I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
4
,
A
u
g
u
s
t 2
0
2
6
,
p
p
.
3
8
0
5
~
3
8
1
7
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
4
.
p
p
3
8
0
5
-
3
8
1
7
3805
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Unsuperv
ised v
o
i
ce activity
de
tec
ti
o
n bas
ed on the
e
nv
elo
pe's
fractal dime
nsio
n
Nesrine
Aba
j
a
dd
i
1
,
Yo
u
s
s
ef
E
lf
a
hm
1
,
L
a
ila
E
lm
a
zo
uzi
2
,
I
lha
m
M
o
un
ir
2
,
B
a
dia
M
o
u
ni
r
2
,
Abdelm
a
j
id F
a
rc
hi
1
1
I
M
I
I
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
C
o
mp
u
t
e
r
E
n
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s
a
n
d
Te
c
h
n
o
l
o
g
i
e
s
,
H
a
ssa
n
F
i
r
s
t
U
n
i
v
e
r
si
t
y
,
S
e
t
t
a
t
,
M
o
r
o
c
c
o
2
LA
P
S
S
I
I
La
b
o
r
a
t
o
r
y
,
H
i
g
h
S
c
h
o
o
l
o
f
Te
c
h
n
o
l
o
g
y
,
C
a
d
i
A
y
y
a
d
U
n
i
v
e
r
si
t
y
,
S
a
f
i
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
3
0
,
2
0
2
5
R
ev
is
ed
Ma
y
2
4
,
2
0
2
6
Acc
ep
ted
J
u
n
1
9
,
2
0
2
6
Cu
rre
n
tl
y
,
v
o
ice
a
c
ti
v
it
y
d
e
tec
t
io
n
(VA
D)
is
u
ti
li
z
e
d
i
n
m
a
n
y
field
s,
in
c
lu
d
in
g
f
o
re
n
sic
s,
h
e
a
lt
h
c
a
re
,
a
n
d
m
e
d
icin
e
,
to
d
e
tec
t
v
o
c
a
l
a
n
o
m
a
li
e
s,
a
s
we
ll
a
s
in
tele
c
o
m
m
u
n
ica
ti
o
n
s
a
n
d
m
o
b
i
le
tele
p
h
o
n
y
.
Du
e
t
o
it
s
i
m
p
o
rtan
c
e
a
n
d
th
e
d
iffi
c
u
lt
y
o
f
d
isti
n
g
u
ish
i
n
g
b
e
twe
e
n
sp
e
e
c
h
a
n
d
n
o
n
s
p
e
e
c
h
se
g
m
e
n
ts,
e
sp
e
c
ially
in
n
o
isy
e
n
v
iro
n
m
e
n
ts
(lo
w
si
g
n
a
l
-
to
-
n
o
ise
ra
ti
o
(S
NR))
,
t
h
is
a
re
a
re
m
a
in
s
u
n
d
e
r
c
o
n
ti
n
u
o
u
s
d
e
v
e
lo
p
m
e
n
t.
M
o
st
e
x
isti
n
g
VA
D
a
lg
o
rit
h
m
s
re
q
u
ire
p
re
d
e
fin
e
d
th
re
sh
o
l
d
s
o
r
trai
n
in
g
d
a
ta,
wh
ic
h
re
d
u
c
e
s
th
e
ir
co
m
p
a
ti
b
il
i
ty
.
T
h
is
stu
d
y
p
ro
p
o
se
s
a
n
u
n
su
p
e
rv
ise
d
VA
D
sy
ste
m
th
a
t
u
ti
li
z
e
s
th
e
Ka
tz
a
lg
o
rit
h
m
to
c
a
lcu
late
th
e
fra
c
tal
d
ime
n
sio
n
o
f
e
n
v
e
lo
p
e
s
o
b
tai
n
e
d
th
r
o
u
g
h
a
si
n
g
le
fre
q
u
e
n
c
y
fil
teri
n
g
(S
F
F
)
a
p
p
ro
a
c
h
.
Th
is
m
e
th
o
d
a
ll
o
ws
fo
r
h
i
g
h
tem
p
o
ra
l
a
n
d
f
re
q
u
e
n
c
y
re
so
l
u
ti
o
n
.
Th
e
p
r
o
p
o
se
d
VA
D
a
lg
o
rit
h
m
d
o
e
s
n
o
t
re
q
u
ire
a
n
y
tr
a
in
in
g
d
a
ta
a
n
d
is
su
i
tab
le
fo
r
v
a
r
io
u
s
t
y
p
e
s
o
f
n
o
ise
a
n
d
S
NRs
.
To
e
v
a
l
u
a
te
th
e
e
ffe
c
ti
v
e
n
e
ss
o
f
th
e
p
r
o
p
o
se
d
m
e
th
o
d
,
two
d
iffere
n
t
d
a
tab
a
se
s
a
re
u
se
d
:
th
e
Tex
a
s
In
str
u
m
e
n
ts
M
a
s
sa
c
h
u
se
tt
s
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
(TI
M
IT
)
d
a
tab
a
se
a
n
d
th
e
Kin
g
S
a
u
d
Un
iv
e
rsity
(KSU)
Ara
b
ic
sp
e
e
c
h
d
a
tab
a
se
.
Th
e
e
x
p
e
rime
n
tal
re
su
lt
s
re
v
e
a
l
a
n
a
v
e
ra
g
e
d
e
tec
ti
o
n
a
c
c
u
ra
c
y
o
f
9
6
.
8
6
%
,
d
e
m
o
n
stra
ti
n
g
it
s
c
o
n
si
d
e
ra
b
le
v
a
lu
e
in
v
a
rio
u
s a
p
p
li
c
a
ti
o
n
s.
K
ey
w
o
r
d
s
:
Fra
ctal
d
im
en
s
io
n
Sig
n
al
p
r
o
ce
s
s
in
g
Sin
g
le
f
r
eq
u
e
n
cy
f
ilter
in
g
Sp
ee
ch
/n
o
n
s
p
ee
ch
Un
s
u
p
er
v
is
ed
lear
n
in
g
Vo
ice
ac
tiv
ity
d
etec
tio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Nesr
in
e
Ab
ajad
d
i
I
MI
I
L
ab
o
r
ato
r
y
,
Dep
ar
tm
e
n
t
o
f
E
lectr
ical
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
in
g
Facu
lty
o
f
Scien
ce
s
an
d
T
ec
h
n
o
lo
g
ies,
Hass
an
First Un
iv
er
s
i
ty
R
o
ad
o
f
C
asab
lan
ca
B
.
P: 5
7
7
,
Settat,
Mo
r
o
cc
o
E
m
ail: n
.
ab
ajad
d
i@
u
h
p
.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
Vo
ice
ac
tiv
ity
d
etec
tio
n
(
V
AD
)
is
a
tech
n
iq
u
e
f
o
r
id
e
n
tify
in
g
s
p
ee
c
h
an
d
d
if
f
er
e
n
tiatin
g
it
f
r
o
m
n
o
n
s
p
ee
ch
s
eg
m
e
n
ts
.
I
n
s
ev
er
al
f
ield
s
,
s
u
ch
as
s
p
ee
ch
co
d
in
g
[
1
]
,
d
ig
ital
h
ea
r
i
n
g
aid
s
[
2
]
,
s
p
ee
c
h
r
ec
o
g
n
itio
n
[
3
]
,
[
4
]
,
a
n
d
v
o
ice
-
co
n
tr
o
lled
a
p
p
licatio
n
s
[
2
]
,
VAD
is
a
n
ec
es
s
ar
y
f
ir
s
t step
.
W
h
ile
h
u
m
an
s
ea
s
ily
d
if
f
er
en
tiate
s
p
ee
c
h
f
r
o
m
n
o
n
s
p
ee
ch
,
m
ac
h
in
es
s
u
f
f
e
r
f
r
o
m
th
is
task
.
Fu
r
th
er
m
o
r
e,
p
er
f
o
r
m
an
ce
is
r
e
d
u
ce
d
in
lo
w
s
ig
n
al
-
to
-
n
o
is
e
r
atio
(
SNR
)
s
itu
atio
n
s
,
s
u
ch
as
wh
is
p
er
ed
o
r
s
tr
o
n
g
ly
ac
ce
n
ted
s
p
ee
ch
s
p
o
k
en
at
a
lo
w
v
o
lu
m
e,
e
v
en
wh
e
n
d
ee
p
lear
n
in
g
o
r
a
d
v
an
ce
d
VAD
alg
o
r
ith
m
s
ar
e
em
p
lo
y
e
d
.
Nu
m
er
o
u
s
alg
o
r
ith
m
s
h
av
e
b
e
en
d
escr
ib
ed
i
n
th
e
liter
atu
r
e
t
h
at
ex
h
ib
it
ef
f
icien
t
p
er
f
o
r
m
a
n
ce
wh
en
th
e
SNR
i
s
h
ig
h
,
b
u
t
s
tr
u
g
g
le
in
lo
w
-
SNR
en
v
ir
o
n
m
en
ts
.
W
ea
k
p
h
o
n
em
es
s
u
ch
as
f
r
ica
tiv
es
an
d
n
asals
,
as
well
as
lo
w
-
am
p
litu
d
e
s
eg
m
en
ts
at
th
e
b
eg
in
n
in
g
an
d
en
d
o
f
u
tter
an
ce
s
,
ar
e
ch
allen
g
i
n
g
to
d
etec
t
in
th
ese
co
n
d
itio
n
s
,
lead
in
g
to
p
o
o
r
p
er
f
o
r
m
a
n
ce
[
5
]
.
T
r
ad
itio
n
al
VAD
alg
o
r
ith
m
s
wer
e
b
ased
o
n
ac
o
u
s
tic
ch
ar
ac
ter
is
tics
s
u
ch
as e
n
er
g
y
,
ze
r
o
-
cr
o
s
s
in
g
r
ate
[
6
]
,
a
n
d
a
u
to
co
r
r
elatio
n
f
u
n
ctio
n
[
7
]
.
T
h
es
e
f
ea
tu
r
es p
er
f
o
r
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
8
0
5
-
3
8
1
7
3806
w
e
l
l
i
n
s
p
e
e
c
h
d
e
t
e
ct
i
o
n
w
h
en
S
N
R
i
s
h
i
g
h
,
b
u
t
s
u
f
f
e
r
wh
e
n
S
N
R
i
s
l
o
w
.
L
a
t
e
r
r
es
e
ar
c
h
e
x
p
l
o
r
e
d
m
o
r
e
a
d
v
a
n
c
e
d
f
e
a
t
u
r
e
s
s
u
c
h
as
h
i
g
h
e
r
-
o
r
d
e
r
s
t
a
t
is
t
ic
s
(
HO
S
)
o
f
t
h
e
e
r
r
o
r
s
i
g
n
a
l
i
n
l
i
n
e
a
r
p
r
e
d
i
c
t
i
v
e
c
o
d
i
n
g
(
L
P
C
)
[
8
]
a
n
d
m
e
l
f
r
e
q
u
e
n
c
y
c
e
p
s
t
r
a
l
c
o
ef
f
i
c
i
e
n
ts
(
MF
C
C
)
[
9
]
.
