I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
9
,
No
.
6
,
Dec
em
b
er
201
9
,
p
p
.
4
6
8
4
~4
6
9
5
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v9
i
6
.
p
p
4
6
8
4
-
4695
4684
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e
.
co
m/
jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
I
JE
C
E
Co
ntinuou
s
k
a
nn
a
da
speech seg
m
e
ntatio
n and
spe
ec
h
recog
nition
ba
sed
on thresho
ld
u
sin
g
M
FCC
a
nd V
Q
Va
na
j
a
k
s
h
i P
utt
a
s
w
a
m
y
G
o
w
da
,
M
a
t
hiv
a
na
n
M
urug
a
v
e
lu
,
SenthilK
u
m
a
ra
n T
ha
ng
a
m
u
t
hu
A
CS
Co
ll
e
g
e
o
f
En
g
in
e
e
rin
g
,
V
is
v
e
s
w
a
r
a
y
a
T
e
c
h
n
o
lo
g
ica
l
Un
iv
e
rs
it
y
,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
1
9
,
2
0
1
8
R
ev
i
s
ed
J
u
n
2
0
,
2
0
1
9
A
cc
ep
ted
J
u
n
30
,
2
0
1
9
Co
n
ti
n
u
o
u
s
s
p
e
e
c
h
se
g
m
e
n
tatio
n
a
n
d
it
s
re
c
o
g
n
i
ti
o
n
is p
lay
in
g
i
m
p
o
rtan
t
r
o
le
in
n
a
t
u
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
in
g
.
Co
n
ti
n
u
o
u
s
c
o
n
tex
t
b
a
se
d
Ka
n
n
a
d
a
sp
e
e
c
h
se
g
m
e
n
tatio
n
d
e
p
e
n
d
s
o
n
c
o
n
tex
t,
g
ra
m
m
e
r
a
n
d
se
m
a
n
ti
c
s
ru
les
p
re
se
n
t
i
n
th
e
k
a
n
n
a
d
a
lan
g
u
a
g
e
.
T
h
e
sig
n
i
f
ica
n
t
f
e
a
tu
re
e
x
tra
c
ti
o
n
o
f
k
a
n
n
a
d
a
sp
e
e
c
h
sig
n
a
l
f
o
r
re
c
o
g
n
it
io
n
sy
ste
m
is
q
u
it
e
e
x
c
it
in
g
f
o
r
re
se
a
rc
h
e
rs.
In
th
is
p
a
p
e
r
p
ro
p
o
se
d
m
e
th
o
d
is
d
iv
id
e
d
i
n
t
o
tw
o
p
a
rts.
F
irst
p
a
rt
o
f
th
e
m
e
th
o
d
is
c
o
n
ti
n
u
o
u
s
k
a
n
n
a
d
a
sp
e
e
c
h
sig
n
a
l
se
g
m
e
n
tatio
n
w
it
h
re
sp
e
c
t
to
th
e
c
o
n
tex
t
b
a
se
d
is
c
a
rried
o
u
t
b
y
c
o
m
p
u
ti
n
g
a
v
e
r
a
g
e
sh
o
rt
term
e
n
e
rg
y
a
n
d
it
s
sp
e
c
tral
c
e
n
tro
id
c
o
e
f
f
icie
n
ts
o
f
th
e
sp
e
e
c
h
sig
n
a
l
p
re
se
n
t
in
th
e
sp
e
c
if
ied
w
in
d
o
w
.
T
h
e
se
g
m
e
n
ted
o
u
tp
u
ts
a
re
c
o
m
p
lete
l
y
m
e
a
n
in
g
f
u
l
se
g
m
e
n
tatio
n
f
o
r
d
if
fe
re
n
t
sc
e
n
a
rio
s
w
it
h
les
s
se
g
m
e
n
tatio
n
e
rro
r.
T
h
e
se
c
o
n
d
p
a
rt
o
f
th
e
m
e
th
o
d
is
sp
e
e
c
h
re
c
o
g
n
it
io
n
b
y
e
x
tra
c
ti
n
g
les
s
n
u
m
b
e
r
M
e
l
f
r
e
q
u
e
n
c
y
c
e
p
stra
l
c
o
e
ff
icie
n
ts
w
it
h
les
s
n
u
m
b
e
r
o
f
c
o
d
e
b
o
o
k
s
u
si
n
g
v
e
c
to
r
q
u
a
n
ti
z
a
ti
o
n
.
In
th
is
re
c
o
g
n
it
io
n
is
c
o
m
p
lete
ly
b
a
se
d
o
n
th
re
sh
o
l
d
v
a
lu
e
.
T
h
is
th
re
sh
o
l
d
se
tt
in
g
is
a
c
h
a
ll
e
n
g
in
g
tas
k
h
o
we
v
e
r
th
e
si
m
p
le
m
e
th
o
d
is
u
se
d
to
a
c
h
iev
e
b
e
tt
e
r
re
c
o
g
n
it
i
o
n
ra
te.T
h
e
e
x
p
e
rim
e
n
tal
re
su
lt
s
sh
o
w
s
m
o
re
e
f
ficie
n
t
a
n
d
e
ffe
c
ti
v
e
se
g
m
e
n
tatio
n
w
it
h
h
ig
h
re
c
o
g
n
it
io
n
ra
te
f
o
r
a
n
y
c
o
n
ti
n
u
o
u
s
c
o
n
tex
t
b
a
se
d
k
a
n
n
a
d
a
sp
e
e
c
h
s
ig
n
a
l
w
it
h
d
iffere
n
t
a
c
c
e
n
ts
f
o
r
m
a
le
a
n
d
f
e
m
a
le
th
a
n
th
e
e
x
isti
n
g
m
e
th
o
d
s
a
n
d
a
lso
u
se
d
m
in
im
a
l
f
e
a
tu
re
d
im
e
n
sio
n
s
f
o
r
train
in
g
d
a
ta.
K
ey
w
o
r
d
s
:
F
e
a
tu
re
e
x
trac
ti
o
n
M
in
im
u
m
d
istan
c
e
S
h
o
rt
term
e
n
e
rg
y
S
p
e
c
tral
c
e
n
tro
i
d
S
p
e
e
c
h
re
c
o
g
n
it
i
o
n
S
p
e
e
c
h
se
g
m
e
n
tatio
n
Zero
c
ro
ss
in
g
ra
te
Co
p
y
rig
h
t
©
2
0
1
9
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
P
.
Van
aj
ak
s
h
i,
R
esear
ch
Sch
o
lar
,
AC
S
C
o
lleg
e
o
f
E
n
g
in
ee
r
i
n
g
,
Vis
v
e
s
w
ar
a
y
a
T
ec
h
n
o
lo
g
ical
Un
i
v
er
s
it
y
,
I
n
d
ia
.
E
m
ail:
v
an
aj
a_
g
o
w
d
a
@
r
ed
if
f
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Sp
ee
ch
s
e
g
m
e
n
tatio
n
w
it
h
r
ec
o
g
n
itio
n
is
ch
al
len
g
i
n
g
task
n
o
w
a
d
a
y
s
.
P
r
o
p
o
s
e
d
s
p
ee
ch
s
eg
m
e
n
tatio
n
i
n
th
i
s
p
ap
er
is
th
e
p
r
o
ce
s
s
es
o
f
is
o
lati
n
g
t
h
e
s
p
ee
ch
s
i
g
n
al
w
it
h
co
n
te
x
t
b
ased
in
f
o
r
m
a
tio
n
o
f
th
e
p
ar
tic
u
lar
s
ce
n
ar
io
.
T
h
e
s
eg
m
e
n
tatio
n
o
f
th
e
s
p
ee
c
h
b
ased
o
n
co
n
te
x
t
is
to
id
en
t
if
y
e
n
d
p
o
in
t
s
o
f
th
e
co
n
tex
t
in
f
o
r
m
a
tio
n
b
ased
o
n
w
o
r
d
s
,
s
y
llab
les
o
r
p
h
o
n
e
m
es
in
n
at
u
r
al
la
n
g
ag
e
s
.
T
h
e
s
p
ee
ch
s
e
g
m
e
n
tat
io
n
is
a
s
u
b
p
ar
t
o
f
s
p
ee
ch
r
ec
o
g
n
i
tio
n
s
y
s
te
m
.
I
n
n
at
u
ar
al
la
n
g
u
ag
e
p
r
o
ce
s
s
in
g
co
n
tex
t,
s
e
m
a
n
tics
,
a
n
d
g
r
a
m
m
er
ar
e
v
er
y
i
m
p
o
r
tan
t
to
r
ec
o
g
n
iz
e
th
e
s
p
ee
ch
.
T
h
e
ap
p
licatio
n
s
o
f
au
to
m
a
tic
s
p
ee
ch
r
ec
o
g
n
iti
o
n
s
y
s
te
m
ar
e
u
s
e
d
in
b
r
o
ad
ca
s
t
n
e
w
s
tr
an
s
cr
ip
ti
o
n
,
in
f
o
r
m
atio
n
ex
tr
ac
tio
n
an
d
r
etr
iev
al,
id
en
tit
y
th
e
s
p
ea
k
er
s
v
o
ice
an
d
allo
w
th
e
au
t
h
e
n
ticated
u
s
er
to
u
tili
z
e
th
e
s
er
v
ice
s
an
d
m
an
y
m
o
r
e.
Hig
h
er
lag
is
m
et
h
o
d
is
u
s
e
d
to
ex
tr
ac
t
th
e
f
ea
tu
r
es
o
f
s
p
ee
ch
s
ig
n
al
w
it
h
lin
ea
r
p
r
ed
ictio
n
[
1
]
.
T
h
e
m
et
h
o
d
g
i
v
es
t
w
o
p
r
ed
ictio
n
er
r
o
r
s
,
o
n
e
is
th
e
o
r
d
in
ar
y
co
n
v
e
n
tio
n
li
n
ea
r
p
r
ed
ictio
n
a
n
d
o
th
er
o
n
e
i
s
th
e
d
ela
y
ed
v
er
s
io
n
w
i
th
k
n
u
m
b
er
o
f
s
a
m
p
les
o
f
lin
ea
r
p
r
ed
ictio
n
.
Fu
r
t
h
er
C
o
m
b
i
n
ed
Hig
h
e
r
L
ag
L
in
ea
r
P
r
ed
ictio
n
(
C
HL
L
P
)
m
o
d
el
s
i
m
u
lta
n
eo
u
s
l
y
b
y
ze
r
o
lag
an
d
h
ig
h
er
lag
p
r
ed
ictio
n
.
I
n
C
H
L
L
P
m
o
d
el
th
e
co
s
t
f
u
n
ctio
n
C
H
L
L
P
m
o
d
el
is
eq
u
al
to
th
e
co
s
t
f
u
n
ct
io
n
o
f
co
n
v
en
t
io
n
al
li
n
ea
r
p
r
ed
ictio
n
if
s
ig
n
a
l
is
co
m
p
letel
y
p
er
io
d
ic
w
it
h
f
u
n
d
a
m
e
n
tal
p
er
io
d
len
g
t
h
eq
u
al
to
P
n
u
m
b
er
o
f
s
a
m
p
le
s
an
d
i
f
m
n
u
m
b
er
o
f
s
a
m
p
les
s
elec
ted
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
o
n
tin
u
o
u
s
ka
n
n
a
d
a
s
p
ee
ch
s
eg
men
ta
tio
n
a
n
d
s
p
ee
c
h
r
ec
o
g
n
itio
n
b
a
s
ed
…
(
V
a
n
a
ja
ksh
i P
u
tta
s
w
a
my
Go
w
d
a
)
4685
f
r
o
m
P
.
