
c
h
k
n
o
wn
wou
l
d
b
e
u
s
e
d
a
s
a
c
o
m
m
a
n
d
t
o
o
p
e
r
a
te
t
h
e
r
o
b
o
t
i
c
c
a
r.
M
FCC
i
n
t
h
i
s
s
tu
d
y
wa
s
u
s
e
d
i
n
t
h
e
f
e
a
tu
r
e
e
x
tra
c
ti
o
n
p
ro
c
e
s
s
,
whi
l
e
Eu
c
l
i
d
e
a
n
d
i
s
t
a
n
c
e
w
a
s
a
p
p
l
i
e
d
i
n
th
e
fe
a
tu
re
c
l
a
s
s
i
fi
c
a
t
i
o
n
p
ro
c
e
s
s
o
f
e
a
c
h
s
p
e
e
c
h
t
h
a
t
l
a
te
r
wo
u
l
d
b
e
fo
rw
a
rd
e
d
t
o
th
e
p
a
rt
o
f
d
e
c
i
s
i
o
n
to
g
i
v
e
t
h
e
c
o
n
tro
l
l
o
g
i
c
i
n
ro
b
o
ti
c
m
o
to
r.
Th
e
te
s
t
th
a
t
h
a
s
b
e
e
n
c
o
n
d
u
c
te
d
s
h
o
wed
t
h
a
t
t
h
e
l
o
g
i
c
s
e
ri
e
s
d
e
s
i
g
n
e
d
wa
s
p
re
c
i
s
e
h
e
re
b
y
m
e
a
s
u
ri
n
g
th
e
M
e
l
Fre
q
u
e
n
c
y
W
a
r
p
i
n
g
a
n
d
P
o
wer
Cep
s
tru
m
.
W
i
t
h
th
e
a
c
h
i
e
v
e
m
e
n
t
o
f
l
o
g
i
c
d
e
s
i
g
n
i
n
th
i
s
re
s
e
a
r
c
h
p
ro
v
e
n
wit
h
a
c
o
m
p
a
ri
s
o
n
b
e
twe
e
n
th
e
M
a
tl
a
b
c
o
m
p
u
ta
ti
o
n
a
n
d
X
i
l
i
n
x
s
i
m
u
l
a
ti
o
n
,
i
t
e
n
a
b
l
e
s
t
o
fa
c
i
l
i
ta
te
t
h
e
re
s
e
a
r
c
h
e
r
s
t
o
c
o
n
ti
n
u
e
i
t
s
i
m
p
l
e
m
e
n
ta
t
i
o
n
to
F
PG
A h
a
rd
ware
.
Key
w
ords
:
E
u
c
l
i
d
e
a
n
d
i
s
ta
n
c
e
,
FPGA
,
M
FCC, s
p
e
e
c
h
re
c
o
g
n
i
ti
o
n
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
rig
ht
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
T
he
r
es
ea
r
c
h
on
A
ut
om
ati
c
S
p
ee
c
h
R
ec
og
n
i
t
i
on
(
A
S
R
)
i
s
s
ti
l
l
t
he
m
os
t
i
nte
r
es
t
to
pi
c
f
or
i
ts
v
er
y
i
m
po
r
tan
t
r
o
l
e
i
n
d
a
i
l
y
l
i
f
e
s
uc
h
as
i
n
t
he
nu
m
be
r
of
w
ork
s
tha
t
ha
v
e
b
ee
n
c
on
d
uc
ted
on
S
R
(
S
pe
ec
h
R
ec
og
n
i
ti
on
)
s
uc
h
as
S
m
art
Hom
e
[1
-
4]
,
A
r
ti
f
i
c
i
a
l
I
nte
l
l
i
ge
nc
e
s
uc
h
as
hu
m
an
em
oti
on
al
c
l
as
s
i
f
i
c
ati
o
n
b
as
ed
on
s
p
ee
c
h
r
ec
og
ni
t
i
on
[
5
-
9]
an
d
r
ob
oti
c
a
p
pl
i
c
a
ti
o
n
[
10
],
i
n
t
he
f
i
el
d
of
S
tu
d
en
t
Le
arni
n
g
[
11
-
1
3
]
,
i
n
m
ed
i
c
a
l
s
e
c
tor,
S
R
i
s
us
ed
to
de
t
ec
t
t
he
s
tr
es
s
l
ev
el
of
a
p
ers
on
[1
4
].
T
he
r
ef
ore
,
th
e
op
t
i
m
i
z
ati
on
proc
es
s
i
n
S
R
i
s
s
t
i
l
l
i
n
th
e
progr
es
s
f
or
ob
ta
i
n
i
n
g
the
be
tte
r
ac
c
urac
y
s
uc
h
as
wi
th
s
pe
ec
h
en
ha
nc
e
m
en
t
[1
5
-
1
7
]
or
no
i
s
e
r
ed
uc
ti
on
[
1
8
-
20
].
A
uto
m
ati
c
S
pe
ec
h
Rec
og
ni
ti
on
(
A
S
R)
i
s
a
s
i
gn
a
l
r
ec
o
gn
i
t
i
o
n
proc
es
s
of
s
pe
ec
h
t
o
b
e
a
nu
m
be
r
of
w
ord
order
s
.
T
hi
s
c
on
c
ep
t
h
as
be
en
de
v
el
op
e
d
s
i
nc
e
20
00
s
tart
ed
f
r
om
the
em
ergenc
e
of
Hi
dd
en
M
ark
ov
Mo
d
el
-
ba
s
ed
CMU
S
p
hi
nx
-
N,
G
o
og
l
e
V
oi
c
e
R
ec
og
n
i
t
i
o
n
i
n
A
nd
r
oi
n
i
n
2
01
0.
I
n
20
12
A
p
pl
e
r
el
ea
s
ed
i
ts
a
pp
l
i
c
ati
on
na
m
ed
S
i
r
i
[
21
]
.
S
i
nc
e
the
en
d
of
20
1
7,
th
ere
ha
v
e
be
e
n
m
an
y
a
pp
l
i
c
a
bl
e
ap
pl
i
c
at
i
o
ns
of
s
pe
ec
h
r
ec
o
gn
i
ti
on
al
o
ng
wi
th
th
e
i
nc
r
ea
s
i
ng
us
e
of
s
pe
ec
h
tec
hn
o
l
og
y
i
n re
al
l
i
f
e e
nv
i
r
on
m
en
t.
O
f
s
o
m
e
ap
p
l
i
c
at
i
on
s
of
A
S
R
a
bo
v
e,
s
pe
ec
h
f
ea
tu
r
es
tha
t
are
f
r
eq
ue
nt
l
y
us
ed
i
n
the
ex
tr
ac
ti
on
proc
es
s
are
[
22,
2
3
]
LP
C
(
L
i
n
ea
r
P
r
ed
i
c
t
i
v
e
Cod
es
)
,
MFCC
(
Me
l
F
r
eq
ue
nc
y
Ceps
tr
al
Co
ef
f
i
c
i
en
ts
)
,
P
L
P
(
P
erc
ep
tu
al
L
i
ne
ar
P
r
e
di
c
ti
on
)
,
an
d
P
LP
-
R
A
S
T
A
(
P
L
P
-
Re
l
at
i
v
e
S
pe
c
tr
a).
A
tec
h
ni
c
a
l
r
e
v
i
e
w
t
o
o
bs
erv
e
th
e
pe
r
f
orm
an
c
e
of
the
s
pe
ec
h
f
ea
ture
ex
tr
ac
t
i
on
tec
hn
i
qu
es
(
MFCC
,
LP
C,
P
L
P
,
G
F
CC)
w
i
th
the
c
o
m
bi
na
ti
on
of
i
ts
c
l
as
s
i
f
i
c
at
i
on
tec
h
ni
q
ue
(
DT
W
,
HMM,
ML
P
,
S
V
M,
a
nd
DT
)
ha
v
e
b
ee
n
tes
ted
f
or
S
R
T
am
i
l
S
po
k
en
w
ords
b
y
V
i
m
al
a
[2
4
]
.
B
as
ed
u
po
n
t
he
r
es
u
l
t
of
the
tes
t,
of
5
(
f
i
v
e)
v
ari
et
i
e
s
of
the
m
eth
od
of
f
ea
ture
ex
tr
ac
ti
on
a
nd
c
l
as
s
i
f
i
c
ati
on
,
G
F
CC
m
eth
od
ha
s
b
ee
n
m
ore
ex
c
el
l
e
nt
i
n
c
om
pa
r
i
s
on
t
o
oth
er
al
g
orit
hm
s
[2
5
].
G
urban
h
as
al
s
o
c
on
du
c
te
d
an
ap
proac
h
of
the
MFC
C
ba
s
ed
a
ud
i
o
v
i
s
ua
l
S
R
an
d
c
on
du
c
te
d
the
op
t
i
m
i
z
ati
on
w
i
th
t
w
o
m
eth
od
s
:
CMI
(
Con
di
t
i
o
na
l
Mu
t
ua
l
I
nf
or
m
ati
on
)
al
go
r
i
thm
an
d
MIFS
(
Mu
tua
l
Inf
orm
ati
on
F
ea
t
ure
S
e
l
ec
ti
on
)
a
l
go
r
i
thm
.
T
he
r
es
ul
t
w
as
f
ou
nd
at
be
s
t
i
n
v
er
y
n
oi
s
y
c
on
di
t
i
o
ns
.
A
no
t
he
r
s
tu
d
y
on
S
R
w
as
c
o
nd
uc
t
ed
b
y
G
u
pt
a
[2
6
]
us
i
ng
L
P
C
an
d
LP
CC
as
the
f
ea
ture
ex
tr
ac
ti
on
tec
h
ni
q
ue
s
w
i
t
h
a
r
es
ul
t
s
ho
wi
ng
tha
t
i
n
term
s
o
f
s
pe
ec
h
i
de
nt
i
f
i
c
ati
on
,
the
LP
C
p
aram
ete
r
was
m
ore
prec
i
s
e
c
om
pa
r
ed
t
o
L
P
CC.