I
n
n
o
is
y
c
o
n
d
i
t
i
o
n
s
,
t
h
es
e
c
h
a
r
a
c
t
e
r
i
s
t
i
cs
p
e
r
f
o
r
m
e
d
b
e
t
t
e
r
,
b
u
t
t
h
ey
w
er
e
s
t
ill
i
n
s
u
f
f
ic
ie
n
tl
y
r
e
lia
b
l
e
f
o
r
p
r
ac
t
ica
l u
s
es
.
A
n
o
t
h
e
r
c
lass
o
f
VAD
al
g
o
r
it
h
m
s
[
1
0
]
was in
s
p
i
r
ed
b
y
E
p
h
r
ai
m
a
n
d
Ma
la
h
[
1
1
]
,
wh
i
ch
ass
u
m
es
t
h
at
cl
ea
n
s
p
e
ec
h
a
n
d
n
o
is
e
h
a
v
e
d
i
f
f
e
r
e
n
t
s
tatis
t
ic
al
p
r
o
p
e
r
ti
es.
T
h
ey
a
p
p
l
ie
d
a
Ga
u
s
s
ia
n
d
is
t
r
i
b
u
ti
o
n
m
o
d
el
t
o
V
AD
[
1
2
]
a
n
d
im
p
r
o
v
ed
it
w
it
h
L
a
p
l
ac
e
d
i
s
tr
i
b
u
ti
o
n
s
[
1
3
]
a
n
d
m
i
x
e
d
d
is
t
r
i
b
u
ti
o
n
s
[
1
4
]
.
On
e
o
f
t
h
e
m
a
in
c
h
al
le
n
g
es
w
it
h
th
ese
V
AD
a
lg
o
r
it
h
m
s
is
ca
lcu
lat
in
g
th
e
SNR
,
wh
i
ch
r
eq
u
ir
es
est
im
ati
n
g
t
h
e
n
o
is
e
s
p
ec
t
r
u
m
b
y
ass
u
m
i
n
g
th
at
t
h
e
f
ir
s
t
d
o
ze
n
f
r
a
m
es
c
o
n
tai
n
o
n
l
y
n
o
is
e
.
T
h
er
ef
o
r
e
,
s
o
m
e
r
ese
a
r
c
h
e
r
s
h
av
e
p
r
o
p
o
s
e
d
c
o
m
b
in
in
g
t
h
ese
p
a
r
a
m
e
te
r
s
t
o
i
n
cr
ea
s
e
t
h
e
r
o
b
u
s
t
n
ess
o
f
VA
Ds
i
n
v
a
r
i
o
u
s
n
o
is
y
en
v
i
r
o
n
m
en
ts
o
r
i
n
te
g
r
ati
n
g
th
em
i
n
t
o
d
ee
p
l
ea
r
n
in
g
,
w
h
i
c
h
h
as
b
ee
n
a
d
o
p
ted
i
n
r
ec
en
t
y
e
ar
s
.
Neu
r
a
l
n
etw
o
r
k
m
o
d
els
h
av
e
b
ee
n
ap
p
l
ie
d
i
n
s
p
ee
c
h
a
cti
v
it
y
d
et
ec
ti
o
n
[
1
5
]
,
[
1
6
]
t
o
d
ete
r
m
i
n
e
w
h
e
th
er
a
f
r
a
m
e
co
n
t
ai
n
s
s
p
e
ec
h
o
r
n
o
n
s
p
e
ec
h
.
Al
th
o
u
g
h
s
u
p
er
v
is
e
d
VA
D
s
y
s
te
m
s
h
av
e
a
l
o
w
e
r
r
o
r
r
at
e
an
d
h
i
g
h
ac
c
u
r
a
cy
,
th
e
y
r
e
q
u
i
r
e
a
s
i
g
n
if
ic
an
t
a
m
o
u
n
t
o
f
tr
ai
n
i
n
g
d
ata
to
r
et
r
ai
n
t
h
e
n
e
tw
o
r
k
t
o
d
is
ti
n
g
u
is
h
s
p
e
e
ch
f
r
o
m
n
o
n
s
p
e
ec
h
,
wh
i
ch
l
im
i
ts
th
ei
r
r
ea
l
-
ti
m
e
u
s
e
.
Un
s
u
p
e
r
v
is
e
d
VA
D
is
c
o
m
m
o
n
ly
u
s
e
d
i
n
d
i
g
i
tal
s
p
e
ec
h
p
r
o
ce
s
s
in
g
.
I
ts
p
r
im
ar
y
b
e
n
e
f
it
o
v
e
r
s
u
p
er
v
is
ed
VA
D
te
ch
n
i
q
u
es
is
t
h
at
it
ca
n
d
i
f
f
e
r
e
n
t
iat
e
b
etw
ee
n
s
p
e
ec
h
a
n
d
n
o
n
s
p
ee
c
h
s
eg
m
e
n
ts
wit
h
o
u
t
r
e
q
u
ir
i
n
g
tr
ain
in
g
d
ata
o
r
p
r
i
o
r
k
n
o
wl
e
d
g
e
.
I
n
s
te
ad
,
u
n
s
u
p
er
v
is
ed
VA
D
m
et
h
o
d
s
u
t
ili
ze
ac
o
u
s
tic
f
ea
tu
r
es
t
o
d
et
er
m
i
n
e
th
r
es
h
o
l
d
s
o
r
o
t
h
e
r
p
ar
am
et
er
s
in
t
h
e
ti
m
e
a
n
d
/
o
r
f
r
eq
u
en
c
y
d
o
m
ai
n
s
.
T
o
ex
tr
ac
t
s
p
ee
ch
f
ea
t
u
r
es,
th
e
m
ajo
r
ity
o
f
r
esear
ch
is
b
ased
o
n
s
ig
n
al
p
r
o
ce
s
s
in
g
u
s
in
g
t
h
e
d
is
cr
ete
Fo
u
r
ier
tr
an
s
f
o
r
m
(
DFT)
[
1
7
]
o
r
g
a
m
m
ato
n
e
f
ilter
s
[
1
8
]
,
[
1
9
]
.
R
ec
en
tly
,
a
n
ew
ap
p
r
o
ac
h
ca
lled
s
in
g
le
f
r
eq
u
e
n
cy
f
ilter
in
g
(
SF
F)
[
2
0
]
h
as
o
f
f
er
ed
an
alter
n
ativ
e
t
o
p
r
e
v
io
u
s
m
eth
o
d
s
.
T
o
d
is
tin
g
u
is
h
s
p
ee
ch
f
r
o
m
n
o
n
s
p
ee
ch
r
e
g
io
n
s
,
SF
F
ex
tr
ac
ts
th
e
s
ig
n
al'
s
tem
p
o
r
al
en
v
elo
p
e
at
ea
ch
f
r
eq
u
e
n
cy
.
I
t
ac
h
iev
es
th
is
b
y
d
eter
m
in
in
g
t
h
e
p
ar
am
eter
c
o
n
to
u
r
as a
f
u
n
ctio
n
o
f
tim
e
a
f
ter
co
m
p
u
tin
g
th
e
v
ar
ia
n
ce
an
d
m
ea
n
o
f
th
e
s
ig
n
al'
s
weig
h
ted
en
er
g
y
at
ea
ch
f
r
eq
u
en
cy
.
T
h
e
m
eth
o
d
was
also
u
s
ed
in
[
2
1
]
,
wh
ic
h
p
r
o
p
o
s
ed
r
ep
lacin
g
th
e
SF
F
ap
p
r
o
ac
h
with
a
q
u
asi
-
q
u
ad
r
a
tu
r
e
f
ilter
b
an
k
to
o
b
tain
en
v
elo
p
es
an
d
u
tili
ze
d
h
is
to
g
r
am
s
an
d
esti
m
ato
r
s
o
f
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
s
to
d
eter
m
in
e
th
e
d
etec
tio
n
th
r
esh
o
ld
.
No
tab
ly
,
th
e
SF
F
ap
p
r
o
ac
h
h
as
also
b
ee
n
ap
p
lied
to
s
p
ee
ch
in
tellig
ib
ilit
y
en
h
a
n
ce
m
en
t
[
2
2
]
,
e
p
o
ch
e
x
tr
ac
tio
n
f
r
o
m
em
o
tio
n
al
s
p
e
ec
h
[
2
3
]
,
a
n
d
o
th
er
s
im
ilar
ap
p
licatio
n
s
.
T
h
is
s
u
g
g
ests
th
at
it is
a
p
r
o
m
is
in
g
ap
p
r
o
ac
h
.
T
h
is
p
ap
er
aim
s
to
d
ev
elo
p
a
r
eliab
le,
u
n
s
u
p
er
v
is
ed
VAD
s
y
s
tem
b
ased
o
n
th
e
f
r
ac
tal
d
im
en
s
io
n
o
f
en
v
elo
p
es.
T
h
e
e
n
v
elo
p
es
ar
e
ca
lcu
lated
at
ea
c
h
f
r
eq
u
e
n
cy
u
s
in
g
th
e
SF
F
ap
p
r
o
ac
h
.
T
h
e
f
r
ac
tal
d
im
en
s
io
n
is
a
m
ath
em
atica
l
to
o
l
wid
ely
u
s
ed
in
v
ar
io
u
s
s
cien
tific
f
ield
s
.
Pre
v
io
u
s
s
tu
d
ies
h
av
e
esti
m
ated
th
e
f
r
ac
tal
d
im
en
s
io
n
o
f
elec
tr
o
ca
r
d
io
g
r
am
(
E
C
G)
s
ig
n
als
in
[
2
4
]
,
[
2
5
]
to
aid
in
class
if
icatio
n
.
Nu
m
er
o
u
s
tech
n
iq
u
es
h
av
e
b
ee
n
d
e
v
elo
p
e
d
to
m
ea
s
u
r
e
th
e
f
r
ac
tal
d
im
e
n
s
io
n
,
in
cl
u
d
in
g
th
o
s
e
p
r
o
p
o
s
ed
b
y
Katz
[
2
6
]
,
Hig
u
c
h
i
[
2
7
]
,
Petr
o
s
ian
[
2
8
]
,
Ma
r
ag
o
s
[
2
9
]
,
an
d
t
h
e
am
p
litu
d
e
s
ca
le
ap
p
r
o
ac
h
[
3
0
]
.
I
n
p
ar
ticu
lar
,
two
alg
o
r
ith
m
s
,
Hig
u
ch
i a
n
d
Katz
,
h
av
e
b
ee
n
s
tu
d
ied
f
o
r
th
eir
e
f
f
ec
tiv
en
ess
in
VADs
[
3
1
]
.