I
n
C
H
L
L
P
i
f
ze
r
o
lag
p
r
ed
ictio
n
m
ea
n
s
m
=
0
,
s
i
m
u
l
tan
eo
u
s
l
y
w
eig
h
ted
b
y
s
ca
lar
.
So
th
is
m
et
h
o
d
is
a
n
e
w
s
p
ec
tr
al
m
o
d
eli
n
g
m
eth
o
d
ca
n
b
e
u
s
ed
as
a
m
o
d
er
n
s
p
ea
k
er
r
ec
o
g
n
itio
n
s
y
s
te
m
w
i
th
d
if
f
er
e
n
t
n
o
i
s
y
en
v
ir
o
n
m
e
n
t
s
.
Sp
ea
k
er
in
d
ep
en
d
e
n
t
f
o
r
T
el
u
g
u
la
n
g
u
a
g
e
f
o
r
co
n
ti
n
u
o
u
s
w
o
r
d
s
M
FC
C
(
m
e
l
f
r
eq
u
en
c
y
ce
p
s
tr
a
l
co
ef
f
icie
n
t
s
)
[
2
]
an
d
DW
T
ar
e
u
s
ed
to
e
x
tr
ac
t
th
e
f
ea
t
u
r
es
o
f
co
n
t
in
u
o
u
s
s
p
ee
ch
t
h
en
clas
s
i
f
ied
t
h
es
e
ex
tr
ac
ted
f
ea
t
u
r
es
u
s
i
n
g
HM
M
b
ased
n
eu
r
al
n
et
w
o
r
k
.
T
h
e
MFC
C
an
d
DW
T
co
m
b
i
n
ed
m
et
h
o
d
s
ar
e
u
s
e
f
u
l
to
ex
tr
ac
t
f
ea
tu
r
es
o
f
t
h
e
s
p
ee
ch
s
ig
n
al.
W
av
elet
p
ac
k
et
d
ec
o
m
p
o
s
itio
n
ar
e
u
s
ed
to
r
e
m
o
v
e
t
h
e
n
o
i
s
e
p
r
esen
t
i
n
th
e
f
ea
t
u
r
e
o
f
th
e
s
p
ee
ch
s
i
g
n
a
l.D
W
T
is
o
late
th
e
h
ig
h
er
a
n
d
l
o
w
f
r
eq
u
e
n
c
y
b
an
d
s
p
r
ese
n
t i
n
th
e
s
p
ee
ch
s
i
g
n
a
l.
So
h
ig
h
f
r
eq
u
en
c
y
b
an
d
s
ar
e
co
n
s
id
er
ed
as
a
u
s
e
f
u
l
f
ea
t
u
r
es.
T
h
is
p
r
o
ce
s
s
g
iv
es
lo
w
w
o
r
d
er
r
o
r
r
ate
in
s
p
ee
ch
r
ec
o
g
n
itio
n
s
y
s
te
m
.
T
h
e
m
eth
o
d
e
n
h
a
n
ce
t
h
e
w
h
i
s
p
er
r
ec
o
g
n
itio
n
[
3
]
b
y
ex
tr
ac
tin
g
a
n
e
w
r
o
b
u
s
t
ce
p
s
tr
al
f
e
atu
r
es
a
n
d
p
r
ep
r
o
ce
s
s
in
g
b
ased
o
n
d
e
m
i
s
in
g
a
u
to
en
co
d
er
.
T
ea
s
er
en
er
g
y
b
ased
ce
p
s
tr
al
f
ea
tu
r
es
ar
e
m
o
r
e
r
o
b
u
s
t
t
h
a
n
MFC
C
f
o
r
w
h
is
p
er
ed
d
escr
ip
tio
n
DD
A
E
R
P
E
C
C
f
ea
t
u
r
e
ex
tr
ac
tio
n
s
ig
n
i
f
ica
n
tl
y
i
m
p
r
o
v
es
th
e
r
ec
o
g
n
itio
n
r
ate
co
m
p
ar
e
MF
C
C
,
G
MM
,
HM
M.
T
h
is
p
r
o
p
o
s
ed
m
et
h
o
d
i
m
p
r
o
v
es
m
o
r
e
th
a
n
3
1
%
th
a
n
tr
ad
itio
n
a
l
m
et
h
o
d
s
.
T
y
p
icall
y
a
u
to
en
co
d
er
h
as
in
p
u
t
la
y
er
w
h
ic
h
is
t
h
e
o
r
ig
in
al
f
ea
tu
r
e
v
ec
to
r
o
n
e
o
r
m
o
r
e
h
id
d
en
la
y
er
w
h
ic
h
ar
e
T
r
an
s
f
o
r
m
ed
f
ea
tu
r
es
o
u
t
o
f
t
h
o
s
e
h
id
d
en
la
y
e
r
w
h
ic
h
m
atc
h
es
i
n
p
u
t
la
y
er
f
o
r
r
ec
o
n
s
tr
u
ctio
n
.
T
E
C
C
b
ased
f
ea
t
u
r
es
p
r
ed
icts
t
h
e
f
ac
ts
t
h
at
T
eo
p
er
ato
r
a
n
d
g
a
m
m
et
o
n
e
f
il
ter
b
an
k
t
o
d
escr
ib
e
w
h
i
s
h
p
er
ch
ar
ac
ter
s
tics
.
So
,
b
ec
au
s
e
o
f
t
h
ese
t
h
e
ac
h
ie
v
ed
w
o
r
d
r
ec
o
g
n
itio
n
r
ate
i
s
9
3
%.
W
o
r
d
b
o
u
n
d
ar
y
d
etec
tio
n
is
u
s
ed
to
s
ep
ar
ate
th
e
w
o
r
d
f
r
o
m
Gu
j
ar
at
Sp
ee
ch
.
T
h
is
p
ap
er
ac
h
iev
e
s
e
n
d
p
o
in
t
d
etec
tio
n
[
4
]
in
Gu
j
ar
ath
s
p
ee
ch
r
ec
o
g
n
itio
n
s
y
s
te
m
w
i
th
t
h
e
p
r
ese
n
ce
o
f
b
ac
k
g
r
o
u
n
d
n
o
is
e.
I
t
s
ep
ar
ates
th
e
s
ilen
t p
o
r
tio
n
s
o
f
t
h
e
s
p
ee
ch
.
So
t
h
at
n
o
is
e
i
s
r
ed
u
ce
d
.
T
h
i
s
w
o
r
d
b
o
u
n
d
ar
y
d
etec
tio
n
u
s
e
s
t
w
o
al
g
o
r
ith
m
to
d
etec
t
an
d
p
o
in
t
ex
p
licitl
y
a
n
d
i
m
p
licitl
y
.
E
x
p
lic
it
e
n
d
p
o
in
t
d
etec
tio
n
u
s
ed
ativ
e
b
ef
o
r
e
r
ec
o
g
n
itio
n
an
d
i
m
p
licit
e
n
d
p
o
in
ts
ar
e
u
s
ed
a
f
ter
s
p
ee
ch
p
r
o
ce
s
s
to
d
etec
t
en
d
p
o
in
t.th
is
m
et
h
o
d
ab
le
to
d
etec
t
w
ea
k
f
r
icati
v
e
in
s
i
g
n
al
to
n
o
is
e
r
atio
co
n
d
iti
o
n
.
T
h
e
H
y
b
r
id
is
m
o
d
el
u
s
ed
f
o
r
m
ax
i
m
u
m
Ga
u
s
s
ia
n
m
i
x
t
u
r
e
c
o
n
tin
u
o
u
s
T
a
m
il
Sp
ee
ch
r
ec
o
g
n
itio
n
[
5
]
.
T
h
is
m
et
h
o
d
i
m
p
r
o
v
e
s
ac
cu
r
a
c
y
u
p
to
3
%
er
r
o
r
r
ate
u
p
to
4
%
co
m
p
ar
e
to
th
e
e
x
i
s
ti
n
g
s
y
s
te
m
.
T
h
i
s
m
o
d
el
i
s
u
s
ed
i
n
s
p
ee
ch
to
te
x
t
co
n
v
er
s
io
n
i
n
v
ar
io
u
s
ap
p
licatio
n
.
I
n
t
h
is
L
P
C
,
MFC
C
,
L
P
a
r
e
u
s
ed
to
e
x
tr
ac
t
th
e
f
ea
t
u
r
e
s
.
T
h
is
m
e
th
o
d
i
s
an
u
n
s
u
p
er
v
is
ed
m
eth
o
d
t
o
an
al
y
s
is
o
f
d
ata
an
d
co
n
s
tr
u
ctio
n
m
o
d
elin
g
.
So
th
is
p
o
r
tio
n
d
ata
p
o
in
ts
b
et
w
ee
n
ze
r
o
an
d
o
n
e.
T
h
ese
v
al
u
es
ar
e
as
s
ig
n
ed
b
ased
o
n
th
e
clu
s
ter
s
,
ce
n
tr
e
an
d
d
ata
p
o
in
ts
T
o
r
ec
o
g
n
ize
is
o
lated
Ka
n
n
a
d
a
w
o
r
d
s
tr
ai
n
ed
HM
M
m
o
d
el
an
d
v
iter
b
i
al
g
o
r
ith
m
f
o
r
d
ec
o
d
in
g
p
r
o
ce
s
s
[
6
]
.
MFC
C
ar
e
co
m
p
u
ted
in
f
r
o
n
ten
d
p
r
o
ce
s
s
i
n
g
.
T
h
is
p
r
o
p
o
s
es to
co
m
p
ar
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
p
h
o
n
e
lev
el
a
n
d
s
y
llab
le
lev
el
ac
o
u
s
tic
m
o
d
el
f
o
r
s
m
all
to
m
ed
i
u
m
s
ized
k
a
n
n
d
a
la
n
g
u
a
g
e
v
o
ca
b
u
lar
y
.
Av
er
ag
e
w
o
r
d
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
9
7
%
f
o
r
s
y
llab
le
lev
e
l
m
o
d
elin
g
,
9
8
.
6
%
f
o
r
p
h
o
n
e
lev
el
m
eo
d
elin
g
.
Sp
ee
ch
co
d
in
g
s
etu
p
h
as b
ee
n
d
o
n
e
u
s
i
n
g
HT
K
to
o
l.
T
h
e
en
tire
d
atab
ase
tr
ain
in
g
an
d
test
in
g
s
a
m
p
les ar
e
u
s
ed
to
b
u
ild
b
y
u
s
i
n
g
HM
M.
C
o
f
u
s
io
n
m
atr
ix
i
s
u
s
ed
to
an
al
y
s
e
an
d
in
ter
p
r
et
th
e
r
es
u
lt
s
at
th
e
w
o
r
d
lev
el.
HM
M
an
d
No
r
m
a
l
Fi
t
[
7
]
m
e
th
o
d
is
u
s
ed
f
o
r
co
n
ti
n
u
o
u
s
s
p
ee
ch
r
ec
o
g
n
itio
n
.
Vo
ice
d
etec
t
io
n
b
ased
o
n
co
m
p
u
tin
g
d
y
n
a
m
ic
th
r
es
h
o
ld
an
d
ce
p
s
tr
u
m
co
ef
f
icie
n
t
s
ar
e
ex
tr
ac
ted
as
a
f
ea
tu
r
e
o
f
v
o
ice.
T
h
e
B
au
m
-
W
alsh
alg
o
r
it
h
m
i
s
u
s
ed
f
o
r
tr
ain
ed
d
atab
ase
an
d
No
r
m
al
Fit
te
h
n
iq
u
e
is
u
s
ed
to
lab
el
th
e
s
p
ee
ch
.
T
h
is
m
eth
o
d
test
ed
f
o
r
f
i
v
e
l
an
g
u
a
g
es.
I
n
an
a
v
er
ag
e
ac
cu
r
atio
n
r
ate
9
5
%.
