I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
9
,
No
.
2
,
Au
g
u
s
t
2020
,
p
p
.
63
~
72
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct.
v
9
i2
.
p
p
63
-
72
63
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Speech
enha
ncem
ent
using
mo
di
fie
d
wiene
r f
il
ter b
a
sed M
M
SE
a
nd speech
prese
nce proba
bility
es
tima
tion
V
.
Vij
a
y
a
s
ri
B
o
lis
et
t
y
1
,
U.
Y
eduk
o
nd
a
lu
2
,
I
.
Sa
nti
p
ra
bh
a
3
1
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
E
n
g
i
n
e
e
rin
g
,
Ad
i
t
y
a
Co
ll
e
g
e
o
f
E
n
g
in
e
e
rin
g
a
n
d
Te
c
h
n
o
l
o
g
y
,
I
n
d
ia
2
El
e
c
tro
n
ics
&
Co
m
m
u
n
ica
ti
o
n
E
n
g
i
n
e
e
rin
g
De
p
a
rtme
n
t,
S
ri
Va
sa
v
i
E
n
g
i
n
e
e
rin
g
Co
ll
e
g
e
,
In
d
ia
3
De
p
a
rtme
n
t
o
f
El
e
c
tro
n
ics
&
Co
m
m
u
n
ica
ti
o
n
En
g
in
e
e
rin
g
,
U
n
iv
e
rsity
Co
ll
e
g
e
o
f
En
g
in
e
e
ri
n
g
,
JN
TUK,
In
d
ia
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Mar
6
,
2
0
19
R
ev
is
ed
Dec
1
2
,
2
0
19
Acc
ep
ted
J
an
1
2
,
2
0
20
In
t
h
e
p
re
se
n
t
-
d
a
y
c
o
m
m
u
n
ica
ti
o
n
s
sp
e
e
c
h
si
g
n
a
ls
g
e
t
c
o
n
tam
in
a
t
e
d
d
u
e
to
v
a
rio
u
s
so
rts
o
f
n
o
ise
s
th
a
t
d
e
g
ra
d
e
th
e
sp
e
e
c
h
q
u
a
li
t
y
a
n
d
a
d
v
e
rse
ly
imp
a
c
ts
sp
e
e
c
h
re
c
o
g
n
it
i
o
n
p
e
rf
o
rm
a
n
c
e
.
To
o
v
e
rc
o
m
e
th
e
se
issu
e
s,
a
n
o
v
e
l
a
p
p
ro
a
c
h
fo
r
sp
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
u
sin
g
M
o
d
if
ied
Wi
e
n
e
r
fi
lt
e
rin
g
is
d
e
v
e
lo
p
e
d
a
n
d
p
o
we
r
sp
e
c
tru
m
c
o
m
p
u
tatio
n
is
a
p
p
l
ied
f
o
r
d
e
g
ra
d
e
d
sig
n
a
l
to
o
b
tain
th
e
n
o
ise
c
h
a
ra
c
teristics
fro
m
a
n
o
isy
sp
e
c
tru
m
.
I
n
n
e
x
t
p
h
a
se
,
M
M
S
E
tec
h
n
i
q
u
e
is
a
p
p
li
e
d
wh
e
re
G
a
u
ss
ian
d
istri
b
u
ti
o
n
o
f
e
a
c
h
sig
n
a
l
i
.
e
.
o
r
ig
i
n
a
l
a
n
d
n
o
is
y
sig
n
a
l
is
a
n
a
l
y
z
e
d
.
Th
e
G
a
u
ss
ian
d
istr
ib
u
ti
o
n
p
r
o
v
i
d
e
s
sp
e
c
tru
m
e
stim
a
ti
o
n
a
n
d
sp
e
c
tral
c
o
e
fficie
n
t
p
a
ra
m
e
ter
s
wh
ich
c
a
n
b
e
u
se
d
fo
r
p
r
o
b
a
b
i
li
stic
m
o
d
e
l
fo
rm
u
latio
n
.
M
o
re
o
v
e
r,
a
-
p
r
io
ri
-
S
NR
c
o
m
p
u
tati
o
n
is
a
lso
i
n
c
o
r
p
o
ra
ted
fo
r
c
o
e
fficie
n
t
u
p
d
a
ti
o
n
a
n
d
n
o
ise
p
r
e
se
n
c
e
e
stim
a
ti
o
n
wh
ich
o
p
e
ra
tes
sim
il
a
r
to
th
e
c
o
n
v
e
n
t
io
n
a
l
VA
D.