An
ac
cu
r
ate
VAD
s
y
s
tem
is
cr
u
cial
in
v
ar
i
o
u
s
s
cien
tific
f
ield
s
,
s
u
ch
as
au
d
io
f
o
r
en
s
ics,
m
ed
ical
d
iag
n
o
s
is
,
an
d
s
p
ee
ch
r
ec
o
g
n
it
io
n
ap
p
licatio
n
s
.
T
h
e
VAD
al
g
o
r
ith
m
s
m
u
s
t
b
e
s
u
itab
le
f
o
r
v
ar
io
u
s
n
o
is
e
ty
p
es
an
d
SNR
lev
els,
a
n
d
p
r
ef
e
r
ab
l
y
s
h
o
u
ld
n
o
t n
ee
d
an
y
tr
ain
in
g
d
ata
o
r
p
r
io
r
k
n
o
wled
g
e
o
f
n
o
is
e.
T
h
e
n
o
v
elty
o
f
th
is
s
tu
d
y
is
th
e
u
s
e
o
f
f
r
ac
tal
d
im
en
s
io
n
an
aly
s
is
o
f
th
e
en
v
elo
p
es
ex
tr
ac
ted
b
y
SF
F
[
2
0
]
to
d
ev
elo
p
a
n
ew
ap
p
r
o
ac
h
f
o
r
VAD.
T
h
e
f
r
ac
tal
d
im
en
s
io
n
'
s
r
o
b
u
s
tn
ess
to
n
o
is
e
an
d
o
th
er
d
is
to
r
tio
n
s
is
an
ad
v
an
tag
e
o
f
ap
p
ly
in
g
it
to
VAD.
I
t
ca
p
tu
r
e
s
th
e
o
v
er
all
c
o
m
p
lex
ity
o
f
th
e
s
ig
n
al,
r
ath
er
th
an
r
ely
in
g
o
n
s
p
ec
if
ic
s
p
ec
tr
al
o
r
tem
p
o
r
al
f
ea
tu
r
es
th
at
m
ay
b
e
af
f
ec
ted
b
y
n
o
is
e
o
r
o
th
er
d
i
s
to
r
tio
n
s
.
Ad
d
itio
n
ally
,
u
s
in
g
th
e
SF
F
ap
p
r
o
ac
h
,
th
e
en
v
el
o
p
e
is
ca
lcu
lated
with
h
ig
h
tem
p
o
r
al
a
n
d
f
r
eq
u
en
c
y
r
eso
lu
tio
n
,
wh
ich
im
p
r
o
v
es
th
e
ac
cu
r
ac
y
o
f
th
e
f
r
ac
tal
d
im
en
s
io
n
an
d
,
h
e
n
ce
,
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
E
x
is
tin
g
VAD
alg
o
r
ith
m
s
u
s
e
p
r
ed
ef
in
e
d
o
r
ad
ju
s
ted
th
r
esh
o
ld
s
d
ep
en
d
i
n
g
o
n
th
e
t
y
p
e
o
f
n
o
is
e
an
d
SNR
lev
els
[
2
0
]
,
[
3
1
]
–
[
3
3
]
.
I
n
co
n
tr
ast,
th
e
m
eth
o
d
a
u
to
m
atica
lly
d
eter
m
in
es
th
e
t
h
r
esh
o
ld
o
f
ea
ch
en
v
elo
p
e,
r
eg
ar
d
less
o
f
th
e
SNR
lev
el
o
r
n
o
is
e
ty
p
e.
Mo
r
e
o
v
er
,
th
e
alg
o
r
ith
m
r
eq
u
ir
es
f
ewe
r
en
v
elo
p
es
co
m
p
ar
ed
to
th
e
VAD
m
eth
o
d
b
ased
o
n
th
e
SFF
p
r
o
p
o
s
ed
in
[
2
0
]
,
t
h
er
eb
y
r
ed
u
cin
g
co
m
p
lex
ity
an
d
m
ak
in
g
it
s
u
itab
le
f
o
r
v
a
r
io
u
s
ap
p
licatio
n
s
.
T
h
is
p
ap
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
ws:
s
ec
tio
n
2
d
etails
th
e
p
r
o
p
o
s
ed
VAD
m
eth
o
d
,
wh
ic
h
in
clu
d
es
SF
F,
co
m
p
u
tatio
n
o
f
f
r
ac
tal
d
im
en
s
io
n
,
an
d
au
to
m
atic
th
r
esh
o
ld
esti
m
atio
n
.
T
h
e
r
e
s
u
lts
o
b
tain
ed
ar
e
p
r
esen
ted
in
s
ec
tio
n
3
,
w
h
er
e
two
s
p
ee
ch
d
atab
ases
ar
e
ev
alu
ated
.
Ad
d
itio
n
ally
,
a
d
is
cu
s
s
io
n
is
p
r
o
v
id
ed
to
an
aly
ze
an
d
co
m
p
a
r
e
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
m
eth
o
d
.
Fin
all
y
,
s
ec
tio
n
4
c
o
n
clu
d
es with
th
e
f
in
d
in
g
s
.
2.
M
E
T
H
O
D
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
s
u
g
g
e
s
ted
ap
p
r
o
ac
h
is
ass
ess
ed
u
s
in
g
th
e
T
ex
as
I
n
s
tr
u
m
e
n
ts
Ma
s
s
ac
h
u
s
etts
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
(
T
I
M
I
T)
[
3
4
]
a
n
d
Kin
g
Sau
d
Un
i
v
er
s
ity
(
KSU
)
[
3
5
]
,
[
3
6
]
d
ata
b
ases
,
wh
ich
wer
e
r
ec
o
r
d
e
d
at
d
if
f
er
e
n
t
s
am
p
lin
g
f
r
eq
u
en
cies
(
1
6
k
Hz
an
d
4
8
k
Hz,
r
esp
ec
tiv
ely
)
.
T
o
en
s
u
r
e
a
co
n
s
is
ten
t
s
am
p
lin
g
f
r
e
q
u
en
c
y
,
th
e
KSU
s
o
u
n
d
s
ar
e
d
o
wn
s
am
p
led
to
1
6
k
Hz.
T
h
e
SF
F
tech
n
iq
u
e
i
s
u
s
ed
to
o
b
tain
th
e
s
ig
n
al
en
v
elo
p
es
at
th
e
d
esi
r
ed
f
r
eq
u
e
n
cy
.
T
h
e
s
tep
s
f
o
r
ca
lcu
latin
g
th
e
s
ig
n
al
en
v
e
lo
p
es
v
ia
SF
F
ar
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Un
s
u
p
ervis
ed
vo
ice
a
ctivity
d
etec
tio
n
b
a
s
ed
o
n
th
e
e
n
ve
lo
p
e
's fra
cta
l d
imen
s
io
n
(
N
esr
in
e
A
b
a
ja
d
d
i
)
3807
d
escr
ib
ed
in
th
e
f
o
llo
win
g
s
u
b
s
ec
tio
n
.
T
o
p
r
ev
en
t
i
n
f
o
r
m
at
io
n
lo
s
s
,
a
ze
r
o
-
p
ad
d
i
n
g
was
im
p
lem
en
ted
at
th
e
en
d
o
f
t
h
e
en
v
el
o
p
e.
T
o
in
cr
e
ase
th
e
ac
cu
r
ac
y
o
f
s
eg
m
en
t
d
etec
tio
n
,
a
s
h
o
r
te
r
win
d
o
w
i
s
p
r
ef
er
r
ed
(
1
0
m
s
)
.
T
h
e
f
r
ac
tal
d
im
en
s
io
n
is
co
m
p
u
ted
f
o
r
ea
ch
en
v
elo
p
e
f
r
am
e,
allo
win
g
f
o
r
a
u
to
m
ati
c
s
p
ee
ch
s
eg
m
en
t
class
if
icatio
n
.
T
h
e
au
to
m
atic
th
r
esh
o
ld
ca
lcu
latio
n
s
tep
s
ar
e
ex
p
lain
ed
as
f
o
llo
ws
.
Fig
u
r
e
1
illu
s
tr
ates
th
e
g
en
er
al
b
lo
c
k
d
iag
r
am
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
.
Fig
u
r
e
1
.
A
g
en
er
al
b
l
o
ck
d
iag
r
am
o
f
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
2
.
1
.
E
nv
elo
pes
ca
lcula
t
io
n
T
o
o
b
tain
t
h
e
en
v
el
o
p
e
f
o
r
a
s
p
ec
if
ic
f
r
eq
u
en
cy
,
a
tech
n
iq
u
e
k
n
o
wn
as
SF
F
is
em
p
lo
y
ed
[
2
0
]
.
T
h
e
s
p
ee
ch
s
ig
n
al
is
m
u
ltip
lied
b
y
a
co
m
p
lex
s
in
u
s
o
id
,
a
n
d
th
e
f
ilter
ed
f
r
eq
u
e
n
cy
is
s
h
if
ted
u
s
in
g
a
s
in
g
le
-
p
o
le
f
ilter
p
lace
d
n
ea
r
th
e
u
n
it
cir
c
le
(
r
>
0
.
9
9
)
at
2
⁄
,
wh
er
e
r
ep
r
es
en
ts
th
e
s
am
p
lin
g
f
r
e
q
u
en
c
y
.
T
h
is
en
s
u
r
es
th
at
th
e
f
ilter
r
em
ain
s
s
tab
le
an
d
co
n
s
is
ten
t
ac
r
o
s
s
all
f
r
eq
u
e
n
cies,
r
esu
ltin
g
in
a
m
o
r
e
ac
c
u
r
ate
en
v
elo
p
e
an
d
im
p
r
o
v
e
d
r
eso
lu
tio
n
at
ea
ch
f
r
eq
u
en
c
y
.
T
o
co
m
p
u
te
th
e
am
p
litu
d
e
en
v
elo
p
e
o
f
th
e
s
ig
n
al
at
a
d
esire
d
f
r
eq
u
e
n
cy
,
th
e
f
o
llo
win
g
s
tep
s
ar
e
p
er
f
o
r
m
e
d
.
T
h
e
s
h
if
te
d
s
ig
n
al
[
]
at
f
r
eq
u
en
c
y
k
is
as (
1
)
.
̃
[
,
]
=
[
]
−
2
̅
(
1
)
T
h
e
d
esire
d
f
r
eq
u
en
cy
is
d
en
o
ted
b
y
̅
,
wh
er
e
k
r
an
g
es
f
r
o
m
0
to
K
-
1
,
an
d
n
r
an
g
es
f
r
o
m
1
to
.
T
h
e
to
tal
n
u
m
b
er
o
f
s
am
p
les
an
d
f
r
eq
u
e
n
cies
ar
e
r
ep
r
esen
ted
b
y
N
an
d
K,
r
esp
ec
tiv
ely
.
Af
ter
s
h
if
tin
g
th
e
s
ig
n
al,
it is
f
ilter
ed
by
a
s
in
g
le
-
p
o
le
f
ilter
with
a
tr
a
n
s
f
er
f
u
n
ctio
n
s
p
ec
if
ied
as
(
2
)
.