T
h
e
ex
p
er
im
en
tal
r
esu
lts
s
h
o
ws
th
at
s
ize
o
f
m
e
m
o
r
y
r
ed
u
ce
s
b
ec
au
s
e
o
f
No
r
m
al
Fit
v
al
u
es.
T
h
e
MFC
C
is
u
s
ed
[
8
]
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
f
o
r
tr
ain
in
g
d
ata.
T
h
e
Vec
to
r
Qu
an
tizatio
n
is
u
s
ed
f
o
r
clu
s
ter
i
n
g
Sp
ea
k
er
I
n
d
ep
en
d
e
n
t
Kan
n
ad
a
Sp
ee
ch
R
ec
o
g
n
iti
o
n
.
VQ
1
a
n
d
VQ
2
is
u
s
ed
f
o
r
clu
s
ter
i
n
g
p
u
r
p
o
s
e.
T
h
e
Sp
ee
ch
R
ec
o
g
n
itio
n
er
r
o
r
d
ec
r
ea
s
es
f
r
o
m
2
.
5
to
1
.
5
.
I
n
ca
s
e
o
f
VQ
1
a
n
d
VQ
2
ar
e
th
e
t
w
o
clu
s
ter
i
n
g
tech
n
iq
u
es,
VQ
1
b
ased
o
n
b
i
n
ar
y
s
p
litt
i
n
g
al
g
o
r
ith
m
an
d
VQ
2
b
ased
o
n
lar
g
es
t
a
v
er
ag
e
d
is
to
r
tio
n
.
A
p
p
li
n
g
L
i
n
ea
r
d
is
cr
i
m
i
n
an
t
a
n
al
y
s
is
(
L
D
A
)
[
9
]
a
n
d
m
ax
i
m
u
m
li
k
eli
h
o
o
d
tr
an
s
f
o
r
m
atio
n
o
n
MF
C
C
to
e
x
tr
ac
t
f
ea
t
u
r
es
o
f
s
p
ee
ch
an
d
i
n
p
u
t
t
h
ese
f
ea
t
u
r
es
to
C
o
n
v
o
lu
tio
n
Ne
u
r
al
N
et
w
o
r
k
(
C
NN)
to
i
m
p
r
o
v
e
r
o
b
u
s
t
n
ess
o
f
s
p
ee
ch
r
ec
o
g
n
itio
n
.
T
h
is
i
m
p
r
o
v
es t
h
e
r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
.
T
h
e
p
r
o
p
o
s
e
d
m
eth
o
d
o
r
g
an
iz
ed
in
to
t
w
o
p
ar
ts
;
1
)
Sp
ee
ch
s
eg
m
e
n
tatio
n
2
)
R
ec
o
g
n
itio
n
;
T
h
e
f
ir
s
t
p
ar
t
o
f
th
e
p
ap
er
is
co
n
ti
n
u
o
u
s
k
a
n
n
ad
a
s
p
ee
ch
s
eg
m
e
n
tat
io
n
b
ased
o
n
co
n
tex
t
an
d
is
o
l
ate
k
a
n
n
ad
a
le
tter
s
f
r
o
m
co
n
ti
n
u
o
u
s
k
a
n
n
ad
a
s
p
e
ec
h
w
h
ic
h
co
n
tain
s
o
n
l
y
k
a
n
n
ad
a
letter
s
s
p
ee
c
h
s
ig
n
al.
T
h
i
s
ca
n
b
e
ac
h
ie
v
ed
b
y
d
etec
tio
n
o
f
v
o
iced
an
d
u
n
v
o
i
ce
d
s
p
ee
ch
s
ig
n
al
b
as
ed
o
n
co
m
p
u
ti
n
g
t
h
e
av
er
a
g
e
en
er
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
o
f
ea
ch
f
r
a
m
e
o
f
t
h
e
Ka
n
n
ad
a
s
p
ee
ch
s
i
g
n
al.
Av
er
ag
e
e
n
e
r
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
co
ef
f
icien
ts
ar
e
f
u
t
h
er
su
b
j
ec
ted
to
a
m
ed
ia
n
f
ilter
.
T
h
e
o
u
tp
u
t
o
f
t
h
e
m
ed
ian
f
ilter
co
ef
f
icien
t
ar
e
u
s
ed
to
s
et
th
e
th
er
s
h
o
ld
s
.
T
h
ese
th
r
es
h
o
ld
s
ar
e
u
s
ed
to
s
eg
m
e
n
t
th
e
co
n
ti
n
u
o
u
s
Ka
n
n
ad
a
s
p
ee
ch
s
ig
n
al
b
ased
o
n
c
o
n
tex
t.
T
h
e
s
ec
o
n
d
p
ar
t
o
f
th
e
p
ap
er
is
to
d
eter
m
i
n
e
th
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
o
f
th
e
s
eg
m
en
ted
s
p
ee
ch
s
i
g
n
al
u
s
i
n
g
t
h
r
es
h
o
ld
b
ased
MFC
C
a
n
d
VQ
i
n
a
n
a
u
to
m
a
tic
s
p
ee
ch
r
ec
o
g
n
i
tio
n
(
ASR
)
s
y
s
te
m
.
T
h
e
th
r
e
s
h
o
ld
b
ased
MFC
C
a
n
d
VQ
i
s
u
s
ed
to
tr
ain
s
p
ee
ch
d
ata
s
et.
T
h
e
m
e
th
o
d
s
u
s
es
less
n
u
m
b
e
r
o
f
MFC
C
an
d
le
s
s
n
u
m
b
er
o
f
co
d
eb
o
o
k
o
f
VQ
g
iv
e
s
b
etter
r
esu
lt
s
t
h
an
t
h
e
ex
is
ti
n
g
m
et
h
o
d
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
9
,
No
.
6
,
Dec
em
b
er
2
0
1
9
:
4
6
8
4
-
4
6
9
5
4686
2.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
T
h
e
f
ir
s
t
p
ar
t
o
f
t
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
co
n
tai
n
s
co
n
ti
n
u
o
u
s
co
n
te
x
t
b
ased
Ka
n
n
a
d
a
s
p
ee
ch
s
eg
m
e
n
tatio
n
an
d
is
o
lated
Ka
n
n
ad
a
Ak
s
h
ar
a
(
m
ea
n
s
letter
)
f
r
o
m
co
n
ti
n
u
o
u
s
Kan
n
ad
a
s
p
e
ec
h
w
h
ic
h
co
n
tain
s
th
e
u
tter
an
ce
o
f
Kan
n
ad
a
Ak
s
h
ar
a
o
n
l
y
.
S
ec
o
n
d
p
ar
t
o
f
m
et
h
o
d
d
escr
ib
es
th
e
co
n
ti
n
u
o
u
s
co
n
te
x
t
b
ased
Kan
n
ad
a
s
p
ee
ch
s
eg
m
e
n
t
s
a
n
d
is
o
lated
Kan
n
ad
a
ak
s
h
ar
a
s
p
ee
ch
r
ec
o
g
n
itio
n
s
y
s
te
m
u
s
i
n
g
t
h
r
es
h
o
ld
b
ased
MFC
C
an
d
VQ
m
eth
o
d
s
.
Sp
ee
ch
s
eg
m
e
n
tatio
n
p
ar
t
is
th
is
p
ap
er
is
ca
r
r
ied
o
u
t
b
y
d
etec
tin
g
th
e
p
r
esen
ce
o
f
t
h
e
v
o
iced
an
d
u
n
v
o
iced
s
p
ee
ch
us
i
n
g
a
v
er
a
g
e
s
h
o
r
t
ti
m
e
e
n
er
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
s
.
T
h
e
m
ed
ia
n
f
ilter
i
s
u
s
ed
to
s
m
o
o
th
e
n
th
e
a
v
er
a
g
e
s
h
o
r
t
ti
m
e
e
n
er
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
s
co
ef
f
icie
n
ts
o
f
t
h
e
s
p
ee
ch
s
ig
n
al
a
n
d
t
h
r
es
h
o
ld
h
as
b
ee
n
s
et
b
ased
o
n
t
h
e
p
r
o
b
a
b
ilit
y
d
en
s
it
y
f
u
n
ctio
n
(
p
d
f
)
o
f
t
h
e
o
u
tp
u
t
o
f
th
e
m
ed
ian
f
ilter
co
ef
f
icie
n
t
s
.
T
h
en
co
n
te
x
b
ase
d
s
p
ee
ch
s
eg
m
e
n
tatio
n
p
er
f
o
r
m
ed
b
ased
o
n
th
e
t
h
r
esh
o
ld
le
v
els
.
2
.
1
.
Av
er
a
g
e
s
ho
rt
t
i
m
e
ene
rg
y
(
ST
E
)
T
h
e
Kan
n
ad
a
s
p
ee
ch
s
ig
n
al
is
d
ec
o
m
p
o
s
ed
in
to
a
n
u
m
b
er
o
f
f
r
a
m
es
b
y
m
u
ltip
li
n
g
w
in
d
o
w
f
u
n
ct
io
n
o
f
len
g
t
h
L
u
s
i
n
g
(
1
)
.
T
h
en
ea
ch
f
r
a
m
e
av
er
a
g
e
s
h
o
r
t ti
m
e
e
n
er
g
y
[
1
0
-
11]
is
co
m
p
u
ted
u
s
i
n
g
(
2
)
.
(
)
(
)
(
)
w
s
i
s
k
w
i
k
(
1
)
w
h
er
e
()
sk
is
k
a
n
n
ad
a
s
p
ee
ch
s
i
g
n
al,
()
wi
r
ep
r
esen
ts
h
a
m
m
i
n
g
w
i
n
d
o
w
f
u
n
c
t
io
n
,
w
h
ich
i
s
s
h
i
f
te
d
ac
r
o
s
s
th
e
s
p
ee
ch
s
ig
n
al
to
o
b
tain
f
r
am
es
a
n
d
()
w
si
is
th
e
w
i
n
d
o
w
ed
s
p
ee
ch
s
i
g
n
a
l
2
1
1
()
L
w
i
E
s
i
L
(
2
)
2
.
2
.
Sp
ec
t
ra
l
ce
ntr
o
id
f
ea
t
u
re
s
T
h
e
s
p
ec
tr
al
ce
n
tr
o
id
(
SC
)
m
ea
s
u
r
es
f
r
eq
u
e
n
c
y
an
d
m
ag
n
i
t
u
d
e
o
f
t
h
e
p
ar
ticu
lar
s
p
ec
tr
al
b
in
u
s
in
g
th
e
Di
s
cr
ete
Fo
u
r
ier
T
r
an
s
f
o
r
m
.
T
h
e
s
p
ec
tr
al
ce
n
tr
o
i
d
co
n
tain
s
m
o
r
e
en
er
g
y
a
b
o
v
e
an
d
b
elo
w
th
e
f
u
n
d
e
m
a
n
tal
f
r
eq
u
e
n
c
y
,
wh
ich
i
s
al
m
o
s
t
t
h
e
a
v
er
ag
e
en
e
r
g
y
o
f
th
e
s
p
ec
tr
al
b
i
n
.
U
s
u
all
y
t
h
e
s
p
ee
ch
s
i
g
n
al
h
as
as
y
m
m
etr
ic
in
n
au
r
e
ab
o
u
t
th
e
p
itc
h
r
an
g
e.
T
h
e
ac
cu
r
ac
y
o
f
p
er
ce
p
tio
n
in
s
p
ee
ch
s
ig
n
a
l
in
t
h
e
f
o
r
m
o
f
r
a
m
p
f
u
n
ctio
n
,
s
o
th
at
it
g
iv
e
s
m
o
r
e
ac
cu
r
ate
p
er
ce
p
tio
n
in
b
o
th
lo
w
er
an
d
h
i
g
h
er
f
r
eq
u
e
n
cies
o
f
t
h
e
s
p
ec
tr
al
b
in
.