O
n
t
he
oth
er
h
an
d,
i
n
r
el
i
a
bi
l
i
t
y
an
d
r
ob
us
tne
s
s
,
L
P
CC
w
as
m
ore
ex
c
el
l
en
t.
R
es
ea
r
c
h
s
tud
y
grou
p
on
S
pe
ec
h
a
nd
La
ng
u
ag
e
T
ec
hn
ol
og
y
f
or
three
y
ea
r
s
s
uc
c
es
s
f
ul
l
y
di
d
a
r
es
e
a
r
c
h
on
ho
w
t
he
ef
f
ec
t
of
s
tr
es
s
w
as
on
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
F
P
G
A
-
ba
s
e
d i
mp
l
em
en
t
ati
on
of
s
pe
ec
h
r
ec
og
n
i
ti
on
f
o
r
r
ob
oc
ar c
on
tr
o
l
...
(
B
ay
ua
j
i
K
urn
i
ad
ha
n
i
)
1915
the
s
pe
ec
h
pro
du
c
ti
on
wi
th
the
a
pp
r
oa
c
he
s
of
HMM
a
nd
MFCC
[2
7
].
O
the
r
S
R
was
wi
th
hi
dd
e
n
Ma
r
c
ov
m
od
el
tha
t
i
s
t
he
e
x
pe
r
i
m
e
nt
c
on
du
c
te
d
b
y
F
a
r
s
i
[2
8
],
the
r
es
ul
t
s
h
o
w
e
d
tha
t
t
he
H
MM
proc
es
s
ne
e
de
d
to
s
t
i
l
l
b
e
do
ne
w
i
th
G
A
(
G
en
eti
c
A
l
g
orit
hm
)
f
or
a
go
od
r
es
ul
t.
T
o
c
om
pa
r
e
[2
7
],
an
ot
he
r
s
tud
y
on
MFCC
c
on
du
c
t
ed
b
y
Mo
ha
n
[2
9
]
was
do
n
e
b
y
c
om
bi
ni
ng
MF
CC
w
i
t
h
D
TW
al
g
orit
hm
s
ho
wed
t
h
e
go
o
d
r
es
ul
ts
i
n
the
proc
es
s
of
S
R
w
i
th
te
n
f
ea
tures
f
or
ea
c
h
w
ord
i
n
tr
ai
n
i
ng
ph
as
e.
B
as
ed
on
the
c
om
pa
r
i
s
on
f
r
om
a
nu
m
be
r
of
s
tud
i
es
a
bo
v
e
(
MFCC
,
L
P
C,
P
L
P
,
G
F
CC)
to
th
i
s
s
tud
y
,
MFCC
was
us
ed
f
or
the
proc
es
s
of
f
ea
ture
ex
tr
ac
t
i
on
on
t
he
s
p
ee
c
h
s
i
g
na
l
as
c
o
m
m
on
l
y
i
t
ha
d
th
e
h
i
g
h
pe
r
f
orm
an
c
e
r
ate
a
nd
l
o
w
c
om
pl
ex
i
t
y
[
30
].
I
n
t
hi
s
r
es
ea
r
c
h,
a
protot
y
p
e
of
S
R
s
y
s
t
e
m
w
as
m
ad
e
to
F
P
G
A
t
ha
t
l
ate
r
w
o
ul
d
be
us
ed
as
an
i
np
ut
f
or
the
r
o
bo
t
i
c
c
ar
(
f
urther
r
es
ea
r
c
h).
O
ne
of
ex
am
pl
es
o
f
w
ord
r
ec
og
n
i
t
i
on
s
y
s
tem
ap
p
l
i
c
at
i
on
c
an
be
d
on
e
i
n
a
s
i
m
pl
e
r
ob
o
t
c
ar.
T
he
r
ob
ot
i
s
ab
l
e
t
o
di
f
f
erenti
a
te
t
he
wor
d
p
r
on
ou
nc
ed
b
y
the
us
ers
.
B
y
ap
p
l
y
i
n
g
t
he
w
ord
r
ec
og
n
i
ti
on
s
y
s
tem
i
n
t
he
r
ob
ot
i
c
c
ar,
the
n
t
h
e
us
ers
c
ou
l
d
c
on
tr
ol
th
e
d
i
r
ec
ti
on
of
r
o
bo
t
m
oti
o
n
wi
th
ou
t
a
ne
e
d
to
to
uc
h
t
h
e
bu
tto
n
or
be
i
ng
c
l
os
e
to
the
r
ob
ot.
2.
M
el
F
r
equ
enc
y
Cepst
r
u
m Co
eff
ici
ent
s (M
F
CC)
F
ea
ture
ex
tr
ac
ti
o
n
i
s
a
pro
c
es
s
to
de
t
erm
i
ne
a
v
al
ue
or
v
ec
t
or
tha
t
c
an
be
us
ed
as
an
ob
j
ec
t
or
i
nd
i
v
i
du
al
i
d
e
nti
f
i
er
th
at
s
ub
s
e
qu
e
ntl
y
wi
l
l
b
e
us
ed
i
n
the
c
l
as
s
i
f
i
c
ati
o
n
proc
es
s
[
31
,
32
].
MFCC
an
al
y
s
i
s
i
s
a
s
tan
da
r
d
m
eth
o
d
[3
3
]
us
ed
to
r
e
pres
en
t
t
h
e
pa
r
a
m
ete
r
o
f
s
ou
nd
s
i
gn
al
.
T
he
m
ec
ha
ni
s
m
of
MFCC
i
s
ba
s
ed
up
o
n
th
e
d
i
f
f
erenc
e
of
f
r
eq
ue
n
c
y
th
at
c
a
n
b
e
c
ap
tured
b
y
h
um
an
ea
r
s
c
o
m
m
on
l
y
s
tat
ed
i
n
th
e
s
c
al
e
of
Me
l
(
ori
gi
na
te
d
f
r
om
Me
l
od
y
)
i
n
whi
c
h
the
s
ou
nd
s
i
g
na
l
w
o
ul
d
be
f
i
l
tered
i
n
l
i
n
ea
r
i
n
Me
l
f
r
eq
ue
nc
y
s
c
a
l
e
th
at
i
s
f
or
the
l
o
w
f
r
eq
ue
nc
y
l
es
s
tha
n
1
K
H
z
an
d
l
og
arit
hm
i
c
al
l
y
f
or
hi
gh
f
r
eq
u
en
c
y
m
ore
tha
n
1
K
H
z
[3
4
]
.
T
he
bl
oc
k
di
ag
r
am
f
or the
proc
es
s
of
f
ea
ture e
x
tr
ac
ti
on
us
i
ng
MFCC
i
s
pr
es
en
te
d
i
n Fi
gu
r
e 1
.
F
i
gu
r
e
1.
MF
CC b
l
oc
k
di
ag
r
am
A
s
s
ho
w
n
i
n
F
i
g
ure
1
MFCC
c
on
s
i
s
ts
of
s
o
m
e f
ol
l
o
wi
ng
c
o
m
pu
tat
i
on
a
l
s
te
ps
:
S
tep
1:
P
r
ep
r
oc
es
s
i
ng
A
s
the
i
ni
t
i
a
l
s
tep
,
th
e
a
na
l
og
s
i
gn
a
l
i
s
pa
s
s
ed
thro
ug
h
HP
F
em
ph
as
i
z
es
or
k
no
wn
as
pre
-
em
ph
as
i
s
[3
5]
Its
pu
r
p
os
e
i
s
to
i
nc
r
ea
s
e
the
en
erg
y
of
s
i
gn
a
l
[3
6
]
th
us
,
i
ts
o
utp
ut
be
c
om
es
the
o
ne
as
i
n
(
1
)
:
Y
(
)
=
(
)
−
∝
(
−
1
)
(
1
)
ℎ
0
.
9
≤
∝
≤
1
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
, N
o.
4
,
A
u
gu
s
t
2
01
9
:
1
9
14
-
1
922
1916
S
tep
2:
F
r
am
e B
l
oc
k
i
ng
an
d
W
i
nd
ow
i
ng
S
pe
ec
h
s
i
gn
a
l
e
nte
r
s
t
o
th
e
proc
es
s
of
s
ho
r
t
f
r
am
e
or
f
r
a
m
e
bl
oc
k
i
ng
w
i
t
h
t
he
d
urati
o
n
of
10
-
30
m
s
on
pu
r
po
s
e
f
or
A
DC.
Ho
wev
er,
i
n
i
ts
proc
es
s
,
al
i
as
i
n
g
or
s
pe
c
t
r
al
l
ea
k
ag
e
or
di
s
c
on
t
i
nu
e
f
r
eq
ue
n
tl
y
oc
c
urs
.
F
or
thi
s
,
t
o
c
op
e
wi
t
h
th
i
s
prob
l
em
,
i
t
m
us
t
f
i
r
s
tl
y
s
tart
wi
t
h
wi
nd
o
w
i
ng
pr
oc
es
s
be
f
ore
the
s
i
g
na
l
c
o
nti
nu
es
t
o
F
F
T
ph
as
e.
If
de
f
i
ne
d,
th
e
w
i
nd
o
w
as
f
un
c
ti
on
of
(
)
,
0
≤
≤
(
−
1
)
i
s
de
pe
nd
e
nt
up
o
n
the
N
v
al
u
e
i
n
whi
c
h
N
r
ef
ers
to
the
nu
m
be
r
of
s
a
m
pl
es
i
n
i
ts
f
r
a
m
e.
T
he
i
np
ut
s
i
gn
a
l
en
ters
to
the
w
i
nd
o
w
i
ng
proc
es
s
w
i
t
h
the
c
on
v
o
l
ut
i
o
na
l
c
on
c
ep
t.
T
he
f
un
c
ti
on
of
wi
nd
o
w
us
ed
i
n
th
i
s
s
tud
y
i
s
Ham
m
i
ng
w
i
nd
o
wi
ng
as
s
ho
w
n
i
n
(
2
)
[3
7,
3
8
]:
(
)
=
0
.