Ho
we
v
e
r,
c
o
n
v
e
n
ti
o
n
a
l
VA
D
sc
h
e
m
e
is
b
a
se
d
o
n
t
h
e
h
a
rd
t
h
re
sh
o
l
d
w
h
ich
is
n
o
t
c
a
p
a
b
le
to
d
e
ri
v
e
sa
ti
sfa
c
to
ry
p
e
rfo
rm
a
n
c
e
a
n
d
a
so
ft
-
d
e
c
isio
n
b
a
se
d
th
re
sh
o
ld
is
d
e
v
e
lo
p
e
d
fo
r
imp
r
o
v
i
n
g
th
e
p
e
rfo
rm
a
n
c
e
o
f
sp
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t.
An
e
x
ten
s
iv
e
sim
u
latio
n
stu
d
y
is
c
a
rried
o
u
t
u
sin
g
M
ATLAB
sim
u
lati
o
n
to
o
l
o
n
N
OIZE
US
sp
e
e
c
h
d
a
tab
a
se
a
n
d
a
c
o
m
p
a
ra
ti
v
e
stu
d
y
is
p
re
se
n
ted
wh
e
re
p
ro
p
o
s
e
d
a
p
p
ro
a
c
h
is
p
r
o
v
e
d
b
e
tt
e
r
in
c
o
m
p
a
riso
n
with
e
x
isti
n
g
tec
h
n
i
q
u
e
.
K
ey
w
o
r
d
s
:
A
-
p
r
io
r
i
-
SNR
Gau
s
s
ian
d
is
tr
ib
u
tio
n
No
is
e
tr
ac
k
in
g
PSD
SPP
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
:
V.
Vijay
asri
B
o
lis
etty
,
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
,
Ad
ity
a
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
an
d
T
ec
h
n
o
l
o
g
y
,
Ad
ity
a
Nag
ar
,
ADB R
d
,
Su
r
a
m
p
alem
,
An
d
h
r
a
Pra
d
esh
,
5
3
3
4
3
7
,
I
n
d
ia.
E
m
ail:
v
asu
d
h
a.
twee
ty
@
g
m
ai
l.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
No
w
-
a
-
d
ay
s
,
s
p
ee
c
h
co
m
m
u
n
icatio
n
s
h
av
e
d
r
awn
d
em
a
n
d
with
th
e
ad
v
en
t
o
f
d
if
f
er
e
n
t
b
r
o
ad
b
an
d
an
d
m
u
ltime
d
ia
a
p
p
licatio
n
s
in
r
ap
id
u
s
ag
e
at
d
if
f
er
e
n
t
e
n
v
ir
o
n
m
en
ts
.
I
n
s
u
c
h
s
ce
n
ar
i
o
s
,
p
r
eser
v
in
g
th
e
in
f
o
r
m
atio
n
in
s
p
ee
ch
s
ig
n
al
p
lay
s
an
im
p
o
r
tan
t
r
o
le
wh
i
ch
h
as
im
p
ac
t
o
n
th
e
e
f
f
icie
n
t
co
m
m
u
n
icatio
n
.
Mo
r
eo
v
er
,
d
u
e
to
l
o
w
-
co
s
t
d
ig
ital
s
ig
n
al
p
r
o
ce
s
s
o
r
s
an
d
m
e
m
o
r
y
c
h
ip
s
,
s
p
ee
ch
p
r
o
ce
s
s
in
g
ap
p
licatio
n
s
,
v
o
ice
co
m
m
u
n
icatio
n
s
y
s
tem
s
an
d
v
o
ice
r
ec
o
g
n
itio
n
s
y
s
tem
s
ar
e
wid
ely
u
tili
ze
d
in
v
ar
io
u
s
r
ea
l
-
tim
e
ap
p
licatio
n
s
ce
n
ar
io
s
[
1
]
.
Var
io
u
s
c
o
m
m
u
n
icatio
n
d
e
v
ices
also
h
av
e
b
ee
n
m
an
u
f
ac
tu
r
e
d
i.e
.
m
o
b
ile
p
h
o
n
es
an
d
m
ics
etc
.
wh
ich
ca
n
b
e
u
s
ed
f
o
r
in
d
o
o
r
as
well
as
o
u
td
o
o
r
ap
p
licatio
n
s
.