T
h
e
ea
ch
f
r
a
m
e
o
f
th
e
s
p
ec
tr
al
ce
n
tr
o
id
o
f
s
ize
N
is
d
ef
i
n
e
d
in
(
3
)
1
1
()
()
2
N
k
j
N
k
j
s
j
M
S
m
Sm
SC
f
(
3
)
W
h
er
e
()
k
Sm
is
t
h
e
F
FT
o
f
w
i
n
d
o
w
ed
s
eq
u
en
ce
o
f
t
h
e
s
p
ee
c
h
s
ig
n
al
o
f
s
ize
N
s
a
m
p
les,
(
2
)
s
f
M
N
is
th
e
w
id
th
o
f
t
h
e
ea
ch
s
p
ec
tr
al
b
in
an
d
s
f
is
th
e
s
a
m
p
lin
g
f
r
eq
u
en
c
y
o
f
th
e
s
p
ee
ch
s
i
g
n
al.
T
h
e
m
u
ltip
licat
io
n
f
ac
to
r
j
in
(
3
)
r
ef
er
s
to
th
e
p
e
r
ce
p
tio
n
o
f
s
p
ee
ch
s
i
g
n
al
as a
r
a
m
p
f
u
n
cti
o
n
.
2
.
3
.
M
edia
n
f
ilte
r
a
nd
t
hersh
o
ld s
et
t
in
g
T
h
e
m
ed
ian
f
i
lter
is
u
s
ed
f
u
th
er
to
s
m
o
o
t
h
en
an
d
r
etai
n
a
n
y
ab
r
u
p
t
ch
a
n
g
es
w
it
h
in
2
L
o
f
av
er
ag
e
en
er
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
co
ef
f
icie
n
t
s
.
W
h
er
e
L
is
t
h
e
len
g
th
o
f
f
ilter
.
I
n
th
i
s
p
ap
er
le
n
g
t
h
o
f
t
h
e
f
il
ter
is
5
.
Sin
ce
it
i
s
a
n
o
n
liear
f
ilter
i
t
w
ill
n
o
t
s
m
o
o
th
e
n
s
t
h
e
n
o
is
e
co
m
p
o
n
en
ts
p
r
ese
n
ts
i
n
t
h
e
a
v
er
ag
e
e
n
er
g
y
a
n
d
s
p
ec
tr
al
ce
n
tr
o
id
co
ef
f
icie
n
t
s
.
T
h
e
m
ed
ian
f
i
lter
o
u
tp
u
ts
ar
e
u
s
ed
t
o
s
et
t
h
e
t
h
r
es
h
o
ld
s
b
ase
d
o
n
th
e
p
r
o
b
ab
ilit
y
d
en
s
it
y
f
u
n
ctio
n
(
p
d
f
)
o
f
th
e
f
ilt
er
o
u
tp
u
t
co
ef
f
icie
n
ts
.
T
h
e
s
e
t
h
er
s
h
o
ld
s
ar
e
u
s
ed
to
id
e
n
ti
f
y
t
h
e
co
n
te
x
t
o
f
th
e
s
p
ee
c
h
s
ig
n
al
i
n
ap
p
r
o
p
r
iate
m
a
n
n
er
.
E
n
er
g
y
t
h
r
es
h
o
ld
(
E
T
)
an
d
s
p
ec
tr
al
ce
n
tr
o
id
th
r
es
h
o
ld
(
ST)
s
ettin
g
i
s
r
eq
u
ir
ed
to
s
eg
m
e
n
t
t
h
e
co
n
ti
n
u
o
u
s
k
an
n
ad
a
s
p
ee
ch
s
i
g
n
a
l.
B
o
th
th
e
t
h
r
e
s
h
o
ld
ca
n
b
e
co
m
p
u
ted
b
y
ta
k
i
n
g
th
e
h
i
s
to
g
r
a
m
o
f
th
e
ST
E
an
d
Sp
ec
tr
al
C
en
tr
o
id
o
f
ea
ch
f
r
am
e.
T
w
o
f
la
g
s
f
1
an
d
f
2
ar
e
s
ettin
g
b
y
co
m
p
ar
in
g
en
er
g
y
w
it
h
E
T
an
d
ce
n
tr
o
id
w
it
h
ST
.
Dep
en
d
in
g
o
n
t
h
e
f
i
n
al
f
la
g
,
th
e
s
p
ee
ch
s
eg
m
e
n
tati
o
n
is
ac
h
iev
ed
b
ased
o
n
th
e
co
n
tex
t
o
f
th
e
s
ce
n
ar
io
.
Fin
all
y
ea
c
h
f
r
a
m
e
o
f
th
e
s
p
e
ec
h
is
s
ep
ar
ated
w
it
h
v
o
iced
an
d
u
n
v
o
iced
s
p
ee
ch
b
ased
o
n
co
n
tex
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
o
n
tin
u
o
u
s
ka
n
n
a
d
a
s
p
ee
ch
s
eg
men
ta
tio
n
a
n
d
s
p
ee
c
h
r
ec
o
g
n
itio
n
b
a
s
ed
…
(
V
a
n
a
ja
ksh
i P
u
tta
s
w
a
my
Go
w
d
a
)
4687
2
.
4
.
Z
er
o
cr
o
s
s
i
ng
ra
t
e
a
nd
end po
int
det
ec
t
io
n
Z
er
o
C
r
o
s
s
i
n
g
R
ate
g
iv
e
s
in
f
o
r
m
at
io
n
o
f
r
ap
id
l
y
ch
a
n
g
i
n
g
o
f
th
e
s
p
ee
ch
s
ig
n
al
f
r
o
m
p
o
s
itiv
e
to
n
eg
at
iv
e.
If
m
o
r
e
n
u
m
b
er
ze
r
o
cr
o
s
s
in
g
s
m
ea
n
s
th
e
s
p
ee
ch
s
i
g
n
al
co
n
tai
n
s
th
e
h
i
g
h
f
r
eq
u
en
c
y
in
f
o
r
m
atio
n
[
1
2
]
.
I
f
it
is
les
s
th
e
s
ig
n
al
co
n
tain
lo
w
f
r
eq
u
en
c
y
in
f
o
r
m
at
io
n
.
T
h
u
s
ze
r
o
cr
o
s
s
in
g
s
i
s
u
s
ed
i
n
th
is
p
ap
er
to
id
en
ti
f
y
th
e
v
o
ic
ed
an
d
u
n
v
o
iced
s
p
ee
ch
s
ig
n
a
l
w
h
ic
h
i
s
h
elp
f
u
l
to
s
eg
m
e
n
t
s
th
e
g
iv
e
n
s
ig
n
al.
I
n
a
g
i
v
en
f
r
a
m
e
t
h
e
s
p
ee
c
h
s
i
g
n
al
i
s
co
n
s
id
er
ed
as
n
o
n
-
s
tat
io
n
ar
y
s
i
g
n
al
an
d
I
t is d
ef
in
ed
in
(
4
)
.
1
1
(
)
(
)
(
)
2
N
CR
l
Z
n
s
l
w
n
l
N
(
4
)
Z
er
o
C
r
o
s
s
i
n
g
R
at
es
(
Z
C
R
)
is
u
s
ed
to
d
etec
t
th
e
v
o
ice
a
ctiv
it
y
in
t
h
e
s
p
ee
ch
s
i
g
n
al,
th
e
s
i
g
n
a
l
w
h
et
h
er
it
is
a
s
p
ee
ch
h
as
s
p
o
k
en
v
o
ice
o
r
s
ilen
t.
T
h
e
Z
C
R
u
s
ed
in
th
is
p
ap
e
r
,
to
d
etec
t
th
e
en
d
p
o
in
t
o
f
th
e
s
p
ee
ch
s
i
g
n
al
w
it
h
in
t
h
e
co
n
tex
t.
Z
er
o
cr
o
s
s
in
g
r
ate
is
is
o
latin
g
t
h
e
letter
ex
ac
tl
y
f
r
o
m
co
n
tin
u
o
u
s
s
p
ee
ch
.
Z
er
o
cr
o
s
s
in
g
is
p
la
y
i
n
g
i
m
p
o
r
tan
t
r
o
le
in
th
is
asp
ec
t
to
s
ep
ar
ate
in
d
iv
id
u
al
letter
s
.
B
y
m
a
s
k
in
g
u
n
v
o
iced
s
p
ee
ch
is
co
n
s
id
er
ed
as
ze
r
o
an
d
v
o
iced
s
p
ee
ch
i
s
m
ai
n
tai
n
ed
as
it
is
i
n
t
h
e
o
r
ig
i
n
al
s
p
ee
c
h
s
i
g
n
al.
Fu
r
t
h
er
ea
ch
letter
s
ar
e
is
o
late
d
w
it
h
th
eir
e
n
d
p
o
in
ts
u
s
i
n
g
s
h
o
r
t ti
m
e
e
n
er
g
y
an
d
ze
r
o
cr
o
s
s
in
g
r
ates.
3.
SPEE
CH
RE
CO
G
N
I
T
I
O
N
C
o
n
te
x
t
b
ased
r
ec
o
g
n
itio
n
an
d
Kan
n
ad
a
Var
n
a
m
ala
an
d
Kan
n
d
a
alp
h
ab
et
r
ec
o
g
n
i
tio
n
s
ar
e
p
r
o
p
o
s
ed
f
r
o
m
co
n
ti
n
u
o
u
s
Kan
n
ad
a
s
p
ea
ch
s
ig
n
al.
Me
l
Fre
q
u
e
n
c
y
C
ep
s
tr
al
C
o
e
f
f
icien
ts
(
MFC
C
)
an
Vec
to
r
Qu
a
n
tis
at
io
n
(
VQ)
b
ased
f
ea
t
u
r
e
ex
tr
ac
tio
n
s
ar
e
p
r
o
p
o
s
ed
.
3
.
1
.
M
el
f
re
qu
ency
ce
ps
t
ra
l c
o
ef
f
icient
s
Me
l
Fre
q
u
en
c
y
C
ep
s
tr
al
C
o
ef
f
icien
t
s
(
MFC
C
)
is
o
n
e
o
f
t
h
e
e
f
f
icien
t
an
d
ef
f
ec
ti
v
e
s
ig
n
i
f
ica
n
t
f
ea
tu
r
e
ex
tr
ac
tio
n
m
e
th
o
d
[
1
3
-
15]
u
s
ed
in
s
p
ee
ch
r
ec
o
g
n
i
tio
n
s
y
s
te
m
.
T
h
e
Me
l
Fre
q
u
e
n
c
y
s
ca
le
i
s
n
o
n
l
in
ea
r
w
h
ic
h
r
ep
r
esen
t
s
b
ased
o
n
t
h
e
s
p
ee
c
h
f
r
eq
u
en
c
y
r
a
n
g
e.
U
s
u
all
y
n
o
n
li
n
ea
r
f
r
eq
u
en
c
y
r
a
n
g
e
ep
er
ce
p
tio
n
o
f
s
p
ee
ch
s
ig
n
al
s
ar
e
r
ep
r
esen
ted
b
y
th
e
MFC
C
co
ef
f
ic
ien
t
s
.
T
h
e
s
p
ee
ch
s
i
g
n
a
l
i
s
p
as
s
in
g
t
h
r
o
u
g
h
a
b
an
d
p
ass
f
ilter
to
o
b
tain
MFC
C
co
ef
f
i
c
ien
t
s
i
n
w
h
ic
h
h
ig
h
er
b
an
d
f
r
eq
u
en
cie
s
a
n
d
cr
itical
b
a
n
d
s
ar
e
en
h
a
n
ce
d
an
d
t
h
en
p
as
s
th
r
o
u
g
h
a
in
v
er
s
e
Fa
s
t
F
o
u
r
ier
T
r
an
s
f
o
r
m
(
F
FT
)
.