54
−
0
.
46
(
2
−
1
)
(
2)
where
0
≤
≤
(
−
1
)
.
T
he
de
term
i
na
ti
o
n o
f
th
e
nu
m
be
r
of
f
r
a
m
e l
en
gth
s
ho
ul
d b
e
i
n
the
f
ol
d o
f
2
N
t
o f
ac
i
l
i
tat
e t
h
e FF
T
proc
es
s
ex
i
s
ti
n
g i
n t
h
e n
ex
t
bl
oc
k
.
S
tep
3:
DF
T
(
Di
s
c
r
ete
Fo
uri
er T
r
an
s
f
or
m
)
T
he
ou
tpu
t
s
i
g
na
l
f
r
o
m
Ham
m
i
ng
w
i
n
do
w
i
s
i
n
the
ti
m
e
do
m
ai
n.
T
o
f
ac
i
l
i
t
ate
the
m
ea
s
ure
m
en
t
i
n
th
e
f
urther
proc
es
s
Me
l
-
f
i
l
ter
ba
n
k
,
the
n
th
e
s
i
gn
al
i
s
tr
an
s
f
or
m
e
d
m
ath
e
m
ati
c
al
l
y
f
r
om
di
s
c
r
et
e
ti
m
e
do
m
ai
n
to
the
f
r
eq
u
en
c
y
us
i
ng
DF
T
m
eth
od
[
3
9].
M
ea
n
whi
l
e,
the
a
l
go
r
i
thm
us
ed
to
do
tr
an
s
f
or
m
ati
on
i
s
c
al
l
e
d
as
F
F
T
.
Ma
the
m
ati
c
al
l
y
,
DF
T
c
an
be
f
or
m
ul
ate
d a
s
f
ol
l
o
w
s
(
3) [4
0,
41
]:
[
]
=
∑
[
]
.
−
2
−
1
=
0
(
2
)
=
0
,
1
,
2
,
…
.
,
(
−
1
)
B
y
do
i
n
g
F
F
T
proc
es
s
,
the
n
i
t
c
an
ob
t
ai
n
th
e
v
a
l
ue
of
X
[k
]
as
the
r
e
s
ul
t
of
the
tr
an
s
f
orm
ati
on
f
r
o
m
F
F
T
r
ep
r
es
en
ti
ng
e
ac
h
v
al
ue
of
x
(
n)
fr
o
m
the
i
np
u
t
s
i
gn
a
l
.
F
r
o
m
the
i
n
pu
t
s
i
gn
a
l
i
n
whi
c
h
ea
c
h
of
the
v
a
l
ue
s
i
s
th
e
r
ep
r
es
en
tat
i
o
n
of
ba
s
i
c
f
r
eq
ue
nc
y
f
r
o
m
the
i
np
ut
s
i
g
na
l
. X
[k
] i
s
c
o
m
m
on
l
y
c
a
l
l
ed
as
s
pe
c
tr
um
or peri
o
do
gram
.
S
tep
4:
Me
l
Fr
eq
u
en
c
y
F
i
l
te
r
B
an
k
T
hi
s
ph
as
e
i
s
th
e
f
i
l
ter
i
ng
p
r
oc
es
s
fr
o
m
f
r
eq
ue
nc
y
s
p
e
c
tr
u
m
of
X
[
k
]
i
n
ea
c
h
f
r
am
e
us
i
ng
a
nu
m
be
r
of
M
f
i
l
ter
ba
nk
s
. T
hi
s
f
i
l
ter
i
s
m
ad
e
b
y
f
ol
l
o
wi
ng
t
he
pe
r
c
ep
t
i
on
of
m
el
fr
eq
ue
nc
y
s
c
al
e
r
ep
r
es
en
te
d
to
b
e
the
f
un
c
t
i
on
of
tr
i
an
g
l
e
f
i
l
ter
f
un
c
ti
on
an
d
m
el
s
c
al
e
f
r
eq
ue
nc
y
i
s
ob
ta
i
ne
d
f
r
o
m
the
r
es
ul
t
of
the
c
on
v
ers
i
on
of
l
i
ne
ar
f
r
eq
ue
nc
y
.
F
or
the
l
i
ne
ar
f
r
eq
ue
nc
y
(
f
Hz
)<
1
k
Hz
,
i
t
i
s
c
on
v
ert
ed
to
be
f
Hz
w
h
i
l
e
i
f
f
Hz
>
1
k
H
z
,
t
he
n
i
t
i
s
c
on
v
erted
i
nto
th
e
(
4
)
pres
en
ted
as
f
ol
l
o
w
s
[2
9],
[
35
],
[4
2
] :
(
)
=
2595
10
(
1
+
700
)
(
3
)
w
arpi
ng
proc
es
s
to
th
e
s
i
g
na
l
i
n
th
e
f
r
eq
ue
nc
y
do
m
ai
n
wi
l
l
r
es
u
l
t
i
n
the
v
a
l
u
e
of
Me
l
S
pe
c
tr
um
c
oe
f
f
i
c
i
en
ts
th
r
ou
g
h t
h
e p
r
o
c
es
s
as
s
ho
wn i
n (
5
)
as
f
ol
l
o
w
s
:
=
10
(
∑
|
(
)
|
(
)
−
1
=
0
)
(
4
)
X
i
i
s
t
he
v
a
l
u
e
of
f
r
eq
ue
nc
y
s
pe
c
tr
um
to
i
,
N
i
s
the
n
um
be
r
of
c
oe
ff
i
c
i
en
ts
of
F
F
T
,
an
d
H
i
(
f)
i
s
the
f
i
l
t
er v
al
u
e t
o
i
on
t
he
f
r
eq
ue
nc
y
s
p
ot
f
.
S
tep
5:
C
ep
s
tr
um
A
t
th
i
s
ph
as
e,
Me
l
C
ep
s
tr
u
m
w
ou
l
d
be
c
on
v
ert
ed
i
nto
the
t
i
m
e
do
m
ai
n
us
i
ng
D
i
s
c
r
ete
Cos
i
ne
T
r
an
s
f
or
m
(
DC
T
)
.
T
he
r
es
ul
t
i
s
c
al
l
ed
as
M
el
F
r
eq
ue
nc
y
Ce
ps
tr
um
Coe
ff
i
c
i
en
ts
(
MFCC
)
as
s
ho
wn
i
n
(
6
)
[41
,
4
2]
i
n
w
h
i
c
h
n
i
s
the
n
um
be
r
of
c
oe
f
f
i
c
i
en
t
an
d
M
i
s
the
n
u
m
be
r
o
f
f
i
l
ter
ba
nk
s
.
T
he
w
ord
c
ep
s
tr
u
m
i
s
orig
i
na
t
ed
f
r
o
m
the
wor
d
s
pe
c
tr
um
tha
t
i
s
r
ev
ers
ed
i
n
i
ts
f
i
r
s
t
s
y
l
l
ab
l
e
tha
t
i
s
s
pe
c
to
c
e
ps
[4
3].
Ceps
tr
um
i
s
t
he
po
w
er
s
pe
c
tr
um
ob
ta
i
ne
d
m
ath
e
m
ati
c
al
l
y
throug
h t
h
e l
og
ar
i
thm
i
c
c
om
pu
tat
i
on
[
44
,
45
].
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
F
P
G
A
-
ba
s
e
d i
mp
l
em
en
t
ati
on
of
s
pe
ec
h
r
ec
og
n
i
ti
on
f
o
r
r
ob
oc
ar c
on
tr
o
l
...
(
B
ay
ua
j
i
K
urn
i
ad
ha
n
i
)
1917
=
∑
(
)
[
(
−
1
2
)
]
=
1
(
5
)
3
. Digit
al
Des
ign
and
S
p
e
ec
h
Re
cog
n
it
ion
S
imu
l
ati
o
n
T
hi
s
s
ec
ti
on
d
i
s
c
us
s
es
i
n
d
eta
i
l
e
d
ab
ou
t
the
de
s
i
gn
of
the
bl
oc
k
of
s
pe
ec
h
r
ec
og
ni
t
i
on
(
S
R)
s
y
s
tem
.
T
he
m
ai
n
pa
r
t
c
on
s
i
s
ts
of
au
d
i
o
c
o
de
c
i
nte
r
f
ac
e,
f
ea
ture
ex
tr
ac
ti
o
n
us
i
n
g
MF
CC
an
d
c
l
as
s
i
f
i
c
ati
o
n
us
i
ng
E
u
c
l
i
de
an
di
s
ta
nc
e.
I
t
a
l
s
o
d
i
s
c
us
s
es
ab
ou
t
the
s
i
m
ul
ati
o
n
of
the
di
gi
t
a
l
l
og
i
c
d
es
i
g
n
t
o
F
P
G
A
us
i
n
g
th
e
X
i
l
i
nx
a
pp
l
i
c
at
i
on
.
I
n
t
he
be
g
i
nn
i
n
g
of
c
on
di
ti
o
n,
t
he
s
y
s
t
em
w
i
l
l
wai
t
f
or
the
i
np
ut
f
r
om
the
s
w
i
tc
h
to
do
tr
a
i
n
i
n
g
or
r
ec
og
n
i
z
i
ng
.
In
th
e
tr
ai
ni
n
g
m
o
de
,
th
e
c
om
i
ng
s
ou
nd
w
i
l
l
f
ac
e
the
proc
e
s
s
of
f
ea
ture
ex
tr
ac
ti
on
u
s
i
ng
MFCC
.
T
he
r
es
ul
t
o
f
the
f
ea
ture
ex
tr
ac
ti
on
w
i
l
l
b
e
i
n
the
f
orm
o
f
c
ep
s
tr
al
c
o
ef
f
i
c
i
en
t
s
t
ored
i
n
the
d
ata
b
as
e.
In
th
e
r
ec
og
ni
z
i
ng
m
od
e,
the
s
o
un
d
c
om
i
ng
t
o
th
e
s
y
s
t
em
w
i
l
l
f
ac
e
the
f
ea
t
ure
ex
tr
ac
ti
on
proc
es
s
.