[
]
=
1
1
+
−
1
(
2
)
T
h
e
o
u
tp
u
t o
f
th
e
f
ilter
in
g
p
r
o
ce
s
s
in
(
3
)
.
T
o
en
s
u
r
e
s
tab
ilit
y
ac
r
o
s
s
all
f
r
eq
u
en
cies,
th
e
p
ar
am
eter
r
is
ch
o
s
en
to
b
e
g
r
ea
ter
th
an
0
.
9
9
.
T
h
e
r
esu
ltin
g
o
u
tp
u
t,
[
,
]
,
is
a
c
o
m
p
lex
n
u
m
b
er
an
d
ca
n
b
e
r
e
p
r
esen
ted
in
p
o
lar
f
o
r
m
u
s
in
g
(
4
)
.
T
h
e
SF
F
en
v
elo
p
e,
d
e
n
o
ted
b
y
[
,
]
,
is
d
ef
in
ed
in
(
5
)
.
I
n
co
n
tr
ast,
th
e
p
h
a
s
e,
d
en
o
ted
by
[
,
]
,
is
d
ef
in
ed
u
s
in
g
(
6
)
,
wh
er
e
"r
"
an
d
"i"
d
en
o
te
t
h
e
r
ea
l
a
n
d
im
ag
in
a
r
y
co
m
p
o
n
en
ts
o
f
t
h
e
co
m
p
le
x
n
u
m
b
er
,
r
esp
ec
tiv
ely
.
[
,
]
=
−
[
,
−
1
]
+
̃
[
,
]
(
3
)
[
,
]
=
[
,
]
[
,
]
(
4
)
[
,
]
=
√
2
[
,
]
+
2
[
,
]
(
5
)
[
,
]
=
−
1
(
[
,
]
[
,
]
)
(
6
)
I
n
th
is
r
esear
ch
,
th
e
SF
F
m
eth
o
d
ca
n
o
b
tain
th
e
en
v
elo
p
e
at
an
y
d
esire
d
f
r
eq
u
en
cy
.
Ho
wev
er
,
f
o
r
p
r
ac
tical
ap
p
licatio
n
s
,
th
e
en
v
elo
p
e
was
ca
lcu
lated
ev
er
y
1
0
0
Hz
(
∆f
=1
0
0
)
with
in
th
e
r
an
g
e
o
f
1
0
0
Hz
to
/
2
⁄
,
wh
ich
co
v
er
s
th
e
u
s
ef
u
l
s
p
e
ctr
al
b
an
d
o
f
s
p
ee
c
h
.
T
h
is
r
es
u
lted
in
a
to
tal
o
f
8
0
en
v
elo
p
es
th
at
v
ar
ied
o
v
er
tim
e,
co
m
p
ar
e
d
to
a
p
r
ev
io
u
s
s
tu
d
y
[
2
0
]
th
at
u
s
ed
1
8
5
en
v
elo
p
es.
Usi
n
g
f
ewe
r
e
n
v
el
o
p
es
ca
n
r
ed
u
ce
th
e
m
eth
o
d
'
s
im
p
lem
en
tatio
n
tim
e
an
d
m
em
o
r
y
r
eq
u
ir
e
m
en
ts
.
2
.
2
.
F
r
a
ct
a
l dim
ens
io
ns
a
nd
t
hresh
o
ld ca
lcula
t
io
n
I
n
th
is
s
tu
d
y
,
th
e
f
r
ac
tal
d
im
e
n
s
io
n
s
[
,
]
wer
e
co
m
p
u
ted
f
o
r
th
e
f
r
am
e
m
in
th
e
en
v
elo
p
e
k
.
T
h
e
f
r
ac
tal
d
im
en
s
io
n
s
wer
e
o
b
tain
ed
b
y
co
m
p
u
tin
g
t
h
e
r
atio
o
f
th
e
n
o
r
m
alize
d
len
g
th
[
,
]
o
f
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
8
0
5
-
3
8
1
7
3808
en
v
elo
p
e
to
th
e
m
ax
im
u
m
d
is
tan
ce
b
etwe
en
th
e
f
ir
s
t
p
o
i
n
t
an
d
an
y
o
th
e
r
p
o
in
t
o
n
th
e
en
v
elo
p
e.
T
h
e
len
g
th
[
,
]
(
7
)
was
d
eter
m
in
ed
b
y
s
u
m
m
in
g
u
p
th
e
d
is
tan
ce
s
b
e
twee
n
two
s
u
cc
ess
iv
e
p
o
in
ts
u
s
in
g
th
e
co
o
r
d
in
ates X
an
d
Y
o
f
ea
ch
p
o
in
t f
o
r
th
e
f
r
a
m
e
m
o
f
th
e
ℎ
en
v
elo
p
e.
[
,
]
=
∑
√
(
[
,
+
1
]
−
X
[
,
]
)
2
+
(
[
,
+
1
]
−
Y
[
,
]
)
2
−
1
=
1
(
7
)
Her
e,
=
1
.
2
…
.
−
1
,
an
d
r
ep
r
esen
ts
th
e
s
am
p
le
n
u
m
b
er
f
o
r
th
e
f
r
am
e
m
.
As
well,
t
h
e
p
la
n
ar
e
x
te
n
t
[
,
]
is
d
ef
in
ed
in
(
8
)
.
[
,
]
=
(
√
(
[
,
+
1
]
−
X
[
,
1
]
)
2
+
(
[
,
+
1
]
−
Y
[
,
1
]
)
2
)
(
8
)
T
o
o
b
tain
th
e
[
,
]
,
th
e
len
g
th
[
,
]
,
an
d
p
la
n
ar
ex
te
n
t
[
,
]
wer
e
n
o
r
m
alize
d
b
y
th
e
av
er
ag
e
d
is
tan
ce
[
,
]
,
wh
ich
is
th
e
d
is
tan
ce
b
etwe
en
co
n
s
ec
u
tiv
e
p
o
in
ts
o
n
th
e
en
v
el
o
p
e.
T
h
e
e
q
u
atio
n
f
o
r
th
e
a
v
er
ag
e
d
is
tan
ce
is
g
iv
en
b
y
(
9
)
.
[
,
]
=
(
√
(
[
,
+
1
]
−
X
[
,
]
)
2
+
(
[
,
+
1
]
−
X
[
,
]
)
2
)
(
9
)
T
h
i
s
a
p
p
r
o
a
c
h
,
d
e
f
i
n
e
d
b
y
t
h
e
K
a
t
z
a
l
g
o
r
it
h
m
,
d
e
t
e
r
m
i
n
es
t
h
e
f
r
a
c
t
a
l
d
i
m
e
n
s
i
o
n
[
,
]
c
a
l
c
u
l
a
t
e
d
u
s
i
n
g
(
1
0
)
.
[
,
]
=
l
o
g
10
(
[
,
]
[
,
]
)
l
o
g
10
(
[
,
]
[
,
]
)
(
1
0
)
T
h
e
f
r
ac
tal
d
im
en
s
io
n
s
ar
e
u
s
ed
to
d
if
f
er
en
tiate
b
etwe
en
s
p
ee
ch
an
d
s
ilen
ce
f
r
am
es.
Fig
u
r
e
2
illu
s
tr
ates
th
e
f
ir
s
t
en
v
elo
p
e
o
f
a
s
p
ee
ch
s
ig
n
al
f
r
o
m
th
e
KSU
d
atab
ase
a
n
d
its
co
r
r
esp
o
n
d
in
g
f
r
ac
tal
d
im
en
s
io
n
with
th
e
VAD
d
ec
is
io
n
.
Sp
ec
if
ically
,
Fig
u
r
e
2
(
a)
p
r
esen
ts
th
e
f
ir
s
t
s
p
ee
ch
en
v
elo
p
e,
wh
ile
Fig
u
r
e
2
(
b
)
s
h
o
ws
th
e
f
r
ac
tal
d
im
en
s
io
n
o
f
th
e
f
ir
s
t e
n
v
elo
p
e
(
b
lu
e
lin
e
)
to
g
eth
e
r
w
ith
th
e
co
r
r
esp
o
n
d
in
g
VAD
d
e
cisi
o
n
(
r
ed
lin
e)
.
(
a)
Fig
u
r
e
2
.
E
x
am
p
le
o
f
a
VAD
d
ec
is
io
n
f
o
r
t
h
e
f
ir
s
t e
n
v
el
o
p
e
b
ased
o
n
its
f
r
ac
tal
d
im
e
n
s
io
n
:
(
a)
f
ir
s
t e
n
v
el
o
p
e
o
f
s
p
ee
ch
o
f
th
e
KSU
d
atab
as
e
an
d
(
b
)
t
h
e
f
r
ac
tal
d
im
en
s
io
n
o
f
th
e
f
ir
s
t e
n
v
elo
p
e
an
d
its
V
AD
d
ec
is
io
n
2
.
3
.
L
o
g
ic
decisi
o
n
2
.
3
.
1
.
T
hresh
o
ld det
er
m
ina
t
i
o
n
T
o
d
eter
m
in
e
th
e
en
v
elo
p
e'
s
t
h
r
esh
o
ld
d
ec
is
io
n
(
Fig
u
r
e
2
)
au
to
m
atica
lly
,
was
em
p
lo
y
ed
th
e
m
eth
o
d
s
u
g
g
ested
in
[
3
7
]
,
[
3
8
]
was
em
p
lo
y
ed
t
o
esti
m
ate
th
e
f
ir
s
t
s
ig
n
if
ican
t
ch
an
g
e
(
I
n
d
)
o
f
th
e
s
o
r
ted
f
r
ac
tal
d
im
en
s
io
n
[
,
]
.
As ex
p
r
ess
ed
in
(
1
1
)
,
th
e
th
r
esh
o
l
d
is
ca
lcu
lated
b
y
av
er
ag
in
g
th
e
f
r
ac
tal
d
im
en
s
io
n
s
f
o
r
ea
ch
en
v
elo
p
e
.
T
h
e
av
er
a
g
in
g
is
p
er
f
o
r
m
e
d
u
p
t
o
th
e
f
i
r
s
t sig
n
if
ican
t c
h
an
g
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Un
s
u
p
ervis
ed
vo
ice
a
ctivity
d
etec
tio
n
b
a
s
ed
o
n
th
e
e
n
ve
lo
p
e
's fra
cta
l d
imen
s
io
n
(
N
esr
in
e
A
b
a
ja
d
d
i
)
3809
ℎ
(
)
=
(
(
[
,
1
:
Ind
]
)
)
(
1
1
)
2
.
3
.
2
.