So
th
at
th
e
s
p
ee
c
h
p
er
ce
p
tio
n
an
al
y
s
is
ar
e
ac
cu
r
atel
y
co
n
s
id
er
as
a
f
ea
tu
r
e
ex
tr
ac
tio
n
f
o
r
r
ec
o
g
n
itio
n
s
y
s
te
m
.
T
h
e
co
n
tin
u
o
u
s
s
p
ee
c
h
s
i
g
n
al
is
d
i
v
id
ed
in
to
N
n
u
m
b
er
o
f
f
r
a
m
e
s
w
i
t
h
m
n
u
m
b
er
o
f
s
a
m
p
le
s
.
T
h
is
p
r
o
ce
s
s
in
g
i
s
d
o
n
e
b
y
w
i
n
d
o
w
in
g
ea
c
h
in
d
iv
id
u
al
f
r
a
m
e
w
it
h
Ha
m
m
i
n
g
w
i
n
d
o
w
tec
h
n
iq
u
e.
T
h
e
s
ig
n
al
()
si
m
u
ltip
l
y
in
g
th
e
w
i
n
d
o
w
f
u
n
ct
io
n
()
wi
to
o
b
tain
th
e
w
in
d
o
w
ed
s
p
ee
c
h
s
i
g
n
a
l
()
w
si
as g
i
v
e
n
in
(
5
)
.
(
)
(
)
(
)
,
w
s
i
s
i
w
i
1
im
(
5
)
T
h
e
f
r
eq
u
en
c
y
a
n
al
y
s
i
s
o
f
w
i
n
d
o
w
ed
s
eq
u
en
ce
i
s
co
m
p
u
ted
u
s
i
n
g
d
is
cr
e
te
Fo
u
r
ier
tr
an
s
f
o
r
m
(
DFT
)
in
(
6
)
.
2
1
(
)
(
)
j
j
k
N
N
ww
j
S
k
s
j
e
(
6
)
T
h
e
tr
ian
g
u
lar
b
an
d
o
f
f
r
eq
u
e
n
cies a
r
e
o
b
tain
ed
u
s
i
n
g
Me
l
-
f
il
ter
b
an
k
s
i
n
(
7
)
.
2
5
9
5
l
o
g
1
700
m
e
l
f
f
(
7
)
3
.
2
.
Vec
t
o
r
q
ua
ntiz
a
t
io
n (
V
Q
)
Vec
to
r
Qu
an
t
izatio
n
i
s
o
n
e
o
f
th
e
m
o
s
t
i
m
p
o
r
ta
n
t
m
eth
o
d
o
f
d
is
tan
ce
m
ea
s
u
r
e
b
et
w
ee
n
t
h
e
test
d
ata
an
d
tr
ain
ed
d
ata
s
et
in
au
to
m
a
tic
s
p
ee
ch
r
ec
o
g
n
itio
n
.
B
ased
o
n
t
h
e
m
in
i
m
u
m
d
i
s
tan
ce
m
ea
s
u
r
e
m
en
t
,
it
i
s
ea
s
y
to
r
ec
o
g
n
is
e
th
e
te
s
t
d
ata
p
r
esen
t
in
t
h
e
tr
ain
ed
d
ata
s
et.
VQ
is
th
e
o
n
e
o
f
th
e
m
et
h
o
d
to
r
e
d
u
ce
th
e
n
u
m
b
er
o
f
s
ig
n
i
f
ica
n
t
d
i
m
e
n
s
io
n
s
o
f
in
p
u
t
d
ata.
So
th
at,
it
m
atc
h
es
t
h
e
u
n
k
n
o
w
n
m
o
d
el
s
in
a
v
er
y
s
i
m
p
le
m
an
n
er
b
y
r
ed
u
cin
g
t
h
e
d
ata.
T
h
is
V
Q
al
g
o
r
ith
m
cr
ea
tes
8
n
u
m
b
er
o
f
d
i
m
en
s
io
n
s
i
n
t
h
is
p
ap
er
,
w
h
ic
h
p
r
o
d
u
ce
s
a
s
et
o
f
clu
s
ter
ce
n
ter
s
s
p
r
ea
d
th
e
d
is
t
an
ce
s
p
ac
e
d
ep
en
d
in
g
o
n
th
e
s
p
ee
ch
.
Sig
n
al
f
ea
tu
r
e
s
.
T
h
en
ca
teg
ar
i
s
e
an
y
f
ea
tu
r
e
v
ec
to
r
to
o
n
e
o
f
th
ese
cl
u
s
ter
s
an
d
b
y
u
s
i
n
g
t
h
ese
clu
s
ter
n
u
m
b
er
as
an
i
n
p
u
t
f
ea
t
u
r
e
v
ec
to
r
.
C
o
m
p
ar
in
g
[
1
6
,
1
7
]
t
w
o
s
eq
u
e
n
ce
s
o
f
i
n
teg
er
s
v
e
cto
r
s
th
an
t
h
e
e
n
tire
o
r
ig
in
al
v
ec
to
r
s
.
o
n
e
o
f
th
e
ad
d
itio
n
al
ad
v
an
ta
g
e
is
to
co
m
p
u
te
th
e
d
is
ta
n
ce
b
et
w
ee
n
th
e
p
air
s
o
f
clu
s
ter
s
a
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
9
,
No
.
6
,
Dec
em
b
er
2
0
1
9
:
4
6
8
4
-
4
6
9
5
4688
th
e
E
u
cl
id
ea
n
d
is
ta
n
ce
m
ea
s
u
r
e
b
et
w
ee
n
t
h
eir
co
r
r
esp
o
n
d
in
g
ce
n
ter
s
o
f
t
h
e
v
ec
to
r
s
.
T
h
is
is
v
er
y
s
i
m
p
le
to
v
ie
w
t
h
e
lo
o
k
ed
u
p
tab
le
to
m
ea
s
u
r
e
th
e
d
i
s
tan
ce
b
w
t
w
ee
n
th
e
c
lu
s
ter
s
a
n
d
n
o
t
r
eq
u
ir
ed
an
y
ad
d
itio
n
al
co
m
p
u
tatio
n
s
to
m
ea
s
u
r
e
d
is
ta
n
ce
.
I
n
o
r
d
er
to
m
a
k
e
t
h
e
VQ
m
et
h
o
d
s
i
m
p
le,
a
s
et
o
f
clu
s
ter
ce
n
ter
s
is
d
e
f
in
ed
as
a
co
d
eb
o
o
k
b
ec
au
s
e
it
p
r
o
d
u
ce
s
f
ea
t
u
r
e
v
ec
to
r
s
in
to
s
i
n
g
le
v
al
u
e.
T
h
e
co
d
eb
o
o
k
s
ize
r
e
f
er
s
to
t
h
e
n
u
m
b
er
o
f
clu
s
ter
s
i
n
th
e
co
d
eb
o
o
k
.
I
f
a
n
y
s
o
r
t
o
f
in
f
o
r
m
at
io
n
i
s
lo
s
t
w
h
e
n
V
Q
is
[
1
8
-
2
0
]
m
et
h
o
d
is
u
s
ed
to
en
co
d
e
a
n
i
n
p
u
t
v
ec
to
r
s
eq
u
en
ce
.
T
h
en
g
r
o
u
p
in
g
d
is
s
i
m
ilar
p
o
in
ts
an
d
r
ep
r
esen
tin
g
ev
er
y
c
lu
s
ter
m
e
m
b
er
b
y
co
m
p
u
ti
n
g
th
e
m
ea
n
clu
s
ter
s
.
T
h
e
lo
s
s
o
f
in
f
o
r
m
at
io
n
is
t
h
e
d
if
f
er
en
c
e
b
et
w
ee
n
t
h
e
o
r
ig
i
n
al
in
p
u
t
v
ec
to
r
s
eq
u
en
ce
a
n
d
th
e
q
u
an
t
is
ed
v
ec
to
r
s
eq
u
e
n
c
e.
T
h
e
m
ea
n
v
al
u
e
o
f
t
h
is
d
if
f
er
en
ce
is
th
e
VQ
d
i
s
to
r
tio
n
.
By
in
cr
ea
s
esi
n
g
th
e
s
ize
o
f
t
h
e
co
d
eb
o
o
k
lea
d
s
to
d
ec
r
ea
s
es in
VQ
d
is
to
r
tio
n
.
3
.
3
.
T
hresh
o
ld s
et
t
ing
T
h
e
o
u
tp
u
t
o
f
t
h
e
co
d
eb
o
o
k
is
u
s
ed
to
s
et
th
e
t
h
r
es
h
o
ld
v
al
u
e
to
r
ec
o
g
n
ize
we
t
h
er
test
s
p
e
ec
h
s
i
g
n
al
is
p
r
esen
t
o
r
n
o
t
in
th
e
tr
ain
i
n
g
d
ataset.
T
h
e
m
i
n
i
m
u
m
v
alu
e
in
th
e
co
d
e
b
o
o
k
is
u
s
ed
as
th
r
es
h
o
ld
v
alu
e
b
u
t
th
is
th
r
es
h
o
ld
s
h
o
u
ld
b
e
les
s
t
h
e
h
a
lf
o
f
t
h
e
av
er
a
g
e
v
a
lu
e
in
th
e
co
d
eb
o
o
k
,
th
en
o
n
l
y
th
e
t
est
s
p
ee
ch
s
i
g
n
a
l
is
allo
w
ed
to
te
s
t
i
n
th
e
tr
ain
i
n
g
d
ata
s
et
o
th
er
w
i
s
e
te
s
t
s
p
ee
ch
s
ig
n
al
is
n
o
t
p
r
ese
n
t
i
n
t
h
e
tr
ain
i
n
g
d
ata
s
e
t.
On
ce
t
h
e
test
s
p
ee
c
h
s
i
g
n
al
i
s
allo
w
ed
to
test
,
it
l
o
o
k
s
o
n
l
y
t
h
e
m
i
n
i
m
u
m
d
i
s
tan
ce
v
ec
to
r
,
th
e
m
i
n
i
m
u
m
d
is
tan
ce
v
ec
to
r
s
p
ee
ch
is
r
ec
o
g
n
ize
a
s
a
test
s
ig
n
al
s
p
ee
c
h
.
3
.
4
.
P
r
o
po
s
ed
m
o
del a
lg
o
rit
h
m
Co
nte
x
t
ba
s
ed
v
o
ice
det
ec
t
io
n:
Ste
p
1:
I
n
p
u
t d
ata
co
n
tin
u
o
u
s
k
an
n
ad
a
s
p
ee
ch
s
i
g
n
al
1.
T
h
e
s
p
ee
ch
s
ig
n
al
o
f
s
a
m
p
li
n
g
f
r
eq
u
e
n
c
y
s
f
Hz
an
d
h
a
m
m
i
n
g
2.
w
i
n
d
o
w
len
g
t
h
(
N
)
=
s
tep
s
ize
=
0
.
0
5
0
s
f
.
3.
C
o
m
p
u
te
n
u
m
b
er
o
f
f
r
a
m
es
o
f
s
p
ee
c
h
u
s
i
n
g
(
1
)
b
y
s
h
i
f
ti
n
g
t
h
e
w
i
n
d
o
w
t
h
e
ac
r
o
s
s
th
e
en
tire
s
p
ee
ch
s
ig
n
al.
Ste
p
2:
C
o
m
p
u
te
t
h
e
av
er
ag
e
en
er
g
y
o
f
ea
ch
f
r
a
m
e
u
s
in
g
e
q
u
atio
n
(
2
)
.