T
he
n
,
the
c
oe
f
f
i
c
i
en
t
c
ep
s
tr
al
of
i
np
u
t
s
ou
n
d
w
ou
l
d
be
c
om
pa
r
ed
the
c
ep
s
tr
al
c
oe
ff
i
c
i
en
t
i
n
the
da
t
ab
as
e
us
i
ng
E
uc
l
i
de
an
di
s
ta
nc
e
m
eth
od
.
T
he
l
o
w
es
t
v
a
l
u
e
of
E
uc
l
i
d
ea
n
d
i
s
tan
c
e
w
o
ul
d
be
c
l
as
s
i
f
i
e
d
w
i
th
t
he
c
or
r
es
po
nd
i
ng
w
ords
.
F
urt
he
r
,
the
l
og
i
c
on
the
ou
t
pu
t
pi
n
wou
l
d
be
c
on
di
t
i
o
ne
d
i
n
ac
c
ordanc
e
wi
th
t
he
wor
ds
r
ec
og
n
i
z
ed
.
T
he
v
al
u
e
of
ou
t
pu
t
p
i
n
of
thi
s
F
P
G
A
i
s
us
ed
as
the
l
og
i
c
i
n
pu
t
o
n
the
dri
v
er
m
oto
r
to
c
on
tr
ol
the
m
oti
on
of
the
r
ob
oti
c
c
ar
as
s
ho
wn
i
n
T
ab
l
e
1 a
s
f
ol
l
o
w
s
.
T
ab
l
e 1
.
Lo
g
i
c
O
utp
ut
F
P
G
A
I
n
s
t
r
u
c
t
ion
Firs
t
M
o
t
o
r
S
e
c
o
n
d
M
o
t
o
r
Ou
t
p
u
t
L
o
g
ic
o
f
F
P
GA
En
D
ir 1
D
ir 2
En
D
ir 1
D
ir 2
“
R
igh
t
”
1
1
1
1
1
1
111111
“
L
e
f
t
”
1
0
1
1
1
0
101110
“
For
w
a
r
d
”
0
0
0
1
1
0
000110
“
B
a
c
k
w
a
r
d
”
0
0
0
1
0
1
000101
“
S
t
o
p
”
0
0
0
0
0
0
000000
T
he
r
es
ul
t
of
the
s
i
m
ul
at
i
on
ou
tp
ut
the
n
i
s
c
om
pa
r
ed
t
o
t
he
r
es
ul
t
of
t
he
M
A
T
LA
B
.
T
he
s
y
s
t
em
arc
hi
tec
ture
o
f
the
s
pe
ec
h
r
ec
o
gn
i
ti
on
bl
oc
k
i
s
s
ho
wn
i
n
F
i
g
ure
2.
T
he
s
y
s
t
em
arc
hi
tec
ture
c
on
s
i
s
ts
of
f
ou
r
m
ai
n
b
l
oc
k
s
:
m
i
c
r
op
ho
n
e,
a
ud
i
o
c
o
de
c
,
F
P
G
A
,
dri
v
er
m
ot
or
an
d
m
oto
r
DC. In
s
i
de
th
e FP
G
A
, i
t
i
s
a
dd
e
d
wi
th
ei
gh
t
di
gi
t
al
s
eri
es
ac
ti
ng
as
th
e c
ore
s
of
th
e s
y
s
tem
tho
s
e
are
c
od
ec
i
nte
r
f
ac
e,
pre
proc
es
s
i
ng
,
c
on
tr
o
l
un
i
t,
c
l
oc
k
di
v
i
de
r
,
MF
CC,
d
ata
b
as
e,
E
uc
l
i
de
an
di
s
ta
nc
e a
n
d o
ut
pu
t
l
og
i
c
.
F
i
gu
r
e
2.
A
r
c
hi
tec
t
ure of
pr
op
os
ed
s
pe
ec
h r
ec
og
n
i
ti
on
4
. Re
sult
s
a
n
d
A
n
al
y
s
is
S
i
m
ul
ati
on
of
the
l
og
i
c
s
erie
s
d
es
i
g
n
was
c
on
d
uc
ted
i
n
X
i
l
i
nx
IS
E
P
r
oj
ec
t
N
av
i
ga
tor.
A
l
s
o,
c
om
pu
tat
i
o
n
s
i
m
ul
ati
on
w
as
do
n
e
i
n
MA
T
LA
B
i
n
order
to
o
bta
i
n
the
c
om
p
arin
g
da
t
a
f
r
om
the
d
es
i
g
n
ed
s
y
s
tem
.
T
he
tes
t
w
as
c
on
du
c
te
d
b
y
o
bs
erv
i
ng
an
d
c
om
pa
r
i
n
g
th
e
da
t
a
i
n
e
ac
h
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
, N
o.
4
,
A
u
gu
s
t
2
01
9
:
1
9
14
-
1
922
1918
s
ub
-
s
y
s
t
em
of
s
pe
ec
h
r
ec
o
gn
i
t
i
o
n.
T
he
s
y
nt
he
s
i
s
of
th
e
l
og
i
c
c
i
r
c
ui
t
i
n
th
i
s
pr
op
os
ed
s
y
s
t
em
c
an
be
s
ee
n i
n
F
i
gu
r
e
3
.
F
i
gu
r
e
3.
RT
L De
s
i
gn
s
c
he
m
ati
c
4
.1.
P
r
e
-
emp
h
as
i
s Filt
e
r
T
hi
s
s
erie
s
f
un
c
ti
on
s
to
r
e
du
c
e
th
e
n
oi
s
e
r
a
ti
o
i
n
s
i
gn
a
l
(
S
NR)
.
T
hi
s
f
i
l
t
er
m
ai
nta
i
ns
the
h
i
g
h
f
r
eq
ue
nc
i
es
on
t
he
s
pe
c
tr
um
el
i
m
i
na
te
d
i
n
the
pr
oc
es
s
of
s
ou
nd
pr
od
uc
ti
on
.
F
r
om
the
r
es
u
l
t
of
s
i
m
ul
ati
o
n
t
e
s
t
c
om
pa
r
ed
b
y
m
ea
ns
of
m
an
ua
l
c
a
l
c
ul
ati
on
,
i
t
h
as
be
en
f
ou
nd
a
s
i
m
i
l
ar v
al
ue
i
n
ea
c
h s
am
pl
e o
f
s
i
gn
a
l
. T
he
r
es
u
l
ts
are
s
ho
w
n
i
n T
ab
l
e
2
.
T
ab
l
e 2
.
C
om
pa
r
i
s
on
of
th
e
Res
ul
t
of
Ma
n
ua
l
Ca
l
c
ul
ati
on
o
n t
h
e P
r
e
-
em
ph
as
i
s
Fil
t
er and
S
i
m
ul
ati
on
R
es
ul
t
No
M
a
n
u
a
l
C
a
lc
u
lat
ion
X
il
in
x
S
im
u
la
t
ion
1
380
380
2
-
1
6
6
.
2
5
-
167
3
2
8
2
.
8
7
5
282
4
-
2
5
.
1
8
7
5
-
26
5
-
4
3
6
.
5
6
3
-
437
6
4
3
1
.
5
6
2
5
431
7
2
3
.
7
5
23
8
-
5
1
9
.
2
5
-
520
9
-
9
2
.
1
8
7
5
-
93
10
-
4
2
.
3
1
2
5
-
43
4
.2.
FFT
T
he
c
om
pu
tat
i
on
F
F
T
de
s
i
gn
on
F
P
G
A
i
s
do
ne
s
e
pa
r
ate
l
y
to
r
es
ul
t
i
n
2
p
arts
of
ou
tp
ut.
F
r
o
m
F
i
gu
r
e
4
b
el
o
w
c
a
n
be
s
e
en
th
at
t
he
r
es
ul
t
was
no
t
m
uc
h
di
f
f
erent
as
t
he
c
a
l
c
ul
ati
on
op
erat
i
o
n
i
n
t
hi
s
d
es
i
g
n
d
i
d
no
t
us
e
t
he
f
l
o
ati
ng
po
i
nt
s
y
s
t
em
.
T
he
c
om
pa
r
i
s
on
of
the
r
es
ul
ts
of
F
F
T
i
n t
he
gra
ph
i
c
f
or
m
c
a
n b
e s
e
en
i
n
F
i
gu
r
e
4
.
4.3
. M
el
F
r
equ
enc
y
W
ar
p
i
n
g
Me
l
F
r
eq
u
en
c
y
W
a
r
pi
ng
f
un
c
ti
on
s
as
a
f
i
l
t
er
f
r
o
m
the
s
p
ec
tr
um
of
f
r
eq
ue
nc
y
of
the
ou
t
pu
t
r
es
ul
t
of
F
F
T
.
T
he
m
ul
ti
pl
i
c
ati
on
proc
es
s
was
do
ne
i
n
pa
r
al
l
e
l
to
20
f
i
l
ter
ba
nk
s
to
m
a
k
e
thi
s
proc
es
s
f
as
ter.
T
he
r
es
u
l
ts
f
r
o
m
20
f
i
l
ter
ba
nk
are
i
n
t
he
f
orm
of
m
ag
ni
tud
e
v
al
u
es
as
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
F
P
G
A
-
ba
s
e
d i
mp
l
em
en
t
ati
on
of
s
pe
ec
h
r
ec
og
n
i
ti
on
f
o
r
r
ob
oc
ar c
on
tr
o
l
...
(
B
ay
ua
j
i
K
urn
i
ad
ha
n
i
)
1919
s
ho
w
n
i
n
F
i
gu
r
e
5
.
A
s
s
ee
n
i
n
F
i
g
ure
5
ab
o
v
e,
i
t
c
an
be
s
ee
n
th
at
t
he
f
ea
tur
e
i
n
5
-
1
0
f
r
o
m
the
r
es
u
l
t
of
X
i
l
i
nk
s
i
m
ul
ati
on
a
pp
r
o
ac
he
d
th
e
M
A
T
LA
B
c
om
pu
tat
i
on
.