Vo
ice
a
ct
i
v
it
y
det
ec
t
i
o
n
decisi
o
n
T
h
e
s
p
ee
ch
-
p
r
esen
ce
s
eg
m
en
ts
f
o
r
th
e
en
v
elo
p
e
k
ar
e
th
o
s
e
ab
o
v
e
th
e
th
r
esh
o
ld
ℎ
(
)
,
an
d
th
e
s
p
ee
ch
-
ab
s
en
ce
s
eg
m
en
ts
ar
e
t
h
o
s
e
b
elo
w
th
e
th
r
esh
o
ld
ℎ
(
)
.
T
h
e
en
v
elo
p
e
d
ec
is
io
n
[
,
]
is
as
(
1
2
)
.
[
,
]
=
1
,
[
,
]
>
ℎ
(
)
[
,
]
=
0
,
[
,
]
≤
ℎ
(
)
(
12
)
W
h
er
e
m
p
r
o
v
id
es
th
e
n
u
m
b
e
r
o
f
f
r
am
es
f
o
r
th
e
ℎ
en
v
el
o
p
e,
s
p
ee
ch
-
p
r
esen
ce
s
eg
m
en
ts
ar
e
in
d
icate
d
b
y
1
,
wh
ile
s
p
ee
ch
-
ab
s
en
ce
o
r
n
o
is
e
s
eg
m
en
ts
ar
e
in
d
icate
d
b
y
0
.
On
th
e
b
asis
o
f
th
e
en
v
el
o
p
e'
s
d
ec
is
io
n
s
,
th
e
Han
g
o
v
er
s
ch
em
e
is
ap
p
lied
to
r
ef
in
e
its
o
u
tp
u
t
an
d
en
h
a
n
c
e
its
p
er
f
o
r
m
an
ce
[
1
0
]
,
[
1
2
]
,
[
2
0
]
.
T
h
e
h
a
n
g
o
v
er
th
r
esh
o
ld
is
b
etwe
en
2
0
an
d
4
0
ms
[
2
0
]
.
Fo
r
th
is
s
tu
d
y
,
th
e
s
m
o
o
th
in
g
th
r
esh
o
l
d
em
p
lo
y
ed
is
2
0
m
s
.
T
h
e
f
in
al
VAD
d
ec
is
io
n
,
(
)
,
is
b
ased
o
n
th
e
e
n
v
elo
p
e'
s
d
ec
is
io
n
,
[
,
]
.
Fo
r
clea
n
s
p
ee
ch
,
th
e
alg
o
r
ith
m
r
eq
u
ir
es
a
m
in
im
u
m
o
f
4
0
%
o
f
th
e
en
v
elo
p
es
to
ex
h
ib
it
s
p
ee
ch
ac
tiv
ity
in
o
r
d
er
to
class
if
y
th
e
s
eg
m
en
t
as
clea
n
s
p
ee
ch
[
2
0
]
.
T
h
is
is
d
u
e
to
th
e
f
ac
t
th
at
clea
n
s
p
ee
ch
s
ig
n
als
ty
p
ically
h
av
e
a
h
ig
h
d
y
n
am
ic
r
an
g
e
in
b
o
t
h
tim
e
an
d
f
r
eq
u
e
n
cy
d
o
m
ain
s
.
C
o
n
v
e
r
s
ely
,
f
o
r
n
o
is
y
s
p
ee
ch
,
th
e
alg
o
r
ith
m
r
e
q
u
ir
es
a
m
in
im
u
m
o
f
6
0
%
o
f
th
e
e
n
v
elo
p
es
t
o
ex
h
ib
it
s
p
ee
ch
ac
tiv
ity
in
o
r
d
er
to
class
if
y
th
e
s
eg
m
en
t
as
n
o
is
y
s
p
ee
c
h
[
2
0
]
.
C
o
n
s
eq
u
en
tly
,
if
th
e
p
er
ce
n
tag
e
is
ab
o
v
e
th
e
th
r
esh
o
ld
p
er
ce
n
tag
e
η
,
th
en
th
e
f
in
al
d
ec
is
io
n
(
)
is
1
(
i.e
.
,
s
p
ee
ch
is
p
r
esen
t)
.
Oth
er
wis
e,
th
e
f
in
al
d
ec
is
io
n
is
0
(
i.
e.
,
n
o
s
p
ee
c
h
is
p
r
esen
t)
,
as d
e
f
in
ed
in
(
1
3
)
.
{
(
)
=
1
,
∑
[
,
]
=
1
≥
(
)
=
0
,
∑
[
,
]
=
1
<
(
1
3
)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
is
p
ar
t,
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
d
ev
el
o
p
ed
VAD
m
eth
o
d
is
ev
alu
ate
d
u
s
in
g
two
d
ata
b
ases
,
KSU
an
d
T
I
MI
T
,
an
d
co
m
p
a
r
ed
wit
h
o
th
er
wid
el
y
r
ec
o
g
n
ized
VA
D
alg
o
r
ith
m
s
.
T
h
e
p
er
f
o
r
m
an
c
e
is
test
ed
in
b
o
th
Ar
ab
ic
an
d
E
n
g
lis
h
to
ass
ess
its
ef
f
ec
tiv
en
ess
f
o
r
d
if
f
er
en
t
lan
g
u
ag
es.
Ad
d
itio
n
ally
,
t
h
e
r
o
b
u
s
tn
ess
o
f
t
h
e
m
eth
o
d
is
ev
alu
ated
b
y
in
tr
o
d
u
cin
g
d
if
f
er
e
n
t
ty
p
es
o
f
n
o
is
e
at
v
ar
y
i
n
g
le
v
els
o
f
SNR
.
Fig
u
r
e
3
illu
s
tr
ates
th
e
m
etr
ics em
p
lo
y
ed
t
o
ev
alu
ate
t
h
e
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
VAD
alg
o
r
ith
m
[
3
9
]
.
Fig
u
r
e
3
.
I
ll
u
s
tr
atio
n
o
f
t
h
e
p
e
r
f
o
r
m
a
n
ce
m
etr
ics
3
.
1
.
P
er
f
o
r
m
a
nce
m
e
a
s
ures
I
n
ev
al
u
atin
g
t
h
e
p
e
r
f
o
r
m
an
c
e
o
f
VAD
m
eth
o
d
s
,
f
iv
e
p
ar
a
m
eter
s
p
r
o
p
o
s
ed
b
y
Fre
em
a
n
et
a
l
.
[
3
9
]
ar
e
ty
p
ically
u
s
ed
,
as sh
o
wn
in
Fig
u
r
e
3
a
n
d
d
escr
ib
e
d
as f
o
ll
o
ws
:
i)
C
OR
R
E
C
T
: th
e
VAD
alg
o
r
ith
m
'
s
co
r
r
ec
t d
ec
is
io
n
in
d
etec
tin
g
s
p
ee
ch
.
ii)
Fro
n
t
-
en
d
clip
p
in
g
(
FEC):
m
is
class
if
icatio
n
o
f
s
p
ee
c
h
p
r
esen
ce
as
n
o
n
s
p
ee
ch
at
t
h
e
s
tar
t
o
f
a
s
p
ee
ch
-
p
r
esen
ce
s
eg
m
e
n
t.
iii)
Mid
s
p
ee
ch
clip
p
in
g
(
MS
C
)
:
m
is
class
if
icatio
n
o
f
s
p
ee
ch
p
r
esen
ce
as
n
o
n
s
p
e
ec
h
in
s
id
e
a
s
p
ee
ch
-
p
r
esen
ce
s
eg
m
e
n
t.
i
v
)
C
a
r
r
y
o
v
e
r
(
O
VE
R
)
:
m
i
s
cl
a
s
s
i
f
i
c
a
t
i
o
n
o
f
n
o
n
s
p
e
e
c
h
a
s
s
p
e
e
ch
-
p
r
e
s
e
n
c
e
a
t
th
e
en
d
o
f
a
s
p
e
e
c
h
-
p
r
e
s
e
n
ce
s
e
g
m
e
n
t.
v)
No
is
e
d
etec
ted
as sp
ee
ch
(
NDS)
: m
is
clas
s
if
icatio
n
o
f
n
o
n
s
p
ee
ch
as sp
ee
ch
with
in
a
n
o
n
s
p
ee
ch
s
eg
m
en
t.
T
o
o
b
tain
th
e
p
er
ce
n
tag
e
v
alu
e
o
f
th
ese
p
ar
am
eter
s
,
ea
ch
is
d
iv
id
ed
b
y
th
e
to
tal
f
r
a
m
e
n
u
m
b
er
an
d
th
en
m
u
ltip
lied
b
y
1
0
0
.
T
h
e
s
u
m
o
f
FEC
an
d
MSC
is
r
ef
er
r
ed
to
as
tr
u
e
r
ejec
tio
n
(
T
R
)
,
r
ep
r
esen
tin
g
th
e
p
er
ce
n
tag
e
o
f
s
p
ee
ch
-
p
r
esen
c
e
r
ejec
ted
.
T
h
e
s
u
m
o
f
OVE
R
an
d
NDS
is
r
ef
e
r
r
ed
to
as
f
al
s
e
ac
ce
p
tan
ce
(
FA)
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
8
0
5
-
3
8
1
7
3810
r
ep
r
esen
tin
g
n
o
n
s
p
ee
c
h
th
at
i
s
d
etec
ted
as
s
p
ee
ch
p
r
esen
ce
.
Fo
r
b
etter
p
er
f
o
r
m
a
n
ce
,
th
e
C
OR
R
E
C
T
m
etr
ic
s
h
o
u
ld
b
e
h
ig
h
er
,
wh
ile
T
R
an
d
FA sh
o
u
ld
b
e
m
o
r
e
m
in
iatu
r
e.
3
.
2
.
Da
t
a
ba
s
es
I
n
th
is
wo
r
k
,
two
d
atab
ases
ar
e
u
s
ed
:
th
e
T
I
M
I
T
d
atab
as
e,
wh
ich
is
wid
ely
u
s
ed
in
v
o
ice
VAD
ap
p
licatio
n
s
an
d
s
p
ee
ch
s
y
s
t
em
s
,
d
ev
elo
p
e
d
b
y
T
e
x
as
I
n
s
tr
u
m
en
ts
an
d
th
e
Ma
s
s
ac
h
u
s
etts
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
[
3
4
]
,
a
n
d
th
e
KS
U
s
p
ee
ch
Ar
a
b
ic
d
atab
ase,
cr
e
ated
b
y
th
e
s
p
ee
ch
g
r
o
u
p
(
SG)
at
KSU
[
3
5
]
,
[
3
6
]
.
Fifty
au
d
io
s
am
p
les
o
f
th
e
f
ir
s
t
s
en
ten
ce
f
r
o
m
th
e
T
I
MI
T
d
atab
ase
o
f
ea
ch
s
p
ea
k
er
ar
e
s
elec
ted
to
ass
e
s
s
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
.