Ste
p
3:
C
o
m
p
u
te
2
N
p
o
in
t FFT
o
f
w
i
n
d
o
w
ed
s
eq
u
e
n
ce
o
f
ea
c
h
f
r
a
m
e.
1.
C
o
n
s
id
er
o
n
l
y
N
p
o
in
t FFT
co
ef
f
icien
ts
to
r
ed
u
ce
h
i
g
h
er
s
p
ec
tr
al
co
m
p
o
n
e
n
t
s
.
2.
Sp
ec
tr
al
ce
n
tr
o
id
‘
C
’
o
f
ea
ch
w
i
n
d
o
w
ed
s
eq
u
en
ce
u
s
in
g
eq
u
atio
n
(
3
)
T
h
en
f
in
al
l
y
ce
n
tr
o
id
C
is
2
s
C
C
f
.
Ste
p
4:
Fil
ter
in
g
Fi
lter
th
e
a
v
er
ag
e
e
n
er
g
y
s
eq
u
e
n
ce
an
d
ce
n
tr
o
id
s
eq
u
en
ce
u
s
i
n
g
m
ed
i
an
t
w
ice
o
f
f
ilter
len
g
t
h
o
f
f
i
v
e
an
d
co
m
p
u
te
m
e
a
n
E
=
m
ea
n
(
)
fi
lt
e
r
e
d
E
,
m
e
a
n
C
=
m
ea
n
()
fi
lt
e
r
e
d
C
1.
Fin
d
t
h
e
th
r
es
h
o
ld
u
s
in
g
p
d
f
o
f
en
e
g
y
a
n
d
ce
n
tr
o
id
s
eq
u
e
n
ce
.
2.
C
o
m
p
u
te
th
e
th
r
e
s
h
o
ld
as
th
e
w
ei
g
h
ted
av
er
a
g
e
b
et
w
ee
n
t
w
o
f
ir
s
t
p
d
f
lo
ca
l
m
ax
i
m
a
t
h
en
t
h
r
es
h
o
ld
en
er
g
y
=
2
m
e
a
n
E
.
3.
Si
m
i
lar
l
y
s
tep
3
is
r
ep
ea
ted
f
o
r
ce
n
tr
o
id
s
eq
u
en
ce
.
Ste
p
5:
Set th
e
T
h
r
esh
o
ld
v
al
u
es
1.
Set f
la
g
s
1
f
an
d
2
f
.
2.
1
fE
th
r
es
h
o
ld
en
er
g
y
.
3.
2
fC
.
th
r
es
h
o
ld
C
en
tr
o
id
.
4.
1
ff
&
2
f
.
Ste
p 6
:
Sp
ee
ch
d
etec
tio
n
.
1.
I
n
itiali
s
e
co
u
n
t=1
,
f
la
g
=1
.
2.
Set star
t li
m
it.
3.
I
n
cr
ea
s
e
o
v
er
all
co
u
n
ter
.
I
n
cr
ea
s
e
co
u
n
ter
o
f
t
h
e
c
u
r
r
en
t sp
ee
ch
s
e
g
m
e
n
t.
4.
I
f
atlea
s
t o
n
e
s
eg
m
e
n
t h
a
s
b
ee
n
f
o
u
n
d
i
n
th
e
c
u
r
r
en
t lo
o
p
s
et
en
d
co
u
n
ter
t
h
en
i
n
cr
ea
s
e
o
v
e
r
all
co
u
n
ter
.
5.
Me
r
g
e
o
v
er
lap
p
in
g
s
eg
m
e
n
ts
.
6.
P
lo
t
th
e
s
eg
m
e
n
ted
s
p
ee
ch
s
i
g
n
al
b
y
r
ep
r
esen
ti
n
g
i
n
r
ed
co
l
o
u
r
an
d
p
lay
ea
ch
s
eg
m
e
n
t.
F
in
all
y
w
r
itten
ea
ch
s
e
g
m
e
n
t a
s
.
w
a
v
f
ile.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
o
n
tin
u
o
u
s
ka
n
n
a
d
a
s
p
ee
ch
s
eg
men
ta
tio
n
a
n
d
s
p
ee
c
h
r
ec
o
g
n
itio
n
b
a
s
ed
…
(
V
a
n
a
ja
ksh
i P
u
tta
s
w
a
my
Go
w
d
a
)
4689
7.
Seg
m
en
t th
e
co
n
tin
u
o
u
s
k
a
n
n
a
d
a
s
p
ee
ch
s
ig
n
al
b
ased
o
n
co
n
tex
t
u
s
i
n
g
f
la
g
s
1
f
an
d
2
f
.
8.
I
s
o
late
th
e
k
an
n
ad
a
letter
s
p
e
ec
h
s
i
g
n
al
u
s
i
n
g
Z
C
R
f
r
o
m
c
o
n
tin
u
o
u
s
k
a
n
n
ad
a
s
p
ee
ch
b
a
s
ed
o
n
co
n
te
x
t
w
h
ic
h
co
n
tai
n
s
o
n
l
y
k
an
n
ad
a
l
etter
s
p
ee
ch
s
i
g
n
a
l.
P
lo
t
th
e
is
o
lated
k
a
n
n
ad
a
l
etter
s
p
ee
ch
s
i
g
n
al
b
y
r
ep
r
ese
n
tin
g
in
r
ed
co
lo
u
r
a
n
d
p
la
y
is
o
late
d
k
an
n
ad
a
s
p
ee
ch
.
Fi
n
all
y
w
r
itte
n
ea
ch
let
ter
s
p
ee
ch
as .
w
av
f
ile.
Sp
ee
ch
Rec
o
g
nitio
n
Ste
p
1
:
T
ak
e
s
eg
m
e
n
ted
s
p
ee
ch
s
ig
n
al
as a
tr
ain
in
g
d
ataset
Ste
p
2:
A
p
p
l
y
t
h
e
Fo
u
r
ier
tr
an
s
f
o
r
m
s
eg
m
e
n
ted
s
p
ee
c
h
s
i
g
n
al
Ste
p
3:
Ma
p
th
e
lo
g
a
m
p
l
itu
d
e
s
o
f
th
e
s
p
ec
tr
u
m
o
b
tain
ed
ab
o
v
e
o
n
to
th
e
Me
l
s
ca
le,
u
s
i
n
g
tr
ian
g
u
lar
o
v
er
lap
p
in
g
w
i
n
d
o
w
s
.
Ste
p
4:
T
ak
e
th
e
Dis
cr
ete
C
o
s
i
n
e
T
r
an
s
f
o
r
m
o
f
th
e
li
s
t o
f
Me
l lo
g
-
a
m
p
lit
u
d
es,
as i
f
it
w
er
e
a
s
ig
n
a
l.
Ste
p
5:
T
h
e
MFC
C
s
ar
e
t
h
e
a
m
p
lit
u
d
e
s
o
f
th
e
r
es
u
lti
n
g
s
p
ec
tr
u
m
.
Ste
p
6:
C
alcu
late
MF
C
C
C
o
ef
f
icie
n
t f
o
r
tr
ain
in
g
d
ata
s
et
w
it
h
f
r
eq
u
e
n
c
y
r
ate
1
0
Ste
p
7:
Gen
er
ate
co
d
e
b
o
o
k
f
o
r
ea
ch
s
eg
m
e
n
ted
MFC
C
co
ef
f
icie
n
ts
u
s
i
n
g
v
ec
to
r
Qu
a
n
tizat
io
n
(
VQ)
w
it
h
8
n
u
m
b
er
u
s
i
n
g
eq
u
id
is
ta
n
ce
a
n
d
k
ee
p
th
e
s
e
co
d
eb
o
o
k
s
as a
tr
ain
in
g
d
ata
s
et.
Ste
p
8
:
R
ep
ea
t step
6
an
d
s
tep
7
f
o
r
test
s
p
ee
ch
s
i
g
n
al
Ste
p
9
:
Set th
e
t
h
r
es
h
o
ld
b
y
c
o
m
p
u
ti
n
g
th
e
t
h
e
m
in
i
m
u
m
v
a
lu
e
o
f
co
d
eb
o
o
k
s
in
tr
ain
i
n
g
d
ata
s
e
t.
Ste
p
10:
C
o
m
p
u
te
th
e
a
v
er
ag
e
v
alu
e
o
f
co
d
eb
o
o
k
s
in
tr
ai
n
in
g
d
ata
s
et.
Ste
p
11:
I
f
th
r
es
h
o
ld
v
al
u
e
is
les
s
th
a
n
h
alf
o
f
th
e
a
v
er
ag
e
v
a
lu
e
,
t
h
en
it
ch
ec
k
th
e
te
s
t
s
i
g
n
al
in
t
h
e
tr
ain
i
n
g
d
ata
s
et
o
th
er
w
is
e
tes
t sp
ee
ch
s
ig
n
s
l is n
o
t r
ec
o
g
n
ized
.
Ste
p
12:
C
o
m
p
u
te
d
is
tan
ce
b
et
w
ee
n
tes
t d
ata
w
it
h
tr
ain
in
g
d
ata.
Ste
p
13:
T
h
e
m
i
n
i
m
u
m
d
i
s
tan
ce
v
ec
to
r
s
p
ee
ch
in
t
h
e
tr
ai
n
in
g
d
ata
s
et
is
co
n
s
id
er
ed
as r
ec
o
g
n
ized
Ste
p
14:
Sto
p
4.
RE
SU
L
T
S
A
ND
D
I
SCU
SS
I
O
NS
4
.
1
.
Seg
m
ent
a
t
io
n
T
h
e
Fig
u
r
e
1
s
h
o
w
s
t
h
at
co
n
ti
n
u
o
u
s
o
r
ig
i
n
al
k
a
n
n
ad
a
Sp
ee
ch
s
i
g
n
al
s
e
g
m
en
ted
i
n
to
f
o
u
r
p
ar
ts
.
T
h
e
s
h
o
r
t
ti
m
e
en
er
g
y
o
f
s
p
ee
ch
s
i
g
n
al
an
d
co
r
r
esp
o
n
d
i
n
g
s
p
ec
tr
al
ce
n
tr
o
id
o
f
ea
c
h
s
eg
m
e
n
ted
o
u
tp
u
t
m
en
tio
n
ed
w
it
h
g
r
ee
n
co
lo
u
r
an
d
its
f
ilter
ed
o
u
tp
u
t
w
it
h
b
lu
e
co
lo
u
r
.
T
h
e
m
ed
ian
f
ilte
r
o
u
tp
u
t
i
s
co
m
p
letl
y
s
m
o
o
th
e
n
s
o
t
h
at
a
n
y
d
is
to
r
ti
o
n
p
r
esen
t
in
t
h
e
s
p
ee
c
h
s
ig
n
al
co
m
p
letel
y
eli
m
i
n
ated
.
T
h
e
s
eg
m
e
n
ted
s
p
ee
c
h
s
ig
n
al
co
m
p
letel
y
is
o
lated
w
it
h
v
o
iced
an
d
u
n
v
o
iced
w
i
th
r
esp
ec
t
to
th
e
p
ar
ticu
lar
s
ce
n
ar
io
o
f
th
at
co
n
tex
t.
T
h
e
Fig
u
r
e
2
,
s
h
o
w
s
th
e
co
r
r
esp
o
n
d
in
g
k
a
n
n
ad
a
Sp
ee
ch
s
ig
n
a
l
te
x
t.
E
ac
h
s
eg
m
e
n
ted
o
u
tp
u
t
is
co
m
p
le
tl
y
m
ea
n
in
g
f
u
l
w
i
th
r
esp
ec
t
to
th
e
k
a
n
n
ad
a
s
y
n
ta
tic
,
s
e
m
a
n
tic
an
d
g
r
a
m
etic
r
u
le
s
w
h
ic
h
is
m
e
n
tio
n
ed
as
in
Fig
u
r
e
3
o
f
(
a
-
d
)
u
s
i
n
g
u
n
ic
o
d
e
o
f
k
a
n
n
ad
a
lan
g
u
ag
e.