T
hi
s
s
ho
w
s
tha
t
th
e
l
og
i
c
de
s
i
g
n m
ad
e i
s
prec
i
s
e
.
F
i
g
ure
4.
T
he
G
r
ap
h
of
th
e
c
al
c
ul
a
ti
o
n
r
es
u
l
t o
n 2
5
6
-
po
i
nt
F
F
T
on
th
e M
A
T
LA
B
an
d X
i
l
i
nx
S
i
m
ul
ati
o
n
F
i
gu
r
e
5.
G
r
ap
h o
f
Me
l
Fr
e
qu
en
c
y
W
arpi
ng
r
es
ul
t
on
MA
T
LA
B
an
d S
i
m
ul
ati
on
o
n X
i
l
i
nx
4.4
. C
epst
r
u
m
T
hi
s
bl
oc
k
f
un
c
ti
on
s
to
c
o
n
v
ert
t
he
Me
l
c
ep
s
tr
um
f
r
o
m
the
r
es
ul
t
of
pre
v
i
o
us
b
l
oc
k
i
nto
the
ti
m
e
do
m
ai
n
us
i
ng
Di
s
c
r
ete
Cos
i
ne
T
r
an
s
f
or
m
.
T
he
c
oe
f
f
i
c
i
en
t
v
a
l
u
e
w
as
the
n
c
ha
ng
ed
i
nto
the
r
ep
r
es
en
ta
ti
o
n
of
f
i
x
ed
po
i
nt
16
bi
ts
an
d
s
t
ored
i
n
RO
M.
T
hi
s
bl
oc
k
al
s
o
ha
d
RA
M
t
o
s
tore
the
ou
t
pu
t
f
r
o
m
the
pre
v
i
ou
s
proc
es
s
to
f
ac
i
l
i
t
ate
t
he
proc
es
s
.
T
he
r
es
u
l
ts
o
f
the
bl
oc
k
,
i
f
c
o
m
pa
r
ed
to
the
r
es
u
l
t
of
the
c
al
c
ul
a
ti
on
on
M
A
T
LA
B
a
s
s
ho
w
n
i
n
F
i
g
ure
6,
w
ere
s
ee
n
di
f
f
erent.
T
hi
s
w
as
b
ec
au
s
e
the
c
a
l
c
ul
at
i
o
n
op
erat
i
on
i
n
th
i
s
bl
oc
k
i
nv
ol
v
e
d
th
e
n
um
be
r
s
tha
t
ha
d
s
om
e
di
g
i
ts
af
ter
th
e
c
om
m
a.
Me
an
whi
l
e,
the
r
ep
r
es
en
tat
i
o
n
of
the
nu
m
be
r
s
us
ed
di
d
no
t
ha
v
e
an
y
ac
c
urac
y
i
n n
um
be
r
; th
us
, r
ou
nd
i
n
g o
c
c
urr
ed
t
o t
h
e rep
r
es
en
tat
i
on
of
nu
m
be
r
s
c
l
os
er.
4.5
. D
ec
i
sion
12
c
oe
f
f
i
c
i
en
ts
f
r
o
m
the
r
es
ul
t
of
f
ea
ture
ex
tr
ac
ti
on
s
to
r
ed
i
n
the
da
ta
ba
s
e
at
th
e
t
es
ti
ng
ph
as
e
wi
l
l
be
c
om
pa
r
ed
to
the
c
o
ef
f
i
c
i
en
t
of
the
s
o
un
d
i
n
pu
t
f
ea
t
ure
a
nd
t
he
n
w
i
l
l
be
c
ut
to
d
o
1
2
3
4
5
6
7
8
9
10
N
umb
er
of
F
ea
t
ur
es
2
2.5
3
3.5
4
M
a
g
n
i
t
u
d
e
Mat
l
ab
vs
X
i
l
i
n
k
S
i
mu
l
at
i
o
n
M
a
t
l
a
b
v
s
Xi
l
i
n
k
Si
m
u
l
a
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
, N
o.
4
,
A
u
gu
s
t
2
01
9
:
1
9
14
-
1
922
1920
the
i
ns
tr
uc
ti
on
as
de
term
i
ne
d.
T
he
de
c
i
s
i
on
w
as
t
ak
en
b
y
c
a
l
c
ul
ati
ng
the
c
l
os
es
t
f
ea
ture
c
oe
f
f
i
c
i
en
t
v
al
ue
us
i
n
g
eu
c
l
i
de
an
di
s
t
an
c
e.
T
he
b
l
oc
k
of
da
ta
ba
s
e
m
ad
e
w
as
us
e
d
to
s
tore
o
ne
s
a
m
pl
e
of
e
ac
h
i
ns
tr
uc
ti
o
n
w
ord
.
F
i
v
e
d
ata
s
et
of
f
ea
ture
wer
e
s
a
v
e
d
i
n
the
da
ta
ba
s
e.
O
nc
e
ob
t
ai
n
i
n
g
th
e
c
l
os
es
t
nu
m
be
r
,
the
n
de
c
i
s
i
on
was
tak
en
to
r
eg
u
l
at
e
th
e
c
on
tr
o
l
i
n
the
m
oto
r
i
n t
he
f
orm
at
of
4
bi
t d
a
ta
a
s
s
ho
wn i
n t
h
e f
ol
l
o
w
i
ng
F
i
g
ure 7.
F
i
gu
r
e
6.
G
r
ap
h o
f
on
t
he
C
al
c
ul
ati
on
R
es
ul
t
on
M
A
T
LA
B
an
d
X
i
l
i
nx
S
i
m
ul
ati
on
F
i
gu
r
e
7.
R
es
ul
t
of
th
e s
i
m
ul
at
i
on
on
t
he
ou
tp
ut
l
og
i
c
b
l
oc
k
on
X
i
l
i
nx
5
. Con
clus
ion
T
hi
s
r
es
ea
r
c
h
ha
s
s
uc
c
es
s
f
ul
l
y
m
ad
e
a
d
es
i
g
n
an
d
s
i
m
ul
ati
o
n
of
l
og
i
c
s
eri
es
i
n
a
s
pe
ec
h
r
ec
og
n
i
ti
on
s
y
s
tem
us
i
ng
th
e
MFCC
m
eth
od
an
d
eu
c
l
i
de
a
n
di
s
ta
nc
e
to
c
on
tr
ol
th
e
r
ate
of
r
ob
oti
c
c
ar.
T
he
MFCC
m
eth
od
was
us
ed
t
o
o
bta
i
n
the
f
ea
t
ur
e
f
r
om
the
c
om
m
an
d
i
np
ut
i
n
the
f
or
m
of
s
ou
nd
c
on
s
i
s
ti
n
g
of
r
i
gh
t,
l
ef
t,
f
orw
ard,
ba
c
k
w
ard
an
d
s
to
p.
T
he
r
es
ul
t
of
the
s
i
gn
a
l
f
ea
ture
f
r
o
m
MFCC
f
urtherm
ore
w
as
c
al
c
ul
a
ted
f
or
i
t
s
s
i
m
i
l
arit
y
an
d
c
om
pa
r
ed
wi
th
t
he
s
i
gn
a
l
f
ea
ture i
n
th
e
da
t
ab
as
e u
s
i
ng
e
uc
l
i
d
ea
n d
i
s
tan
c
e
to
g
i
v
e t
he
c
o
ntrol
l
o
gi
c
i
n m
oto
r
.
P
r
oc
es
s
s
i
m
pl
i
f
i
c
ati
on
was
al
s
o
d
on
e
to
o
bta
i
n
t
he
r
es
ou
r
c
e
of
the
F
P
G
A
m
em
ory
as
m
i
ni
m
u
m
as
po
s
s
i
bl
e
b
ut
s
ti
l
l
h
ad
a
go
o
d
pe
r
f
or
m
an
c
e.
V
a
l
i
da
t
i
on
h
as
be
en
c
o
nd
uc
te
d
t
o
tes
t
the
ex
c
el
l
e
nc
e
of
the
s
y
s
t
e
m
m
ad
e.
T
hi
s
tes
t
was
do
ne
b
y
c
om
pa
r
i
ng
t
he
o
utp
u
t
v
a
l
ue
of
l
og
i
c
de
s
i
g
n
tha
t
ha
s
be
en
m
ad
e
i
n
e
ac
h
pa
r
t
of
s
pe
ec
h
r
e
c
og
ni
t
i
o
n
c
om
po
ne
nts
\
w
i
th
the
c
a
l
c
u
l
at
i
o
n
s
i
m
ul
ati
on
on
M
A
T
LA
B
.
T
he
r
e
w
as
a
di
f
f
erenc
e
i
n
th
e
v
al
ue
be
t
ween
the
r
es
u
l
t
of
the
c
om
pu
tat
i
o
n
of
th
e
l
o
gi
c
s
erie
s
an
d
the
MA
T
L
A
B
as
th
e
c
al
c
u
l
at
i
ng
op
e
r
ati
on
i
n
v
o
l
v
ed
the
nu
m
be
r
s
tha
t
ha
d
s
om
e
di
gi
ts
af
ter
the
c
o
m
m
a.
Me
an
w
h
i
l
e,
the
r
ep
r
es
en
tat
i
on
of
the
n
um
be
r
s
us
ed
di
d
no
t
ha
v
e
a
n
y
s
uf
f
i
c
i
en
t
nu
m
be
r
ac
c
urac
y
;
t
hu
s
m
a
k
i
ng
the
r
o
un
d
i
n
g
oc
c
urr
ed
i
n
the
c
l
os
es
t
r
e
pres
en
tat
i
v
e
nu
m
be
r
s
.