I
n
th
ese
en
v
ir
o
n
m
e
n
ts
,
b
ac
k
g
r
o
u
n
d
n
o
is
e
is
p
r
esen
t
wh
ich
g
ets
ac
cu
m
u
lat
ed
with
th
e
o
r
ig
in
al
s
ig
n
al
r
es
u
ltin
g
d
eg
r
a
d
ed
s
p
ee
ch
q
u
ality
an
d
it
m
ay
lead
to
th
e
p
er
f
o
r
m
an
ce
d
eg
r
a
d
atio
n
o
f
v
o
ice
r
ec
o
g
n
itio
n
s
y
s
tem
s
[
2
]
.
I
n
o
u
td
o
o
r
e
n
v
ir
o
n
m
en
ts
lik
e
h
ea
v
y
tr
af
f
ic,
b
u
s
ter
m
in
als,
m
alls
,
etc.
,
p
r
esen
ce
o
f
v
ar
io
u
s
ty
p
es
o
f
b
ac
k
g
r
o
u
n
d
n
o
is
es
co
n
tam
in
a
te
th
e
o
r
ig
in
al
s
ig
n
al
th
at
lea
d
s
to
im
p
r
o
p
er
co
m
m
u
n
icati
o
n
[
3
]
in
r
ea
l
-
tim
e
ap
p
licatio
n
s
lik
e
s
p
ea
k
e
r
r
ec
o
g
n
itio
n
s
y
s
tem
,
m
o
b
ile
co
m
m
u
n
icatio
n
,
h
ea
r
in
g
aid
s
etc.
H
en
ce
n
o
is
e
is
to
b
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
9
,
No
.
2,
Au
g
u
s
t
20
20
:
63
–
7
2
64
esti
m
ated
an
d
r
ed
u
ce
d
to
im
p
r
o
v
e
th
e
s
p
ee
ch
q
u
ality
.
Du
r
in
g
last
d
ec
ad
e,
v
ar
i
o
u
s
en
h
a
n
ce
m
en
t
te
ch
n
iq
u
es
h
av
e
b
ee
n
p
r
esen
ted
f
o
r
s
p
ee
ch
s
ig
n
al
en
h
an
ce
m
e
n
t.
Su
b
s
tan
tially
,
th
ese
tech
n
iq
u
es
ar
e
ca
teg
o
r
ized
s
in
g
le
-
ch
an
n
el
an
d
m
u
lti
-
ch
a
n
n
el
s
p
ee
ch
en
h
an
ce
m
en
t.
Usu
ally
,
b
ac
k
g
r
o
u
n
d
n
o
is
e
f
ea
tu
r
es
ca
n
b
e
id
en
tifie
d
u
s
in
g
s
in
g
le
ch
an
n
el
s
p
ee
ch
e
n
h
an
c
em
en
t
a
n
d
th
e
ef
f
ec
ts
o
f
r
ev
er
b
er
atio
n
ca
n
b
e
r
ed
u
ce
d
th
r
o
u
g
h
m
u
ltich
an
n
el
s
p
ee
ch
en
h
a
n
ce
m
en
t.
T
h
o
u
g
h
,
Mu
ltich
an
n
el
s
ch
em
es
s
h
o
w
s
ig
n
if
ican
t
p
er
f
o
r
m
a
n
ce
f
o
r
s
p
ee
ch
en
h
a
n
ce
m
en
t
wh
en
co
m
p
ar
e
d
to
s
in
g
le
-
ch
an
n
el
s
p
ee
ch
en
h
an
ce
m
en
t,
th
e
en
h
an
ce
m
e
n
t
is
to
b
e
d
o
n
e
th
r
o
u
g
h
ea
ch
m
icr
o
p
h
o
n
e
in
d
iv
id
u
ally
.
Hen
ce
th
e
wo
r
k
is
co
n
cise to
s
in
g
l
e
ch
an
n
el
s
p
ee
ch
en
h
a
n
ce
m
en
t.
T
h
e
s
p
ee
ch
en
h
an
ce
m
en
t
alg
o
r
ith
m
aim
s
to
r
ec
tify
th
e
d
am
ag
ed
in
p
u
t
o
r
o
u
t
p
u
t
s
ig
n
al
an
d
to
in
cr
ea
s
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
co
m
m
u
n
icatio
n
lin
k
.