T
h
e
Fi
g
u
r
e
4
i
llu
s
tr
ates
s
o
m
e
o
f
t
h
e
w
o
r
d
s
p
r
ese
n
t
i
n
th
e
s
e
g
m
en
ted
s
p
ee
ch
s
i
g
n
al
t
ex
t
w
r
itte
n
u
s
in
g
U
n
ico
d
e
o
f
k
an
n
ad
a
la
n
g
u
a
g
e
w
it
h
m
at
lab
R
2
0
1
4
a.
T
h
e
T
a
b
le
1
g
iv
es
th
e
co
n
tex
t
b
ased
co
n
tin
u
o
u
s
k
an
n
ad
a
s
p
ee
ch
s
i
g
n
al
s
e
g
m
e
n
tatio
n
w
it
h
d
if
f
er
e
n
t
ac
ce
n
t
o
f
d
if
f
er
en
t
s
i
g
n
al
s
iz
e
o
f
m
ale
a
n
d
f
e
m
ale
s
p
ee
ch
.
T
h
e
s
eg
m
en
ta
tio
n
al
g
o
r
it
h
m
test
ed
n
ea
r
l
y
1
0
0
d
if
f
er
e
n
t
k
a
n
n
ad
a
s
p
ee
ch
s
i
g
n
al
o
f
f
e
m
ale
an
d
m
ale
w
i
th
d
if
f
er
en
t
ac
ce
n
t.
T
h
is
alg
o
r
ith
m
g
i
v
es
0
.
0
1
%
o
f
s
eg
m
e
n
tatio
n
er
r
o
r
r
ate.
On
l
y
th
r
ee
d
i
f
f
er
en
t
s
p
ee
c
h
o
f
d
if
f
er
en
t
d
u
r
atio
n
i
s
l
is
ted
i
n
T
ab
le
1
.
T
h
e
o
r
ig
in
a
l
s
eg
m
e
n
tatio
n
w
it
h
co
n
tex
b
ased
is
al
m
o
s
t
n
ea
r
l
y
eq
u
a
l
t
o
p
r
ac
tical
s
eg
m
e
n
tatio
n
o
f
co
r
r
ec
t
s
eg
m
en
t
s
.
T
h
e
n
u
m
b
er
o
f
m
is
s
ed
an
d
ex
tr
a
s
eg
m
en
tat
io
n
al
m
o
s
t
n
i
l
an
d
er
r
o
r
r
ate
is
alm
o
s
t
n
eli
g
ib
le
.
So
,
th
e
alg
o
r
ith
m
w
h
ic
h
h
as
b
ee
n
m
e
n
tio
n
ed
in
th
i
s
p
ap
er
g
iv
e
s
co
r
r
ec
t
s
e
g
m
en
tatio
n
w
it
h
le
s
s
er
r
o
r
r
ate.
T
h
e
T
ab
le
2
s
h
o
w
s
th
at
s
e
g
m
en
ta
tio
n
o
f
co
n
tex
t
b
ased
w
it
h
d
if
f
er
en
t
v
o
ca
b
u
la
r
y
f
o
r
m
ale
a
n
d
f
e
m
ale
o
f
d
if
f
er
en
t
ac
ce
n
t
g
i
v
e
s
d
ep
en
d
in
g
o
n
th
e
v
o
ca
b
u
lar
y
s
ize
th
e
s
e
g
m
en
tatio
n
ac
cu
r
ac
y
d
ec
r
ea
s
es
as
v
o
ca
b
u
lar
y
s
ize
in
cr
ea
s
es
b
u
t
p
r
o
p
o
s
ed
alg
o
r
ith
m
g
i
v
es b
ett
er
s
eg
m
e
n
tat
io
n
ac
cu
r
ac
y
ev
e
n
f
o
r
lar
g
e
v
o
ca
b
u
lar
y
s
ize.
T
h
e
Fig
u
r
e
5
s
h
o
w
s
t
h
e
co
n
ti
n
u
o
u
s
k
an
n
ad
a
letter
s
s
p
ee
c
h
s
ig
n
al
an
d
s
e
g
m
e
n
tatio
n
o
f
c
o
n
tin
u
o
u
s
k
an
n
ad
a
letter
s
s
p
ee
c
h
s
ig
n
al
s
h
o
w
s
t
h
e
co
n
t
in
u
o
u
s
k
an
n
ad
a
lette
r
s
s
p
ee
ch
s
i
g
n
al
w
h
ich
co
n
tai
n
s
a
ll 5
2
letter
s
o
f
k
a
n
n
ad
a
la
n
g
u
a
g
e.
T
h
is
s
p
e
ec
h
s
i
g
n
al
letter
s
ar
e
i
s
o
lated
in
Fi
g
u
r
e
5
(
a
-
h
)
w
h
ich
ar
e
m
e
n
tio
n
ed
o
n
l
y
s
ev
e
n
letter
s
w
it
h
r
ed
m
ar
k
i
n
s
p
ee
c
h
s
i
g
n
a
l.
T
h
e
T
ab
le
3
c
o
n
tain
s
t
h
e
s
e
g
m
e
n
tat
io
n
o
f
t
h
e
k
an
n
ad
a
ak
s
h
r
a
f
r
o
m
co
n
ti
n
u
o
u
s
k
an
n
a
d
a
ak
s
h
r
a
s
p
ee
ch
s
ig
n
al
o
f
m
a
le
a
n
d
f
e
m
ale.
Seg
m
e
n
tatio
n
o
f
k
a
n
n
a
d
a
ak
s
h
r
a
i
n
t
h
i
s
ca
s
e
is
th
e
is
o
latio
n
o
f
k
a
n
n
ad
a
ak
s
h
r
a.
T
h
e
co
n
ti
n
u
o
u
s
s
p
ee
ch
s
i
g
n
al
co
n
tain
s
to
tal
5
2
k
a
n
n
ad
a
a
k
s
h
ar
'
s
.
t
h
e
o
r
ig
i
n
al
s
eg
m
e
n
t
s
m
u
s
t
b
e
5
2
ak
s
h
ar
'
s
b
u
t
is
o
lated
k
a
n
n
a
d
a
letter
s
f
r
o
m
t
h
e
alg
o
r
it
h
m
w
h
ic
h
h
as
b
ee
n
m
e
n
tio
n
ed
g
iv
e
s
m
o
r
e
t
h
an
5
2
ak
s
h
ar
'
s
w
h
ic
h
co
n
tai
n
s
co
r
r
ec
t
5
2
ak
s
h
ar
'
s
s
e
g
m
e
n
ts
an
d
ex
tr
a
s
eg
m
e
n
ts
w
h
ich
ar
e
n
o
t
r
eq
u
ir
ed
s
eg
m
e
n
ts
.
T
h
e
er
r
o
r
o
cc
u
r
s
d
u
e
to
th
e
e
x
tr
a
s
e
g
m
e
n
t
s
w
h
ich
i
s
r
ed
u
n
d
an
t,
b
u
t
t
h
e
m
i
s
s
i
n
g
s
e
g
m
en
t
s
n
il
f
o
r
b
o
th
m
al
e
an
d
f
e
m
ale
s
p
ee
ch
s
i
g
n
al.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
9
,
No
.
6
,
Dec
em
b
er
2
0
1
9
:
4
6
8
4
-
4
6
9
5
4690
Seg
m
en
t
-
1
Seg
m
en
t
-
2
Seg
m
en
t
-
3
Seg
m
en
t
-
4
Fig
u
r
e
1
.
C
o
n
ti
n
u
o
u
s
Ka
n
n
ad
a
s
p
ee
ch
S
i
g
n
al
s
eg
m
e
n
ted
in
t
o
f
o
u
r
p
ar
ts
Fig
u
r
e
2
.
Or
ig
in
al
Ka
n
n
ad
a
S
p
ee
ch
s
ig
n
al
te
x
t
Fig
u
r
e
3
.
Kan
n
ad
a
Sp
ee
ch
s
ig
n
al
is
s
eg
m
e
n
ted
in
to
f
o
u
r
p
ar
ts
(
a)
,
(
b
)
,
(
c
)
an
d
(
d
)
Fig
u
r
e
4
.
Un
ico
d
e
r
ep
r
esen
tati
o
n
o
f
Kan
n
ad
a
tex
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
o
n
tin
u
o
u
s
ka
n
n
a
d
a
s
p
ee
ch
s
eg
men
ta
tio
n
a
n
d
s
p
ee
c
h
r
ec
o
g
n
itio
n
b
a
s
ed
…
(
V
a
n
a
ja
ksh
i P
u
tta
s
w
a
my
Go
w
d
a
)
4691
T
ab
le
1
.
C
o
n
tex
t
b
ased
co
n
ti
n
u
o
u
s
Ka
n
n
ad
a
Sp
ee
ch
s
e
g
m
en
tatio
n
w
it
h
d
if
f
er
en
t a
cc
en
t
S
p
e
a
k
e
r
s
N
o
.
o
f
d
i
f
f
e
r
e
n
t
sp
e
a
k
e
r
s
N
o
.
o
f
o
r
i
g
i
n
a
l
se
g
me
n
t
s
N
o
.
o
f
c
o
r
r
e
c
t
se
g
me
n
t
s
N
o
.
o
f
mi
sse
d
se
g
me
n
t
s
N
o
.
o
f
e
x
t
r
a
se
g
me
n
t
s
Er
r
o
r
Rate
F
e
mal
e
w
i
t
h
si
g
n
a
l
_
1
S1
4
4
N
i
l
N
i
l
N
i
l
S2
4
4
N
i
l
N
i
l
N
i
l
S3
4
4
N
i
l
N
i
l
N
i
l
M
a
l
e
w
i
t
h
si
g
n
a
l
_
1
S1
4
4
N
i
l
N
i
l
N
i
l
S2
4
4
N
i
l
N
i
l
N
i
l
S3
4
4
N
i
l
N
i
l
N
i
l
F
e
mal
e
w
i
t
h
si
g
n
a
l
_
2
S1
10
10
N
i
l
N
i
l
N
i
l
S2
11
12
N
i
l
1
1
S3
10
11
N
i
l
1
1
M
a
l
e
w
i
t
h
si
g
n
a
l
_
2
S1
11
12
N
i
l
1
1
S2
10
11
N
i
l
1
1
S3
11
10
1
N
i
l
1
F
e
mal
e
w
i
t
h
si
g
n
a
l
_
3
S1
62
62
N
i
l
N
i
l
N
i
l
S2
63
62
N
i
l
1
1
S3
63
62
N
i
l
1
1
M
a
l
e
w
i
t
h
si
g
n
a
l
_
3
S1
62
62
N
i
l
N
i
l
N
i
l
S2
63
62
N
i
l
1
1
S3
63
62
N
i
l
1
1
T
ab
le
2
.
Seg
m
en
ta
tio
n
o
f
co
n
te
x
t b
ased
Kan
n
ad
a
s
p
ee
ch
s
ig
n
al
w
it
h
d
if
f
er
en
t v
o
ca
b
u
lar
y
(
a)
(
b
)
(
c)
(
d
)
Fig
u
r
e
5
.