Howev
er,
th
i
s
di
f
f
erenc
e
v
al
u
e
w
as
r
e
l
at
i
v
e
i
ns
i
g
ni
f
i
c
an
t
a
nd
the
s
y
s
t
e
m
the
n
s
ti
l
l
ha
d
a
g
oo
d
p
e
r
f
or
m
an
c
e
to
c
om
pa
r
e
the
s
i
gn
a
l
f
ea
t
ures
f
r
o
m
on
e
c
l
as
s
t
o
oth
er
a
s
prov
en
i
n
th
e
r
es
u
l
ts
.
T
he
ne
x
t
r
es
e
arc
h
wi
l
l
be
do
ne
thr
ou
g
h
the
i
m
pl
em
en
tat
i
o
n
to
th
e
F
P
G
A
bo
ar
d
an
d
a
na
l
y
s
i
s
on
the
s
y
n
th
es
i
s
of
l
og
i
c
s
erie
s
to
de
term
i
ne
th
e p
ar
am
ete
r
of
th
e
pe
r
f
or
m
an
c
e i
n I
C de
s
i
gn
.
1
2
3
4
5
6
7
8
9
10
11
12
Number of Features
-10
-6
-2
2
6
10
M
a
g
n
i
t
u
d
e
Matlab Computation vs Xilink Simulation
Matlab
Xilink
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
F
P
G
A
-
ba
s
e
d i
mp
l
em
en
t
ati
on
of
s
pe
ec
h
r
ec
og
n
i
ti
on
f
o
r
r
ob
oc
ar c
on
tr
o
l
...
(
B
ay
ua
j
i
K
urn
i
ad
ha
n
i
)
1921
Ref
er
en
ce
s
[1
]
M
Va
c
h
e
r,
AFl
e
u
ry
,
F
Po
rt
e
t,
J
-
F
Se
r
i
g
n
a
t,
N
No
u
ry
.
Co
m
p
l
e
te
So
u
n
d
a
n
d
Sp
e
e
c
h
Rec
o
g
n
i
ti
o
n
Sy
s
te
m
f
o
r
He
a
l
th
Sm
a
r
t
Ho
m
e
s
:
Ap
p
l
i
c
a
ti
o
n
t
o
t
h
e
Re
c
o
g
n
i
ti
o
n
o
f
A
c
ti
v
i
ti
e
s
o
f
D
a
i
l
y
L
i
v
i
n
g
.
N
e
w
Dev
e
l
o
p
m
e
n
t
s
i
n
B
i
o
m
e
d
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
.
In
T
e
c
h
.
2
0
1
0
.
[2
]
P
Pu
tt
h
a
p
i
p
a
t,
C
W
o
r
a
l
e
r
t,
P
Si
ri
n
i
m
n
u
a
n
k
u
l
.
Sp
e
e
c
h
re
c
o
g
n
i
ti
o
n
g
a
te
wa
y
f
o
r
h
o
m
e
a
u
to
m
a
ti
o
n
o
n
o
p
e
n
p
l
a
tf
o
rm
.
In
t
e
rn
a
t
i
o
n
a
l
Con
fe
re
n
c
e
o
n
El
e
c
tro
n
i
c
s
,
I
n
fo
rm
a
ti
o
n
,
a
n
d
Com
m
u
n
i
c
a
ti
o
n
(ICEIC
)
2
0
1
8
.
H
a
w
a
i
,
USA. 2
0
1
8
:
1
–
4.
[3
]
T
Ay
re
s
,
B
Nol
a
n
.
Vo
i
c
e
a
c
ti
v
a
te
d
c
o
m
m
a
n
d
a
n
d
c
o
n
tro
l
w
i
th
s
p
e
e
c
h
re
c
o
g
n
i
ti
o
n
o
v
e
r
W
i
Fi
.
Sc
i
.
Com
p
u
t.
Pro
g
r
a
m
.
2
0
0
6
;
5
9
(
1
–
2
):
1
0
9
–
1
2
6
.
[4
]
A
M
o
h
a
n
ta
,
VK
M
i
tt
a
l
.
Hum
a
n
Em
o
ti
o
n
a
l
Sta
te
s
Cla
s
s
i
fi
c
a
ti
o
n
Ba
s
e
d
u
p
o
n
Cha
n
g
e
s
i
n
Sp
e
e
c
h
Pro
d
u
c
ti
o
n
F
e
a
t
u
re
s
i
n
Vo
we
l
Re
g
i
o
n
.
In
te
r
n
a
ti
o
n
a
l
Con
fe
r
e
n
c
e
o
n
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
a
n
d
Net
w
o
rk
s
(T
EL
-
NE
T
) 2
0
1
7
.
No
i
d
a
,
In
d
i
a
.
2
0
1
7
.
[5
]
KF
Ak
i
n
g
b
a
d
e
,
O
M
Um
a
n
n
a
,
IA
Al
i
m
i
.
Vo
i
c
e
-
Ba
s
e
d
Doo
r
Ac
c
e
s
s
Con
tro
l
Sy
s
te
m
Us
i
n
g
th
e
M
e
l
Fre
q
u
e
n
c
y
Cep
s
tru
m
Coe
ff
i
c
i
e
n
ts
a
n
d
G
a
u
s
s
i
a
n
M
i
x
tu
re
M
o
d
e
l
.
In
t.
J
.
E
l
e
c
tr.
C
o
m
p
u
t.
En
g
.
2
0
1
4
;
4
(5
):
6
4
3
–
6
4
7
.
[6
]
W
Rao
e
t
a
l
.
In
v
e
s
t
i
g
a
ti
o
n
o
f
fi
x
e
d
-
d
i
m
e
n
s
i
o
n
a
l
s
p
e
e
c
h
re
p
re
s
e
n
t
a
ti
o
n
s
fo
r
re
a
l
-
ti
m
e
s
p
e
e
c
h
e
m
o
ti
o
n
re
c
o
g
n
i
ti
o
n
s
y
s
te
m
.
In
te
rn
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
Oran
g
e
T
e
c
h
n
o
l
o
g
i
e
s
(I
CO
T
) 2
0
1
7
:
197
–
2
0
0
.
[7
]
J
Y
a
d
a
v
,
M
S
Fa
h
a
d
,
KS
Ra
o
.
Ep
o
c
h
d
e
t
e
c
t
i
o
n
fro
m
e
m
o
ti
o
n
a
l
s
p
e
e
a
c
h
s
i
g
n
a
l
u
s
i
n
g
z
e
ro
ti
m
e
w
i
n
d
o
w
i
n
g
.
Sp
e
e
c
h
Co
m
m
u
n
.
2
0
1
7
;
9
6
(Nov
e
m
b
e
r):
1
4
2
–
1
4
9
.
[8
]
HK
Pa
l
o
,
M
N
M
o
h
a
n
ty
.
Cla
s
s
i
fi
c
a
t
i
o
n
o
f
E
m
o
ti
o
n
a
l
Sp
e
e
c
h
o
f
Ch
i
l
d
r
e
n
U
s
i
n
g
Pro
b
a
b
i
l
i
s
ti
c
Neu
ra
l
Net
w
o
rk
.
In
t.
J
.
El
e
c
tr.
Co
m
p
u
t
.
En
g
.
2
0
1
5
;
5
(2
):
3
1
1
–
3
1
7
.
[9
]
FE
G
u
n
a
w
a
n
,
K
Id
a
n
a
n
t
a
.
Pr
e
d
i
c
ti
n
g
t
h
e
L
e
v
e
l
o
f
Em
o
t
i
o
n
b
y
M
e
a
n
s
o
f
In
d
o
n
e
s
i
a
n
Sp
e
e
c
h
Si
g
n
a
l
.
TEL
KO
M
NIKA
.
2
0
1
7
;
1
5
(2
):
6
6
5
–
6
7
0
.
[1
0
]
J
T
w
i
e
fe
l
,
X
Hin
a
u
t,
S
W
e
r
m
t
e
r,
J
T
w
i
e
fe
l
,
X
Hin
a
u
t,
S
.
W
e
rm
te
r
.
S
y
n
t
a
c
t
i
c
Rea
n
a
l
y
s
i
s
i
n
L
a
n
g
u
a
g
e
M
o
d
e
l
s
fo
r
S
p
e
e
c
h
Re
c
o
g
n
i
t
i
o
n
.
I
n
t.
Co
n
f.
Dev
.
L
e
a
rn
.
Ep
i
g
e
n
e
ti
c
Rob
o
t
.
L
i
s
b
o
n
,
Po
rtu
g
a
l
.
2
0
1
7
:
215
–
2
2
0
.
[1
1
]
D
Rec
a
s
e
n
s
,
C
Ro
d
ríg
u
e
z
.
Con
te
x
tu
a
l
a
n
d
s
y
l
l
a
b
i
c
e
ff
e
c
t
s
i
n
h
e
t
e
ro
s
y
l
l
a
b
i
c
c
o
n
s
o
n
a
n
t
s
e
q
u
e
n
c
e
s
.
An
u
l
tr
a
s
o
u
n
d
s
tu
d
y
.
Sp
e
e
c
h
Com
m
u
n
.
2017
;
9
6
(De
c
e
m
b
e
r
):
1
5
0
–
1
6
7
.
[1
2
]
M
J
i
a
,
J
Su
n
,
C
Ba
o
,
C
Rit
z
.
Se
p
a
ra
t
i
o
n
o
f
m
u
l
ti
p
l
e
s
p
e
e
c
h
s
o
u
rc
e
s
b
y
re
c
o
v
e
ri
n
g
s
p
a
rs
e
a
n
d
non
-
s
p
a
rs
e
c
o
m
p
o
n
e
n
t
s
fro
m
B
-
fo
rm
a
t
m
i
c
ro
p
h
o
n
e
re
c
o
rd
i
n
g
s
.
Sp
e
e
c
h
Com
m
u
n
.
2
0
1
7
;
9
6
(M
a
y
):
184
–
1
9
6
.
[1
3
]
M
T
a
h
o
n
,
G
L
e
c
o
rv
e
,
D
L
o
l
i
v
e
.
Can
w
e
G
e
n
e
ra
t
e
E
m
o
t
i
o
n
a
l
Pro
n
u
n
c
i
a
ti
o
n
s
fo
r
Ex
p
re
s
s
i
v
e
Sp
e
e
c
h
Sy
n
th
e
s
i
s
?