Alth
o
u
g
h
th
e
T
I
MI
T
s
en
ten
ce
s
co
n
tain
m
o
r
e
s
p
e
ec
h
th
an
n
o
n
s
p
ee
ch
s
eg
m
en
ts
,
alg
o
r
ith
m
s
g
en
er
all
y
p
er
f
o
r
m
well
in
ter
m
s
o
f
s
p
ee
ch
ac
ce
p
tan
ce
b
u
t m
a
y
h
av
e
p
o
o
r
er
p
er
f
o
r
m
an
ce
in
n
o
n
s
p
ee
ch
r
ejec
tio
n
.
T
o
s
o
lv
e
th
is
is
s
u
e,
a
2
-
s
ec
o
n
d
s
ilen
ce
p
er
io
d
was
ad
d
ed
at
th
e
b
e
g
in
n
in
g
an
d
en
d
o
f
ea
ch
T
I
MI
T
s
en
ten
ce
,
as
d
escr
ib
ed
in
[
2
0
]
,
[
3
7
]
.
Fo
r
th
e
KSU
d
atab
ase,
5
0
r
ec
o
r
d
i
n
g
s
o
f
is
o
lated
d
ig
its
wer
e
co
llected
f
r
o
m
v
ar
io
u
s
m
icr
o
p
h
o
n
es
u
s
ed
in
th
e
o
f
f
ice,
as
th
ey
p
r
o
v
id
ed
clea
r
p
au
s
es
b
etw
ee
n
s
p
o
k
en
wo
r
d
s
,
allo
win
g
f
o
r
an
ac
cu
r
ate
ass
ess
m
en
t
o
f
th
e
alg
o
r
ith
m
'
s
p
er
f
o
r
m
an
ce
in
d
etec
tin
g
s
p
ee
ch
a
n
d
s
ilen
ce
s
eg
m
en
ts
.
T
h
e
T
I
MI
T
d
ataset
was
r
ec
o
r
d
ed
at
a
1
6
k
Hz
s
am
p
lin
g
r
at
e,
wh
er
ea
s
th
e
KSU
d
atab
ase
was
r
ec
o
r
d
ed
at
a
48
k
Hz
s
am
p
lin
g
r
ate.
B
o
th
d
a
tab
ases
ar
e
d
o
wn
s
am
p
led
to
1
6
k
Hz.
3
.
3
.
P
er
f
o
r
m
a
nce
benchm
a
r
k
ing
3
.
3
.
1
.
E
x
perim
ent
a
l set
up
T
h
is
s
ec
tio
n
aim
s
to
ev
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
b
o
th
e
x
is
tin
g
an
d
p
r
o
p
o
s
ed
VAD
a
lg
o
r
ith
m
s
,
u
s
in
g
th
e
p
er
f
o
r
m
an
ce
m
etr
ics
o
u
tlin
ed
ea
r
lier
.
Fo
r
th
is
p
u
r
p
o
s
e,
f
iv
e
VAD
alg
o
r
ith
m
s
wer
e
s
elec
ted
,
n
am
ely
th
e
VAD
f
o
r
ad
ap
tiv
e
m
u
lti
-
r
a
te
o
p
tio
n
2
(
AM
R
2
)
,
wh
ich
w
as
n
o
r
m
alize
d
f
o
r
E
u
r
o
p
ea
n
T
elec
o
m
m
u
n
icatio
n
s
Stan
d
ar
d
s
I
n
s
titu
te
(
E
T
SI)
[
3
2
]
,
th
e
SF
F
ap
p
r
o
ac
h
f
o
r
d
is
cr
im
in
atin
g
s
p
ee
ch
a
n
d
n
o
n
s
p
ee
c
h
(
VAD
-
SFF)
[
2
0
]
,
th
e
u
n
s
u
p
er
v
is
ed
VAD
an
d
class
if
icatio
n
o
f
au
d
io
s
eg
m
en
t
s
u
s
in
g
f
r
ac
tal
d
im
en
s
io
n
s
(
VAD
-
FD)
[
3
1
]
,
an
d
th
e
VAD
alg
o
r
ith
m
u
s
in
g
m
o
d
if
ied
g
lo
b
al
a
n
d
f
u
zz
y
en
tr
o
p
y
(
Fu
zz
y
E
n
)
[
3
3
]
.
T
o
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
d
ev
elo
p
ed
alg
o
r
ith
m
,
n
o
is
e
-
ad
d
ed
s
p
ee
c
h
s
ig
n
als
with
v
ar
io
u
s
SNR
lev
els,
r
an
g
in
g
f
r
o
m
-
5
d
B
to
2
5
d
B
in
in
cr
em
en
ts
o
f
1
0
d
B
,
wer
e
u
tili
ze
d
.
T
h
eir
p
e
r
f
o
r
m
an
ce
w
as e
v
alu
ated
o
n
t
h
e
T
I
MI
T
an
d
KSU
d
atab
ases
.
3
.
3
.
2
.
T
ex
a
s
I
ns
t
rum
ent
s
M
a
s
s
a
chu
s
et
t
s
I
n
s
t
it
ute
o
f
T
ec
hn
o
lo
g
y
benchm
a
r
k
ing
T
h
e
ev
alu
atio
n
r
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
VAD
m
eth
o
d
p
er
f
o
r
m
s
well
o
n
clea
n
T
I
MI
T
s
en
ten
ce
s
,
with
an
av
er
ag
e
c
o
r
r
ec
t
s
co
r
e
o
f
9
6
%
f
o
r
5
0
au
d
io
f
iles
.
T
h
e
r
atio
s
o
f
FEC,
MSC
,
ND
S,
an
d
OVE
R
ar
e
0
.
0
4
%,
2
.
3
2
%,
1
.
2
7
%,
an
d
0
.
3
5
%,
r
esp
ec
tiv
ely
.
T
h
e
an
aly
s
is
s
h
o
ws
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
h
as
a
lo
w
FEC an
d
OVE
R
r
atio
,
w
h
ile
MSC
an
d
NDS
ar
e
b
elo
w
2
.
5
%.
Ho
wev
er
,
e
v
en
with
m
a
n
u
al
s
eg
m
en
tatio
n
o
f
th
e
T
I
MI
T
co
r
p
u
s
,
an
ac
cu
r
ac
y
b
elo
w
1
0
0
%
was
o
b
tain
ed
,
wh
ich
co
u
ld
ex
p
lain
th
e
v
alu
es
o
f
th
e
r
atio
s
.
Fig
u
r
e
4
illu
s
tr
ates
an
e
x
am
p
l
e
o
f
th
e
VAD
d
ec
is
io
n
m
a
d
e
b
y
th
e
p
r
o
p
o
s
ed
m
eth
o
d
(
g
r
ee
n
lin
e)
a
n
d
m
an
u
al
lab
elin
g
(
r
ed
lin
e
)
f
o
r
a
T
I
MI
T
s
p
ee
ch
,
d
em
o
n
s
tr
atin
g
th
at
t
h
e
m
eth
o
d
ac
c
u
r
ately
d
etec
ts
th
e
s
tar
t
(
FEC)
an
d
en
d
(
OVE
R
)
o
f
t
h
e
s
p
ee
ch
,
as
well
as
th
e
s
h
o
r
t
s
ilen
ce
b
etwe
en
s
p
ee
ch
s
eg
m
en
ts
(
MSC
)
,
wh
ich
ju
s
tifie
s
th
e
MSC
v
alu
e.
Ov
er
all,
th
e
r
esu
lt
s
illu
s
tr
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
s
u
g
g
ested
VAD
alg
o
r
ith
m
.
Fig
u
r
e
4
.
E
x
am
p
le
o
f
th
e
VA
D
p
r
o
p
o
s
ed
m
eth
o
d
an
d
th
e
m
an
u
al
lab
elin
g
s
eg
m
e
n
tatio
n
o
f
T
I
MI
T
d
ata
b
ase
Fig
u
r
e
5
c
o
m
p
ar
e
th
e
p
er
f
o
r
m
an
ce
o
f
s
ev
er
al
VAD
m
eth
o
d
s
,
in
clu
d
in
g
AM
R
2
[
3
2
]
,
VAD
-
SFF
[
2
0
]
,
Fu
zz
y
E
n
[
3
3
]
,
VAD
-
FD
[
3
1
]
,
an
d
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
T
h
e
ev
alu
atio
n
was
ap
p
lied
to
T
I
MI
T
s
p
ee
ch
s
ig
n
als
co
r
r
u
p
te
d
b
y
a
d
d
itiv
e
wh
ite
G
au
s
s
ian
n
o
is
e
(
Fig
u
r
e
5
(
a)
)
a
n
d
f
iv
e
n
o
n
-
s
tatio
n
ar
y
n
o
is
e
ty
p
es,
in
clu
d
in
g
p
in
k
(
Fig
u
r
e
5
(
b
)
)
,
le
o
p
ar
d
Fig
u
r
e
5
(
c)
,
r
estau
r
an
t
(
Fig
u
r
e
5
(
d
)
)
,
tr
ain
(
Fig
u
r
e
5
(
e
)
)
an
d
ca
r
(
Fi
g
u
r
e
5
(
f
)
)
,
at
SNR
lev
els
r
an
g
in
g
f
r
o
m
-
5
d
B
to
2
5
d
B
.
T
h
e
m
ax
im
u
m
co
r
r
e
ct
v
alu
es
o
b
tain
ed
f
o
r
2
5
,
1
5
,
5
,
an
d
-
5
d
B
ar
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Un
s
u
p
ervis
ed
vo
ice
a
ctivity
d
etec
tio
n
b
a
s
ed
o
n
th
e
e
n
ve
lo
p
e
's fra
cta
l d
imen
s
io
n
(
N
esr
in
e
A
b
a
ja
d
d
i
)
3811
9
5
.
7
1
%,
9
4
.
9
9
%,
9
3
.
3
2
%,
an
d
8
8
.
1
9
%,
r
esp
ec
tiv
el
y
.
T
ab
le
1
p
r
esen
ts
th
e
av
er
a
g
e
d
etec
t
io
n
ac
cu
r
ac
y
f
o
r
all
n
o
is
e
ty
p
es.
(
a)
(
b
)
(
c)
(
d
)
(
e)
(f)
Fig
u
r
e
5
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
th
e
T
I
MI
T
d
ata
b
ase
as a
p
er
ce
n
tag
e
o
f
C
OR
R
E
C
T
,
F
E
C
,
MSC
,
ND
S,
an
d
OVE
R
f
o
r
v
a
r
io
u
s
n
o
is
e
ty
p
es a
t d
if
f
er
e
n
t SNR
: (
a)
wh
ite
n
o
is
e
,
(
b
)
p
in
k
n
o
is
e
,
(
c
)
leo
p
ar
d
n
o
is
e
,
(
d
)
r
estau
r
a
n
t n
o
is
e,
(
e
)
tr
ain
n
o
is
e,
an
d
(
f
)
ca
r
n
o
is
e
T
ab
le
1
.