(
a)
C
o
n
tin
u
o
u
s
k
a
n
n
ad
a
letter
s
s
p
ee
ch
s
i
g
n
al
,
(
b
)
Ak
s
h
ar
a
’
a
’
,
(
c
)
Ak
s
h
ar
a
’
aa
’
,
(
d
)
A
k
s
h
ar
a
’
e
’
S
p
e
a
k
e
r
s
N
u
mb
e
r
S
p
e
a
k
e
r
s
S
i
z
e
o
f
v
o
c
a
b
u
l
a
r
y
(
s
e
g
me
n
t
s)
S
e
g
me
n
t
a
c
c
u
r
a
c
y
(
%)
Er
r
o
r
r
a
t
e
(
%)
F
e
mal
e
50
1
0
0
99
1
M
a
l
e
50
1
0
0
99
1
F
e
mal
e
50
5
0
0
97
3
M
a
l
e
50
5
0
0
98
2
F
e
mal
e
50
1
0
0
0
96
4
M
a
l
e
50
1
0
0
0
97
3
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
9
,
No
.
6
,
Dec
em
b
er
2
0
1
9
:
4
6
8
4
-
4
6
9
5
4692
(
e)
(
f
)
(
g
)
(h
)
Fig
u
r
e
5
.
(
e
)
A
k
s
h
ar
a
’
ee
’
,
(
f
)
Ak
s
h
ar
a
’
u
’
,
(
g
)
,
Ak
s
h
ar
a
‘
u
u
’
,
(
h
)
Ak
s
h
ar
a
’
r
u
’
T
ab
le
3
.
Kan
n
ad
a
Ak
s
h
ar
a
(
letter
s
)
Sp
ee
ch
Seg
m
e
n
tatio
n
w
it
h
d
if
f
er
en
t a
cc
e
n
t
S
p
e
a
k
e
r
s
N
o
.
o
f
d
i
f
f
e
r
e
n
t
S
p
e
a
k
e
r
s
N
o
.
o
f
O
r
i
g
i
n
a
l
se
g
me
n
t
s
N
o
.
o
f
C
o
r
r
e
c
t
se
g
me
n
t
s
N
o
.
o
f
mi
sse
d
se
g
me
n
t
s
N
o
.
o
f
e
x
t
r
a
se
g
me
n
t
s
S
e
g
me
n
t
a
t
i
o
n
Er
r
o
r
F
e
mal
e
1
52
57
N
i
l
5
5
2
52
54
N
i
l
2
2
3
52
56
N
i
l
4
4
M
a
l
e
1
52
55
N
i
l
3
3
2
52
57
N
i
l
5
5
3
52
56
N
i
l
4
4
4
.
2
.
Rec
o
g
nitio
n
T
h
e
T
a
b
le
4
illu
s
tr
ates
th
e
s
eg
m
e
n
ted
Kan
n
ad
a
letter
s
s
p
ee
ch
r
ec
o
g
n
itio
n
f
o
r
m
ale
an
d
f
e
m
ale.
T
h
e
is
o
lated
Kan
n
ad
a
a
k
s
h
a
r
a
s
p
ee
ch
s
ig
n
al
is
u
s
ed
as
a
tr
ain
g
d
ata
s
et
f
o
r
t
h
e
r
ec
o
g
n
itio
n
s
y
s
te
m
.
T
h
e
r
ec
o
g
n
itio
n
s
y
s
te
m
g
iv
e
s
'
1
'
f
o
r
th
e
t
h
e
r
ec
o
g
n
is
ed
s
eg
m
en
t
w
h
ich
is
p
r
ese
n
t
i
n
t
h
e
tr
ain
e
d
d
ata
s
et
i.e
th
e
test
s
eg
m
e
n
t
o
th
er
w
i
s
e
it i
s
'
0
'
.
T
ab
le
4
.
I
s
o
lated
Kan
n
ad
a
(
A
k
s
h
ar
a)
letter
s
s
p
ee
ch
r
ec
o
g
n
it
io
n
S
p
e
a
k
e
r
s
T
e
st
S
e
g
m
e
n
t
s
T
r
a
i
n
i
n
g
S
e
g
me
n
t
ಅ
ಆ
ಇ
ಈ
ಉ
ಊ
ಅ
ಋ
F
e
mal
e
ಅ
1
0
0
0
0
0
1
0
ಆ
0
1
0
0
0
0
0
0
ಋ
0
0
0
0
0
0
0
1
ಎ
0
0
0
0
0
0
0
0
M
a
l
e
ಇ
0
0
1
0
0
0
0
0
ಈ
0
0
0
1
0
0
0
0
ಋ
0
0
0
0
0
0
0
1
ಔ
0
0
0
0
0
0
0
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
C
o
n
tin
u
o
u
s
ka
n
n
a
d
a
s
p
ee
ch
s
eg
men
ta
tio
n
a
n
d
s
p
ee
c
h
r
ec
o
g
n
itio
n
b
a
s
ed
…
(
V
a
n
a
ja
ksh
i P
u
tta
s
w
a
my
Go
w
d
a
)
4693
T
ab
le
5
c
o
n
tain
s
th
e
co
n
te
x
t
b
ased
co
n
tin
u
o
u
s
Ka
n
n
ad
a
s
p
ee
ch
r
ec
o
g
n
it
io
n
.
T
h
e
s
eg
m
e
n
ted
s
p
ee
ch
s
ig
n
al
is
u
s
ed
as
a
tr
ai
n
g
d
ata
s
et
f
o
r
t
h
e
r
ec
o
g
n
itio
n
s
y
s
te
m
.
T
h
e
r
ec
o
g
n
iti
o
n
s
y
s
te
m
g
i
v
es
'
1
'
f
o
r
th
e
r
ec
o
g
n
ized
s
e
g
m
e
n
t
w
h
ic
h
i
s
p
r
esen
t
i
n
t
h
e
tr
ai
n
ed
d
ata
s
et
i.e
th
e
test
s
e
g
m
en
t.
Oth
er
w
i
s
e
it
is
'
0
'
.
T
h
is
r
ec
o
g
n
itio
n
s
y
s
te
m
g
iv
e
s
b
etter
r
ec
o
g
n
itio
n
r
ate.
T
h
e
T
a
b
le
6
s
h
o
w
s
th
e
co
m
p
a
r
is
o
n
o
f
s
p
ee
ch
r
ec
o
g
n
itio
n
o
f
d
if
f
er
en
t
f
ea
t
u
r
e
e
x
tr
ac
tio
n
m
eth
o
d
s
o
f
d
if
f
er
e
n
t
lan
g
u
ag
e
s
w
h
ic
h
ar
e
r
ef
er
r
ed
in
th
e
s
e
p
ap
er
s
[
21
,
22
].
T
h
is
tab
le
m
ain
l
y
co
m
p
ar
e
s
t
h
e
M
FC
C
alo
n
g
w
it
h
o
t
h
er
tec
h
n
iq
ue
o
f
f
ea
t
u
r
e
ex
tr
ac
tio
n
m
et
h
o
d
s
u
s
ed
f
o
r
test
an
d
tr
ai
n
g
d
ata
s
et
w
h
ich
ar
e
ar
e
u
s
ed
as
a
in
p
u
t
to
th
e
r
ec
o
g
n
itio
n
s
y
s
t
e
m
.
Sp
ee
ch
s
i
g
n
al
o
f
d
i
f
f
er
e
n
t
lan
g
u
a
g
es
d
ep
en
d
s
o
n
t
h
e
s
y
n
tax
a
n
d
s
y
m
a
n
tic
an
al
y
s
is
o
f
th
at
la
n
g
u
a
g
e.
E
s
p
ec
iall
y
th
e
I
n
d
ian
la
n
g
u
a
g
es
h
av
e
lar
g
e
s
et
s
o
f
v
o
ca
b
u
lar
y
.
T
h
e
r
ec
o
g
n
itio
n
o
f
I
n
d
ian
la
n
g
u
a
g
e
s
p
ee
ch
s
ig
n
a
l
as
s
h
o
w
n
i
n
T
ab
le
5
w
it
h
d
if
f
er
e
n
t
r
ec
o
g
n
itio
n
r
ate.
T
h
e
p
r
o
p
o
s
ed
m
et
h
o
d
w
h
ic
h
h
as
b
ee
n
u
s
ed
g
i
v
es
b
etter
r
ec
o
g
n
it
io
n
r
ate
w
it
h
le
s
s
n
u
m
b
er
o
f
MF
C
C
co
ef
f
ici
en
ts
th
a
n
th
e
o
t
h
er
l
an
g
u
a
g
e.
T
h
e
Fi
g
u
r
e
6
s
h
o
w
s
t
h
e
g
r
ap
h
ical
r
ep
r
ese
n
tatio
n
o
f
r
ec
o
g
n
itio
n
ac
c
u
r
ac
y
o
f
d
i
f
f
er
e
n
t
lan
g
u
a
g
es
[
23
-
25
]
.
T
h
e
m
et
h
o
d
w
h
ic
h
i
s
u
s
ed
i
n
th
is
p
ap
er
f
o
r
Kan
n
ad
a
lan
g
u
a
g
e
s
p
ee
ch
r
ec
o
g
n
itio
n
s
h
o
ws
b
etter
ac
cu
r
ac
y
r
ate
th
a
n
th
e
e
x
is
t
in
g
m
eth
o
d
s
.
T
ab
le
5
.
C
o
n
tex
t
b
ased
co
n
ti
n
u
o
u
s
Ka
n
n
ad
a
s
p
ee
ch
r
ec
o
g
n
it
io
n
T
ab
le
6
.
R
ec
o
g
n
itio
n
ac
c
u
r
ac
y
o
f
d
if
f
er
en
t f
ea
t
u
r
e
ex
tr
ac
tio
n
m
et
h
o
d
s
o
f
d
i
f
f
er
e
n
t la
n
g
u
a
g
es
L
a
n
g
u
a
g
e
s
F
e
a
t
u
r
e
Ex
t
r
a
c
t
i
o
n
M
e
t
h
o
d
s
P
e
r
c
e
n
t
a
g
e
o
f
R
e
c
o
g
n
i
t
i
o
n
r
a
t
e
I
n
d
o
n
e
si
a
n
M
F
C
C
,
C
M
N
_
M
F
C
C
,
P
L
P
8
8
,
9
0
.
8
8
K
o
r
e
a
n
M
F
C
C
(
G
M
M
)
78
En
g
l
i
sh
M
F
C
C
95
A
r
a
b
i
c
M
F
C
C
91
I
n
d
i
a
n
(
T
a
mi
l
)
L
P
C
+
M
F
C
C
89
I
n
d
i
a
n
(
T
e
l
u
g
u
)
L
P
C
+
M
F
C
C
90
I
n
d
i
a
n
(
M
a
l
a
y
a
l
a
m)
L
P
C
+
M
F
C
C
90
I
n
d
i
a
n
(
K
a
n
n
a
d
a
)
L
P
C
+
M
F
C
C
91
I
n
d
i
a
n
(
K
a
n
n
a
d
a
)
P
r
o
p
o
se
d
98
Fig
u
r
e
6
.
R
ec
o
g
n
it
io
n
ac
cu
r
a
c
y
o
f
d
if
f
er
en
t f
ea
t
u
r
e
ex
tr
ac
ti
o
n
m
et
h
o
d
s
o
f
d
if
f
er
en
t la
n
g
u
ag
es
S
p
e
a
k
e
r
s
T
e
st
S
e
g
me
n
t
s
T
r
a
i
n
i
n
g
S
e
g
me
n
t
s
S
e
g
1
S
e
g
2
S
e
g
3
S
e
g
4
S
e
g
1
S
e
g
2
S
e
g
3
S
e
g
4
F
e
mal
e
S
e
g
1
S
e
g
2
S
e
g
3
S
e
g
4
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
M
a
l
e
S
e
g
1
S
e
g
2
S
e
g
3
S
e
g
4
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
1
0
0
0
0
1
0
0
0
0
1
0
0
0
0
1
Evaluation Warning : The document was created with Spire.PDF for Python.