.
IEEE
Tra
n
s
.
Af
fe
c
t.
Co
m
p
u
t.
2
0
1
8
;
3
0
4
5
:
1
-
1.
[1
4
]
K
L
i
,
S
M
a
o
,
X
L
i
,
Z
W
u
,
H
M
e
n
g
.
Au
to
m
a
t
i
c
l
e
x
i
c
a
l
s
tr
e
s
s
a
n
d
p
i
t
c
h
a
c
c
e
n
t
d
e
te
c
ti
o
n
f
o
r
L
2
En
g
l
i
s
h
s
p
e
e
c
h
u
s
i
n
g
m
u
l
ti
-
d
i
s
tr
i
b
u
t
i
o
n
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
.
Sp
e
e
c
h
Com
m
u
n
.
2
0
1
7
;
9
6
(S
e
p
te
m
b
e
r)
:
28
–
3
6
.
[1
5
]
RK
Ka
n
d
a
g
a
tl
a
,
PV.
S
u
b
b
a
i
a
h
.
Sp
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
u
s
i
n
g
M
M
SE
e
s
ti
m
a
t
i
o
n
o
f
a
m
p
l
i
tu
d
e
a
n
d
c
o
m
p
l
e
x
s
p
e
e
c
h
s
p
e
c
tr
a
l
c
o
e
ff
i
c
i
e
n
t
s
u
n
d
e
r
p
h
a
s
e
-
u
n
c
e
rt
a
i
n
ty
.
Sp
e
e
c
h
Com
m
u
n
.
2
0
1
7
;
9
6
(Nov
e
m
b
e
r):
1
0
–
2
7
.
[1
6
]
A
M
i
s
ra
,
J
HL.
Ha
n
s
e
n
.
M
o
d
e
l
l
i
n
g
a
n
d
c
o
m
p
e
n
s
a
ti
o
n
fo
r
l
a
n
g
u
a
g
e
m
i
s
m
a
tc
h
i
n
s
p
e
a
k
e
r
v
e
ri
fi
c
a
ti
o
n
.
Sp
e
e
c
h
Co
m
m
u
n
.
2
0
1
7
;
9
6
(J
a
n
u
a
ry
):
5
8
–
66.
[1
7
]
B
W
i
e
m
,
B
M
M
o
h
a
m
e
d
a
n
o
u
a
r,
P
M
o
w
l
a
e
e
,
B
A
i
c
h
a
.
Uns
u
p
e
rv
i
s
e
d
s
i
n
g
l
e
c
h
a
n
n
e
l
s
p
e
e
c
h
s
e
p
a
r
a
ti
o
n
b
a
s
e
d
o
n
o
p
ti
m
i
z
e
d
s
u
b
s
p
a
c
e
s
e
p
a
ra
ti
o
n
.
S
p
e
e
c
h
Co
m
m
u
n
.
2017
;
9
6
(A
p
ri
l
)
:
9
3
–
101.
[1
8
]
R
So
l
e
y
m
a
n
i
,
I
W
.
S
e
l
e
s
n
i
c
k
,
D
M
.
L
a
n
d
s
b
e
rg
e
r.
SEDA
:
A
tu
n
a
b
l
e
Q
-
fa
c
to
r
w
a
v
e
l
e
t
-
b
a
s
e
d
n
o
i
s
e
re
d
u
c
ti
o
n
a
l
g
o
ri
t
h
m
f
o
r m
u
l
t
i
-
ta
l
k
e
r
b
a
b
b
l
e
.
S
p
e
e
c
h
Co
m
m
u
n
.
2017
;
9
6
(O
c
to
b
e
r)
:
1
0
2
–
1
1
5
.
[1
9
]
Y
T
a
n
g
,
Q
L
i
u
,
W
W
a
n
g
,
T
J
C
o
x
.
A
n
o
n
-
i
n
tru
s
i
v
e
m
e
t
h
o
d
fo
r
e
s
ti
m
a
t
i
n
g
b
i
n
a
u
ra
l
s
p
e
e
c
h
i
n
te
l
l
i
g
i
b
i
l
i
ty
fro
m
n
o
i
s
e
-
c
o
rru
p
te
d
s
i
g
n
a
l
s
c
a
p
t
u
re
d
b
y
a
p
a
i
r
o
f
m
i
c
ro
p
h
o
n
e
s
.
Sp
e
e
c
h
Com
m
u
n
.
2
0
1
7
;
9
6
(
J
u
n
e
)
:
116
–
1
2
8
.
[2
0
]
F
Sa
k
i
,
A
Bh
a
tt
a
c
h
a
ry
a
,
N
Ke
h
ta
rn
a
v
a
z
.
Rea
l
-
t
i
m
e
Si
m
u
l
i
n
k
i
m
p
l
e
m
e
n
ta
t
i
o
n
o
f
n
o
i
s
e
a
d
a
p
t
i
v
e
s
p
e
e
c
h
p
ro
c
e
s
s
i
n
g
p
i
p
e
l
i
n
e
o
f
c
o
c
h
l
e
a
r
i
m
p
l
a
n
t
s
.
Sp
e
e
c
h
Co
m
m
u
n
.
2
0
1
7
;
9
6
(Ap
r
i
l
):
1
9
7
–
2
0
6
.
[2
1
]
S
Nara
g
a
n
,
M
D
G
u
p
ta
.
Sp
e
e
c
h
Fe
a
t
u
re
Ex
tra
c
ti
o
n
T
e
c
h
n
i
q
u
e
:
A
Rev
i
e
w
.
In
t.
J
.
Com
p
u
t.
Sc
i
.
M
o
b
.
Com
p
u
t.
2
0
1
5
;
4
(
3
):
1
0
7
–
114.
[2
2
]
N
Dav
e
.
Fe
a
tu
r
e
Ex
tra
c
ti
o
n
M
e
th
o
d
s
L
PC,
P
L
P
a
n
d
M
FCC
In
Sp
e
e
c
h
R
e
c
o
g
n
i
t
i
o
n
.
In
t.
J
.
Ad
v
.
Re
s
.
En
g
.
Te
c
h
n
o
l
.
2013
;
1
(VI):
1
–
5.
[2
3
]
Chu
l
h
e
e
L
e
e
,
Do
n
g
h
o
o
n
Hy
u
n
,
Eu
i
s
u
n
Cho
i
,
J
i
n
w
o
o
k
G
o
,
a
n
d
Chu
n
g
y
o
n
g
L
e
e
.
O
p
ti
m
i
z
i
n
g
fe
a
t
u
r
e
e
x
tra
c
ti
o
n
f
o
r
s
p
e
e
c
h
re
c
o
g
n
i
ti
o
n
.
IEEE
Tr
a
n
s
.
Sp
e
e
c
h
A
u
d
i
o
Pro
c
e
s
s
.
2
0
0
3
;
1
1
(1
):
8
0
–
8
7
.
[2
4
]
C
Vi
m
a
l
a
a
n
d
V
.
Rad
h
a
.
S
u
i
ta
b
l
e
F
e
a
tu
r
e
Ex
tra
c
ti
o
n
a
n
d
Sp
e
e
c
h
Rec
o
g
n
i
ti
o
n
T
e
c
h
n
i
q
u
e
f
o
r
Is
o
l
a
t
e
d
T
a
m
i
l
Sp
o
k
e
n
W
o
rd
s
.
I
n
t.
J
.
C
o
m
p
u
t.
Sc
i
.
In
f.
T
e
c
h
n
o
l
.
2
0
1
4
;
5
(1
):
3
7
8
–
383.
[2
5
]
M
G
u
rb
a
n
,
J
P
T
h
i
ra
n
.
I
n
fo
r
m
a
t
i
o
n
T
h
e
o
re
t
i
c
F
e
a
tu
r
e
Ex
tra
c
ti
o
n
fo
r
Au
d
i
o
-
Vi
s
u
a
l
Sp
e
e
c
h
R
e
c
o
g
n
i
ti
o
n
.
IEEE
Tra
n
s
.
Si
g
n
a
l
Pr
o
c
e
s
s
.
2
0
0
9
;
5
7
(1
2
)
:
4
7
6
5
–
4
7
7
6
.
[2
6
]
H
G
u
p
ta
,
D
G
u
p
ta
.
L
PC
a
n
d
L
PCC
m
e
th
o
d
o
f
fe
a
tu
re
e
x
t
ra
c
ti
o
n
i
n
Sp
e
e
c
h
Rec
o
g
n
i
ti
o
n
Sy
s
te
m
.
Pro
c
.
6
th
I
n
t.
Co
n
f
.
Clo
u
d
Sy
s
t
.
Bi
g
Da
t
a
En
g
.
Con
f
l
u
.
(2
0
1
6
)
.
No
i
d
a
,
I
n
d
i
a
:
4
9
8
–
502.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 16
93
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
, N
o.
4
,
A
u
gu
s
t
2
01
9
:
1
9
14
-
1
922
1922
[2
7
]
HJ
M
Ste
e
n
e
k
e
n
,
J
HL
Ha
n
s
e
n
.
Sp
e
e
c
h
u
n
d
e
r
s
tre
s
s
c
o
n
d
i
ti
o
n
s
:
o
v
e
rv
i
e
w
o
f
th
e
e
ff
e
c
t
o
n
s
p
e
e
c
h
p
ro
d
u
c
ti
o
n
a
n
d
o
n
s
y
s
te
m
p
e
rfo
rm
a
n
c
e
.
IEEE
In
te
rn
a
ti
o
n
a
l
Con
fe
re
n
c
e
o
n
A
c
o
u
s
ti
c
s
,
S
p
e
e
c
h
,
a
n
d
Si
g
n
a
l
Pro
c
e
s
s
i
n
g
.
Ph
o
e
n
i
x
,
USA
.
1999
:
2
0
7
9
–
2
0
8
2
.
[2
8
]
H
Fa
rs
i
,
R
S
a
l
e
h
.