T
h
e
d
am
ag
ed
s
p
ee
ch
s
ig
n
al
g
en
er
ates
h
u
g
e
tr
o
u
b
l
e
m
ain
ly
f
o
r
s
p
ee
ch
r
ec
o
g
n
itio
n
ap
p
licatio
n
s
.
T
h
e
q
u
ality
an
d
i
n
tellig
ib
ilit
y
o
f
s
p
ee
ch
a
r
e
d
a
m
ag
ed
b
y
th
e
n
o
is
e
in
v
o
lv
ed
in
th
e
s
p
ee
ch
s
ig
n
al.
T
h
e
ter
m
in
tellig
ib
ilit
y
d
en
o
tes
th
e
u
n
d
er
s
tan
d
ab
ilit
y
o
f
f
i
n
al
o
u
tco
m
e
o
f
th
e
sp
ee
ch
s
ig
n
al.
T
h
e
ac
cu
r
ac
y
o
f
th
e
ex
ac
t
co
n
te
n
t
o
f
s
p
ee
ch
s
ig
n
al
is
ter
m
ed
as
q
u
ality
.
T
h
e
d
if
f
er
en
t
ty
p
es
o
f
s
p
ee
ch
en
h
an
ce
m
e
n
t
alg
o
r
ith
m
u
s
ed
to
d
ed
u
ct
n
o
is
e
ar
e
Ad
ap
tiv
e
o
r
n
o
n
-
ad
a
p
tiv
e,
f
r
e
q
u
e
n
cy
o
r
tim
e
d
o
m
ain
[
4
]
.
O
v
er
th
e
last
y
ea
r
s
,
s
ev
er
al
tech
n
iq
u
es
h
av
e
b
ee
n
p
r
ese
n
ted
f
o
r
s
p
ee
ch
s
ig
n
al
en
h
an
c
em
en
t
i.e
.
Sp
ec
tr
al
Su
b
tr
ac
tio
n
,
W
av
elet
b
ased
m
eth
o
d
s
,
m
o
d
el
-
b
ased
tech
n
i
q
u
es
an
d
f
ilter
in
g
tec
h
n
iq
u
es
[
5
,
6
]
.
Fu
r
th
er
m
o
r
e,
s
p
ee
ch
en
h
an
ce
m
en
t
tech
n
i
q
u
es
ca
n
b
e
ca
teg
o
r
ized
as
s
p
ec
tr
al
an
d
tem
p
o
r
al
p
r
o
ce
s
s
in
g
tech
n
iq
u
es.
Acc
o
r
d
in
g
to
t
h
e
s
p
ec
tr
al
d
o
m
ain
p
r
o
ce
s
s
,
co
r
r
u
p
ted
s
ig
n
al
is
p
r
o
ce
s
s
ed
th
r
o
u
g
h
th
e
tr
an
s
f
o
r
m
d
o
m
ai
n
tech
n
iq
u
e
wh
e
r
ea
s
tem
p
o
r
al
p
r
o
ce
s
s
in
g
m
eth
o
d
u
s
es
tim
e
-
d
o
m
ain
an
aly
s
is
f
o
r
im
p
r
o
v
in
g
th
e
q
u
ality
o
f
th
e
s
p
ee
ch
s
ig
n
al.
T
h
e
co
m
p
lete
ar
ticle
is
o
r
g
a
n
i
ze
d
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
es
en
ts
a
b
r
ief
liter
atu
r
e
s
u
r
v
e
y
a
n
d
r
ec
e
n
t
ad
v
an
ce
m
e
n
ts
in
s
p
ee
ch
en
h
a
n
ce
m
en
t
tech
n
iq
u
es.
Sectio
n
3
p
r
esen
ts
p
r
o
p
o
s
ed
s
p
ee
c
h
e
n
h
an
ce
m
e
n
t
s
y
s
tem
m
o
d
elin
g
.