Av
e
r
ag
e
ac
cu
r
ac
y
o
f
d
etec
tio
n
(
%)
at
d
if
f
er
en
t SNR
s
S
N
R
(
d
B
)
V
A
D
met
h
o
d
s
A
M
R
2
[
3
2
]
VAD
-
SFF
[
2
0
]
F
u
z
z
y
E
n
[
3
3
]
VAD
-
F
D
[
3
1
]
P
r
o
p
o
se
d
m
e
t
h
o
d
-
5
7
0
.
1
7
8
2
.
1
5
9
2
.
0
5
8
1
.
3
1
8
4
.
6
8
5
8
3
.
8
5
9
4
.
5
5
9
2
.
3
7
8
6
.
4
7
9
2
.
7
4
15
8
8
.
4
4
9
4
.
6
6
9
2
.
5
3
8
8
.
4
3
9
4
.
2
6
25
9
1
.
4
5
9
5
.
9
4
9
3
.
1
3
8
9
.
9
2
9
5
.
3
7
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
8
0
5
-
3
8
1
7
3812
3
.
3
.
3
.
K
SU benchm
a
rk
ing
Fo
r
b
etter
ac
cu
r
ac
y
,
th
e
r
o
b
u
s
tn
ess
o
f
th
e
alg
o
r
ith
m
was
ev
alu
ated
u
s
in
g
th
e
Ar
a
b
ic
KSU
d
atab
ase,
r
ec
o
r
d
in
g
s
in
an
o
f
f
ice
r
o
o
m
,
an
d
with
d
if
f
er
e
n
t
r
ec
o
r
d
i
n
g
to
o
ls
.
T
h
e
av
er
a
g
e
c
o
r
r
ec
t
s
co
r
e
f
o
r
5
0
clea
n
s
p
ee
ch
s
am
p
les
in
th
e
KSU
d
atab
ase
was
9
8
.
4
8
%,
with
n
eg
lig
ib
le
NDS
an
d
OVE
R
v
a
lu
es.
T
h
e
FEC
an
d
MSC
er
r
o
r
s
ar
e
n
ea
r
ze
r
o
,
0
.
8
8
%
an
d
0
.
6
5
%,
r
esp
ec
tiv
el
y
.
T
h
ese
r
esu
lts
d
em
o
n
s
tr
ate
th
at
th
e
ap
p
r
o
ac
h
p
er
f
o
r
m
s
ex
ce
llen
tly
in
n
o
n
-
is
o
lated
r
o
o
m
s
.
T
o
ev
alu
ate
th
e
r
o
b
u
s
tn
ess
o
f
th
e
ap
p
r
o
ac
h
o
n
th
e
KSU
d
atab
ase,
v
ar
io
u
s
ty
p
es
o
f
n
o
is
e
wer
e
ad
d
ed
,
in
clu
d
in
g
w
h
ite,
r
estau
r
a
n
t,
tr
ain
,
ca
r
,
p
in
k
,
an
d
leo
p
ar
d
n
o
is
e,
at
d
i
f
f
er
en
t
SNR
lev
els
(
2
5
,
1
5
,
5
,
an
d
-
5
d
B
)
.
Fig
u
r
e
6
illu
s
tr
ates
th
e
av
er
ag
e
p
er
f
o
r
m
an
ce
o
f
th
e
KSU
d
atab
ase
with
v
ar
io
u
s
n
o
is
e
lev
els
at
d
if
f
er
e
n
t
SNR
s
.
Fig
u
r
e
7
s
h
o
ws
th
e
ac
cu
r
ac
y
o
f
th
e
s
u
g
g
ested
a
p
p
r
o
ac
h
f
o
r
s
ev
e
r
al
ty
p
es
o
f
n
o
is
e
at
d
if
f
er
en
t
S
NR
s
.
T
h
e
b
est
r
esu
lts
ar
e
o
b
tain
ed
with
ac
cu
r
ac
y
r
ates
o
f
9
6
.
7
7
%,
9
3
.
7
9
%,
9
1
.
0
8
%,
an
d
8
1
.
4
2
%
at
2
5
,
1
5
,
5
,
an
d
-
5
d
B
,
r
esp
ec
tiv
ely
.
T
h
e
r
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
als
o
p
er
f
o
r
m
s
well
f
o
r
Ar
ab
ic
n
o
is
y
s
p
ee
ch
.
Fig
u
r
e
6
.
Per
f
o
r
m
an
c
e
co
m
p
ar
is
o
n
o
f
th
e
KSU
d
atab
ase
as a
v
er
ag
e
f
o
r
d
i
f
f
er
en
t
n
o
is
e
ty
p
e
s
at
v
ar
io
u
s
SNR
Fig
u
r
e
7
.
Av
e
r
ag
e
ac
c
u
r
ac
y
o
f
d
etec
tio
n
(
%)
at
v
a
r
io
u
s
SNR
s
o
f
VAD
-
FD
[
3
1
]
an
d
th
e
p
r
o
p
o
s
ed
m
eth
o
d
3
.
4
.
Dis
cus
s
io
n
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
lev
er
ag
es
th
e
f
r
ac
tal
d
im
en
s
io
n
o
f
en
v
elo
p
es,
d
er
iv
e
d
u
s
in
g
th
e
SF
F
m
eth
o
d
,
to
d
if
f
er
en
tiate
b
etwe
en
s
p
ee
ch
an
d
n
o
n
s
p
ee
ch
s
eg
m
en
ts
.
T
h
e
SF
F
ap
p
r
o
ac
h
p
r
o
v
id
es
h
ig
h
tem
p
o
r
al
an
d
s
p
ec
tr
al
r
eso
lu
tio
n
f
o
r
co
m
p
u
tin
g
th
e
am
p
litu
d
e
e
n
v
elo
p
e
a
t
ea
ch
f
r
e
q
u
en
c
y
o
f
th
e
s
p
ee
c
h
s
ig
n
al.
T
h
e
Katz
alg
o
r
ith
m
is
u
tili
ze
d
to
d
eter
m
in
e
th
e
f
r
ac
tal
d
im
en
s
io
n
,
wh
ich
ca
p
tu
r
es
am
p
litu
d
e
v
a
r
iatio
n
,
an
d
th
e
Katz
alg
o
r
ith
m
d
e
m
o
n
s
tr
ates su
p
er
i
o
r
p
er
f
o
r
m
a
n
ce
in
d
etec
tin
g
s
p
ee
ch
ac
tiv
ity
[
3
1
]
.
T
h
e
r
esu
lts
p
r
esen
ted
in
T
ab
le
1
f
r
o
m
th
e
T
I
M
I
T
d
atab
ase
s
h
o
w
th
at
th
e
p
r
o
p
o
s
ed
ap
p
r
o
a
ch
an
d
th
e
VAD
-
SFF
[
2
0
]
u
n
d
e
r
p
o
s
itiv
e
SNR
co
n
d
itio
n
s
.
At
th
e
s
a
m
e
tim
e,
th
e
Fu
zz
y
E
n
[
3
3
]
m
eth
o
d
o
u
tp
er
f
o
r
m
s
60
65
70
75
80
85
90
95
1
0
0
p
r
o
p
o
se
d
V
A
D
_
F
D
p
r
o
p
o
se
d
V
A
D
_
F
D
p
r
o
p
o
se
d
V
A
D
_
F
D
p
r
o
p
o
se
d
VA
D_
FD
p
r
o
p
o
se
d
V
A
D
_
F
D
w
h
i
t
e
Tr
a
i
n
l
e
o
p
a
r
d
C
a
r
p
i
n
k
C
o
r
r
e
c
t
(
%)
S
N
R
(
d
B)
-5
5
15
25
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Un
s
u
p
ervis
ed
vo
ice
a
ctivity
d
etec
tio
n
b
a
s
ed
o
n
th
e
e
n
ve
lo
p
e
's fra
cta
l d
imen
s
io
n
(
N
esr
in
e
A
b
a
ja
d
d
i
)
3813
th
em
in
ad
v
er
s
e
SNR
co
n
d
itio
n
s
.
Ho
wev
er
,
th
e
Fu
zz
y
E
n
m
eth
o
d
r
eq
u
i
r
es
s
elec
tin
g
an
d
ad
ju
s
tin
g
t
h
e
g
l
o
b
al
th
r
esh
o
ld
b
ased
o
n
th
e
ty
p
e
o
f
n
o
is
e
an
d
t
h
e
SNR
lev
els.
No
tab
ly
,
th
e
ap
p
r
o
ac
h
ac
h
ie
v
es
an
ac
cu
r
ac
y
o
f
8
4
.
6
4
%
with
o
u
t
an
y
p
r
ed
ef
in
e
d
o
r
ad
j
u
s
ted
th
r
esh
o
ld
,
d
em
o
n
s
tr
atin
g
its
p
o
ten
tial
f
o
r
r
ea
l
-
wo
r
ld
a
p
p
licatio
n
s
.
I
n
ad
d
itio
n
,
th
e
al
g
o
r
ith
m
r
ed
u
ce
s
th
e
n
u
m
b
er
o
f
p
r
o
ce
s
s
ed
en
v
elo
p
es
c
o
m
p
ar
e
d
to
VAD
-
SF
F,
wh
ich
em
p
lo
y
s
1
8
2
e
n
v
elo
p
es
an
d
r
elies
o
n
t
h
e
s
tatis
tics
f
ea
tu
r
e
δ(
n
)
to
m
ak
e
a
d
ec
is
io
n
b
ased
o
n
s
tan
d
a
r
d
d
ev
iatio
n
,
m
ea
n
s
,
s
m
o
o
th
in
g
p
ar
am
eter
s
,
an
d
o
th
e
r
p
ar
am
ete
r
s
.
I
n
co
n
tr
ast,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
u
s
es o
n
ly
8
0
en
v
elo
p
es,
m
o
r
e
th
an
5
0
%
r
e
d
u
ctio
n
in
th
e
n
u
m
b
er
o
f
en
v
elo
p
es,
an
d
au
to
m
atica
lly
ca
lc
u
lates
th
e
th
r
esh
o
ld
b
ased
o
n
ly
o
n
th
e
f
r
ac
tal
d
i
m
en
s
io
n
o
f
th
e
e
n
v
elo
p
es.
T
h
e
f
r
ac
tal
d
im
en
s
io
n
ac
c
u
r
ate
ly
d
etec
ts
am
p
litu
d
e
v
ar
iatio
n
,
with
o
u
t
th
e
n
ee
d
f
o
r
o
th
er
p
ar
a
m
eter
s
,
as
in
V
AD
-
SF
F.
T
h
ese
ad
v
an
tag
es
m
ak
e
th
e
d
ev
elo
p
e
d
m
eth
o
d
h
i
g
h
ly
r
eliab
le
a
n
d
s
u
itab
le
f
o
r
f
a
v
o
r
a
b
le
SNR
co
n
d
itio
n
s
,
d
em
o
n
s
tr
atin
g
g
o
o
d
p
er
f
o
r
m
an
ce
f
o
r
a
wid
e
r
an
g
e
o
f
ap
p
licatio
n
s
.