Im
p
l
e
m
e
n
ta
t
i
o
n
a
n
d
o
p
t
i
m
i
z
a
ti
o
n
o
f
a
s
p
e
e
c
h
r
e
c
o
g
n
i
t
i
o
n
s
y
s
t
e
m
b
a
s
e
d
o
n
h
i
d
d
e
n
M
a
rk
o
v
m
o
d
e
l
u
s
i
n
g
g
e
n
e
ti
c
a
l
g
o
ri
t
h
m
.
Ira
n
i
a
n
Co
n
fe
re
n
c
e
o
n
I
n
te
l
l
i
g
e
n
t
Sy
s
te
m
s
(ICIS)
.
Ba
m
,
Ira
n
.
2014
:
1
–
5.
[2
9
]
Bh
a
d
ra
g
i
ri
J
a
g
a
n
M
o
h
a
n
a
n
d
Ram
e
s
h
Ba
b
u
N.
Sp
e
e
c
h
re
c
o
g
n
i
ti
o
n
u
s
i
n
g
M
FCC
a
n
d
DT
W
.
In
te
rn
a
ti
o
n
a
l
Co
n
fe
re
n
c
e
o
n
A
d
v
a
n
c
e
s
i
n
El
e
c
tri
c
a
l
E
n
g
i
n
e
e
r
i
n
g
(I
CAE
E
)
.
Ve
l
l
o
r
e
,
In
d
i
a
.
2
0
1
4
:
1
–
4.
[3
0
]
I
M
c
L
o
u
g
h
l
i
n
.
Ap
p
l
i
e
d
Sp
e
e
c
h
a
n
d
Au
d
i
o
Pro
c
e
s
s
i
n
g
.
Cam
b
r
i
d
g
e
:
C
a
m
b
ri
d
g
e
Uni
v
e
rs
i
ty
Pre
s
s
.
2
0
0
9
;
9
7
8
0
5
2
1
5
1
9
.
[3
1
]
J
Y
a
n
g
,
J
Y
a
n
g
.
G
e
n
e
r
a
l
i
z
e
d
K
–
L
tra
n
s
fo
r
m
b
a
s
e
d
c
o
m
b
i
n
e
d
fe
a
tu
r
e
e
x
tra
c
ti
o
n
.
P
a
tt
e
r
n
Rec
o
g
n
i
t
.
2002
;
3
5
(1
)
:
2
9
5
–
2
9
7
.
[3
2
]
I
G
u
y
o
n
,
A El
i
s
s
e
e
ff
.
An
I
n
tro
d
u
c
ti
o
n
t
o
F
e
a
tu
re
Ex
tra
c
ti
o
n
.
F
e
a
tu
re
E
x
tra
c
ti
o
n
.
2
0
0
6
;
2
0
7
:
1
–
25.
[3
3
]
P
M
o
tl
i
c
e
k
.
Fe
a
tu
re
Ex
tra
c
ti
o
n
i
n
Sp
e
e
c
h
Cod
i
n
g
a
n
d
Re
c
o
g
n
i
ti
o
n
.
Rep
o
r
t
o
f
Ph
D
re
s
e
a
r
c
h
i
n
te
rn
s
h
i
p
i
n
ASP
G
ro
u
p
.
2
0
0
3
:
1
–
50.
[3
4
]
G
SVS
Si
v
a
r
am
,
H
Herm
a
n
s
k
y
.
Sp
a
rs
e
M
u
l
ti
l
a
y
e
r
Pe
rc
e
p
tro
n
fo
r
Ph
o
n
e
m
e
R
e
c
o
g
n
i
t
i
o
n
.
IE
EE
Tra
n
s
.
Au
d
i
o
.
S
p
e
e
c
h
.
L
a
n
g
.
Pro
c
e
s
s
i
n
g
.
2
0
1
2
;
2
0
(
1
):
2
3
–
2
9
.
[3
5
]
L
M
u
d
a
,
M
Be
g
a
m
,
I
El
a
m
v
a
z
u
th
i
.
V
o
i
c
e
Re
c
o
g
n
i
t
i
o
n
Al
g
o
ri
th
m
s
u
s
i
n
g
M
e
l
Fre
q
u
e
n
c
y
Cep
s
tra
l
Coe
ff
i
c
i
e
n
t
(M
FCC) a
n
d
Dy
n
a
m
i
c
T
i
m
e
W
a
r
p
i
n
g
(DT
W
) T
e
c
h
n
i
q
u
e
s
.
J
.
C
o
m
p
u
t.
2
0
1
0
;
2
(3
):
1
3
8
–
1
4
3
.
[3
6
]
A
Při
b
i
l
o
v
á
.
Pre
e
m
p
h
a
s
i
s
i
n
fl
u
e
n
c
e
o
n
h
a
rm
o
n
i
c
s
p
e
e
c
h
m
o
d
e
l
w
i
th
a
u
to
re
g
r
e
s
s
i
v
e
p
a
ra
m
e
te
ri
z
a
ti
o
n
.
Rad
i
o
e
n
g
i
n
e
e
ri
n
g
.
2
0
0
3
;
1
2
(
3
)
:
3
2
–
3
6
.
[3
7
]
R
G
o
l
d
b
e
rg
,
L
Ri
e
k
.
A
p
ra
c
ti
c
a
l
h
a
n
d
b
o
o
k
o
f
s
p
e
e
c
h
c
o
d
e
r
s
,
1
s
t
e
d
.
Bo
c
a
Ra
t
o
n
:
CR
C Pre
s
s
,
2
0
0
0
.
[3
8
]
M
A
Sa
m
a
d
.
A
Nov
e
l
W
i
n
d
o
w
Fu
n
c
t
i
o
n
Y
i
e
l
d
i
n
g
Su
p
p
re
s
s
e
d
M
a
i
n
l
o
b
e
W
i
d
th
a
n
d
M
i
n
i
m
u
m
Si
d
e
l
o
b
e
Pe
a
k
.
In
t.
J
.
Com
p
u
t
.
Sc
i
.
En
g
.
I
n
f.
T
e
c
h
n
o
l
.
2
0
1
2
;
2
(
2
):
9
1
–
1
0
3
.
[3
9
]
KR
G
h
ule
,
RR
Des
h
m
u
k
h
.
Fe
a
tu
re
-
Ex
tra
c
ti
o
n
-
T
e
c
h
n
i
q
u
e
s
-
f
o
r
-
Sp
e
e
c
h
-
Re
c
o
g
n
i
ti
o
n
-
A
-
Rev
i
e
w
.
In
t.
J
.
Sc
i
.
En
g
.
Re
s
.
2
0
1
5
;
6
(5
):
1
4
3
–
147.
[4
0
]
SC
J
o
s
h
i
,
AN
Ch
e
e
ra
n
.
M
AT
L
AB
Ba
s
e
d
F
e
a
tu
r
e
Ex
tra
c
ti
o
n
Us
i
n
g
M
e
l
Fre
q
u
e
n
c
y
Cep
s
tr
u
m
Coe
ff
i
c
i
e
n
ts
f
o
r
Au
t
o
m
a
ti
c
Sp
e
e
c
h
Rec
o
g
n
i
ti
o
n
.
In
t.
J
.
Sc
i
.
E
n
g
.
T
e
c
h
n
o
l
.
Res
.
2
0
1
4
;
3
(6
):
1820
–
1
8
2
3
.
[4
1
]
SE
L
e
v
i
n
s
o
n
.
M
a
th
e
m
a
ti
c
a
l
M
o
d
e
l
s
f
o
r
S
p
e
e
c
h
Te
c
h
n
o
l
o
g
y
.
Chi
c
h
e
s
te
r
,
UK:
J
o
h
n
W
i
l
e
y
&
So
n
s
,
L
t
d
.
2005.
[4
2
]
S
Dhi
n
g
ra
,
G
Nij
h
a
w
a
n
,
P
Pa
n
d
i
t.
I
s
o
l
a
te
d
s
p
e
e
c
h
re
c
o
g
n
i
t
i
o
n
u
s
i
n
g
M
FCC
a
n
d
DT
W
.
In
t
e
rn
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Ad
v
a
n
c
e
d
Res
e
a
r
c
h
i
n
El
e
c
tri
c
a
l
,
El
e
c
tro
n
i
c
s
a
n
d
In
s
tr
u
m
e
n
ta
t
i
o
n
En
g
i
n
e
e
r
i
n
g
.
2
0
1
3
;
2
(8
):
4085
–
4
0
9
2
.
[4
3
]
S
Be
d
i
,
R
Si
n
g
h
.
Ham
m
i
n
g
G
e
n
e
ra
l
i
z
e
d
L
o
w
T
i
m
e
L
i
ft
e
r
e
d
Cep
s
tra
l
An
a
l
y
s
i
s
.
In
t.
J
.
Ad
v
.
Re
s
.
Com
p
u
t.
Sc
i
.
So
ft
w.
En
g
.
2
0
1
7
;
7
(6
):
1
5
7
–
160.
[4
4
]
NA
M
i
c
h
a
e
l
.
S
h
o
rt
-
T
i
m
e
Sp
e
c
t
ru
m
a
n
d
‘
Ce
p
s
tru
m
’
T
e
c
h
n
i
q
u
e
s
fo
r
Vo
c
a
l
-
Pi
t
c
h
Det
e
c
t
i
o
n
.
J
.
Ac
o
u
s
t
.
So
c
.
Am
.
1
9
6
4
;
3
6
(2
):
2
9
6
–
3
0
2
.
[4
5
]
RM
M
a
rti
n
,
CL
Bu
rl
e
y
.
P
o
wer
Cep
s
tru
m
Te
c
h
n
i
q
u
e
W
i
t
h
Ap
p
l
i
c
a
t
i
o
n
to
M
o
d
e
l
Hel
i
c
o
p
t
e
r
Ac
o
u
s
t
i
c
Dat
a
.
NASA T
e
c
h
n
i
c
a
l
Pa
p
e
r
2
5
8
6
.
1
9
8
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.