Fig
u
r
e
8
illu
s
tr
ates
ex
am
p
les
o
f
th
e
VAD
d
ec
is
io
n
r
elate
d
t
o
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
f
o
r
t
h
e
T
I
MI
T
d
atab
ase:
Fig
u
r
e
8
(
a)
clea
n
s
p
ee
ch
,
Fig
u
r
e
8
(
b
)
s
p
ee
ch
d
e
g
r
ad
ed
b
y
wh
ite
n
o
is
e
at
2
5
d
B
,
Fig
u
r
e
8
(
c)
at
1
5
d
B
,
Fig
u
r
e
8
(
d
)
at
5
d
B
,
an
d
Fig
u
r
e
8
(
e)
at
–
5
d
B
.
T
h
e
V
AD
lab
elin
g
ac
r
o
s
s
all
SNR
l
ev
els
in
d
icate
s
g
o
o
d
p
er
f
o
r
m
an
ce
,
wh
ich
ca
n
b
e
att
r
ib
u
ted
to
th
e
h
ig
h
r
eso
lu
tio
n
p
r
o
v
id
e
d
b
y
th
e
SF
F
an
d
th
e
a
u
to
m
atic
th
r
esh
o
ld
co
m
p
u
tatio
n
b
y
f
r
ac
tal
d
im
e
n
s
io
n
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
p
er
f
o
r
m
s
well
f
o
r
p
o
s
itiv
e
SN
R
v
alu
es,
wh
ile
f
o
r
n
eg
ativ
e
SNR
v
alu
es
(
Fig
u
r
e
8
(
e)
)
,
it
ca
n
ac
cu
r
ately
d
etec
t
n
o
n
s
p
ee
ch
s
eg
m
e
n
ts
with
o
u
t
m
is
s
in
g
s
p
ee
ch
.
I
n
th
is
ca
s
e,
ad
ju
s
tin
g
th
e
p
er
ce
n
tag
e
to
s
m
o
o
th
th
e
f
in
al
d
ec
is
io
n
im
p
r
o
v
es th
e
r
esu
lts
.
Similar
ly
,
f
o
r
th
e
KSU
d
atab
a
s
e,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
ig
n
i
f
ican
tly
o
u
t
p
er
f
o
r
m
s
VAD
-
FD,
as
s
h
o
wn
in
Fig
u
r
e
9
.
T
h
e
m
ain
d
i
f
f
er
en
ce
b
etwe
en
th
e
d
ev
elo
p
e
d
alg
o
r
ith
m
an
d
VAD
-
FD
is
th
at
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ca
lcu
lates
th
e
f
r
ac
tal
d
im
en
s
io
n
in
ea
ch
f
r
am
e
o
f
ea
ch
e
n
v
el
o
p
e.
T
h
e
SF
F
ap
p
r
o
ac
h
en
a
b
le
s
th
e
ca
lcu
latio
n
o
f
th
e
en
v
elo
p
e
with
h
ig
h
tem
p
o
r
al
an
d
f
r
eq
u
en
cy
r
eso
lu
tio
n
,
wh
ich
in
tu
r
n
en
h
an
ce
s
th
e
ac
cu
r
ac
y
o
f
t
h
e
VAD
alg
o
r
ith
m
s
b
y
im
p
r
o
v
i
n
g
th
e
f
r
ac
tal
d
im
e
n
s
io
n
.
Fig
u
r
e
9
illu
s
tr
ates
th
e
VAD
d
ec
is
io
n
s
o
f
t
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
f
o
r
th
e
KSU
d
atab
ase:
Fig
u
r
e
9
(
a)
clea
n
s
p
ee
c
h
,
Fig
u
r
e
9
(
b
)
s
p
ee
ch
d
eg
r
a
d
e
d
b
y
wh
ite
n
o
is
e
at
2
5
d
B
,
Fig
u
r
e
9
(
c)
at
1
5
d
B
,
Fig
u
r
e
9
(
d
)
at
5
d
B
,
an
d
Fig
u
r
e
9
(
e)
at
–
5
d
B
.
T
h
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
s
u
cc
ess
f
u
lly
d
etec
ts
s
p
ee
ch
f
o
r
p
o
s
itiv
e
SNR
v
alu
es
(
F
ig
u
r
e
9
(
b
)
to
(
d
)
)
.
Fo
r
n
eg
ativ
e
SNR
v
alu
es
(
Fig
u
r
e
9
(
e)
)
,
it
ac
cu
r
ately
d
e
tects
n
o
n
s
p
ee
ch
s
eg
m
en
ts
with
o
u
t
m
is
s
in
g
s
p
ee
c
h
,
a
n
d
th
e
p
er
ce
n
tag
e
o
f
f
in
al
d
ec
is
io
n
s
r
em
ain
s
ac
ce
p
tab
le.
(
a)
(
b
)
(
c)
(
d
)
(
e)
Fig
u
r
e
8
.
VAD
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
f
o
r
T
I
MI
T
d
atab
ase: (
a)
clea
n
s
p
ee
ch
,
(
b
)
s
p
ee
ch
d
eg
r
ad
e
d
b
y
wh
ite
n
o
is
e
at
2
5
d
B
,
(
c)
at
1
5
d
B
,
(
d
)
at
5
d
B
,
a
n
d
(
e)
at
-
5
d
B
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t 2
0
2
6
:
3
8
0
5
-
3
8
1
7
3814
(
a)
(
b
)
(
c)
(
d
)
(
e)
Fig
u
r
e
9
.
VAD
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
f
o
r
th
e
KSU
d
atab
ase: (
a)
clea
n
s
p
ee
c
h
,
(
b
)
s
p
ee
ch
d
eg
r
a
d
ed
b
y
wh
ite
n
o
is
e
at
2
5
d
B
,
(
c
)
at
1
5
d
B
,
(
d
)
at
5
d
B
,
an
d
(
e
)
at
-
5
d
B
4.
CO
NCLU
SI
O
N
I
n
t
h
i
s
s
t
u
d
y
,
a
r
e
l
i
a
b
le
m
e
th
o
d
f
o
r
V
A
D
w
a
s
d
e
v
e
lo
p
ed
t
h
a
t
ef
f
ec
t
i
v
e
ly
d
i
s
c
r
im
i
n
a
t
es
b
e
t
w
ee
n
s
p
e
e
c
h
an
d
n
o
n
s
p
e
e
ch
s
e
g
m
e
n
t
s
.
T
h
e
m
e
t
h
o
d
o
l
o
g
y
u
s
e
s
a
co
m
b
in
a
t
io
n
o
f
S
F
F
an
d
f
r
a
c
t
a
l
d
i
m
en
s
i
o
n
a
n
a
l
y
s
i
s
.
T
h
e
en
v
e
lo
p
e
s
a
r
e
c
a
l
c
u
l
a
t
ed
u
s
in
g
th
e
S
F
F
a
p
p
r
o
a
ch
,
an
d
th
e
f
r
a
c
t
a
l
d
i
m
e
n
s
i
o
n
o
f
t
h
e
s
e
e
n
v
e
lo
p
e
s
i
s
o
b
t
a
in
ed
u
s
i
n
g
t
h
e
K
a
tz
a
lg
o
r
i
t
h
m
.
T
h
r
e
s
h
o
ld
s
a
r
e
th
e
n
d
e
t
e
r
m
i
n
ed
a
u
to
m
a
t
i
c
a
l
l
y
f
o
r
e
a
ch
e
n
v
e
lo
p
e
to
m
ak
e
t
h
e
f
in
a
l
V
A
D
d
ec
i
s
i
o
n
.
T
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
's
p
e
r
f
o
r
m
a
n
c
e
i
s
ev
a
l
u
a
te
d
u
s
in
g
t
w
o
d
i
f
f
e
r
en
t
d
a
t
a
b
a
s
e
s
,
n
a
m
e
ly
,
t
h
e
T
I
M
I
T
a
n
d
K
S
U
d
a
t
ab
a
s
e
s
.
T
h
e
r
e
s
u
l
t
s
s
h
o
w
t
h
a
t
t
h
e
V
A
D
d
ec
i
s
i
o
n
i
s
r
o
b
u
s
t
f
o
r
n
o
i
s
y
an
d
c
l
e
an
s
p
e
e
ch
.
T
h
e
ap
p
r
o
a
ch
s
ig
n
if
i
ca
n
t
l
y
s
u
r
p
a
s
s
e
d
o
r
m
a
t
c
h
e
d
th
e
p
e
r
f
o
r
m
an
c
e
o
f
o
t
h
er
m
e
t
h
o
d
s
s
u
ch
a
s
A
MR
2
,
V
A
D
-
S
F
F
,
F
u
zz
y
E
n
,
an
d
V
A
D
-
F
D
.
M
o
r
eo
v
er
,
t
h
is
m
e
t
h
o
d
r
e
d
u
c
e
s
m
e
m
o
r
y
a
n
d
t
im
e
c
o
n
s
u
m
p
t
io
n
a
n
d
d
o
e
s
n
o
t
r
e
q
u
i
r
e
a
n
y
t
r
a
in
i
n
g
d
a
t
a
o
r
p
r
i
o
r
k
n
o
w
le
d
g
e
o
f
s
p
e
e
c
h
o
r
n
o
i
s
e
c
h
ar
a
c
t
er
i
s
t
i
c
s
.
T
h
e
d
e
v
e
l
o
p
ed
a
lg
o
r
i
t
h
m
o
f
f
e
r
s
a
n
e
f
f
e
c
t
i
v
e
an
d
ef
f
i
c
i
en
t
s
o
l
u
t
io
n
f
o
r
V
A
D
,
e
n
h
an
c
in
g
th
e
p
e
r
f
o
r
m
a
n
c
e
o
f
s
p
e
e
ch
p
r
o
ce
s
s
i
n
g
ap
p
l
ic
a
t
io
n
s
.
F
in
a
l
l
y
,
i
n
s
h
o
r
t
t
e
r
m
,
f
o
c
u
s
w
i
l
l
b
e
p
la
c
e
d
on
h
y
b
r
id
f
r
a
c
t
a
l
d
i
m
en
s
i
o
n
–
m
a
c
h
in
e
l
e
ar
n
in
g
m
o
d
e
l
s
,
a
n
d
i
n
l
o
n
g
t
e
r
m
f
o
r
r
e
a
l
t
im
e
d
ep
l
o
y
m
en
t
i
n
s
p
e
e
ch
r
e
c
o
g
n
i
t
io
n
s
y
s
t
e
m
s
.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
e
au
th
o
r
s
ex
p
r
ess
th
eir
g
r
atitu
d
e
to
th
e
R
esear
ch
er
s
S
u
p
p
o
r
tin
g
Pro
ject
(
R
SP
2
0
2
3
R
3
4
)
,
Kin
g
Sau
d
Un
iv
er
s
ity
,
R
iy
ad
h
,
Sau
d
i A
r
ab
ia,
f
o
r
th
e
s
u
p
p
o
r
t
r
ec
e
iv
ed
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
e
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.