R
esu
lts
ar
e
an
aly
z
ed
an
d
d
is
cu
s
s
io
n
is
elab
o
r
at
ed
in
s
ec
tio
n
4
,
f
i
n
ally
,
co
n
c
lu
d
in
g
r
em
ar
k
s
a
r
e
p
r
esen
ted
in
th
e
s
ec
tio
n
5
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
C
o
n
v
en
tio
n
al
m
u
ltib
an
d
s
p
ee
c
h
en
h
an
ce
m
en
t
ac
q
u
ir
es
two
o
p
er
atio
n
s
:
o
n
e
is
s
p
litt
in
g
th
e
s
p
ec
tr
u
m
in
to
f
r
e
q
u
en
c
y
b
an
d
s
,
an
d
th
e
o
th
er
is
e
x
ec
u
tin
g
s
p
ee
ch
en
h
an
ce
m
en
t
in
ea
ch
b
a
n
d
i
n
d
ep
e
n
d
en
tly
.
T
h
e
p
o
l
o
-
in
ter
ac
tio
n
p
r
o
b
le
m
in
th
e
s
p
ec
tr
al
d
o
m
ai
n
lead
s
to
t
h
e
s
u
p
p
r
ess
io
n
o
f
f
ew
co
e
f
f
icien
ts
in
th
e
esti
m
atio
n
o
f
clea
n
s
p
ee
ch
b
y
th
e
in
f
lu
e
n
ce
o
f
th
e
f
o
r
m
a
n
ts
in
th
e
n
eig
h
b
o
r
in
g
b
a
n
d
s
an
d
t
h
u
s
g
r
ad
es
in
p
o
o
r
q
u
ality
.
T
o
r
ed
u
ce
th
e
d
o
m
in
atio
n
o
f
s
tr
o
n
g
er
f
o
r
m
an
ts
o
v
er
th
e
n
eig
h
b
o
r
in
g
b
an
d
s
,
th
e
ass
ess
m
en
t
o
f
clea
n
s
p
ee
ch
is
d
o
n
e
b
y
,
in
th
e
tem
p
o
r
al
d
o
m
ain
.
T
h
e
u
n
s
u
p
p
r
ess
ed
s
p
ee
ch
is
f
ilter
ed
in
to
v
ar
io
u
s
eq
u
iv
alen
t
r
ec
tan
g
u
lar
b
an
d
wid
th
b
ased
s
u
b
b
an
d
s
a
n
d
f
o
llo
wed
b
y
en
h
a
n
ce
m
en
t
o
f
s
p
ec
tr
al
s
p
ee
ch
i
n
ea
ch
b
an
d
b
ased
o
n
Dis
cr
ete
C
o
s
in
e
T
r
an
s
f
o
r
m
(
DC
T
)
u
s
in
g
Sp
ec
tr
al
Su
b
tr
ac
tio
n
/Min
im
u
m
Me
an
Sq
u
a
r
e
E
r
r
o
r
(
MM
S
E
)
[
7
]
.
Par
k
et.
al.
[
8
]
d
is
cu
s
s
ed
ab
o
u
t
th
e
u
s
e
o
f
s
p
ee
ch
en
h
an
ce
m
en
t
tech
n
iq
u
e
in
m
o
b
ile
co
m
m
u
n
icatio
n
s
y
s
tem
s
.
Au
th
o
r
s
d
e
v
elo
p
e
d
a
n
ef
f
icie
n
t
s
ch
em
e
f
o
r
n
o
is
e
r
ed
u
ctio
n
wh
ich
ca
n
im
p
r
o
v
e
th
e
p
e
r
f
o
r
m
an
ce
o
f
s
p
ee
ch
r
ec
o
g
n
itio
n
.
Ho
wev
e
r
,
m
o
b
ile
d
e
v
ices
h
av
e
lim
ite
d
ca
p
ac
ity
wh
ich
m
o
tiv
ates
to
d
ev
el
o
p
a
lo
w
co
m
p
lex
s
ch
em
e
f
o
r
n
o
is
e
r
ed
u
ctio
n
.
I
n
t
h
is
wo
r
k
,
a
s
p
ee
ch
co
d
er
is
also
u
tili
ze
d
f
o
r
p
ac
k
et
d
ata
esti
m
atio
n
.
I
n
g
en
er
al,
th
is
wo
r
k
u
s
es
p
itch
in
f
o
r
m
atio
n
f
o
r
c
o
m
b
f
ilter
i
n
g
.
Ad
ap
tiv
e
f
ilter
in
g
tech
n
iq
u
e
h
as
a
s
ig
n
if
ican
t
im
p
ac
t
o
n
th
e
s
p
ee
ch
p
r
o
ce
s
s
in
g
s
y
s
tem
.
Ad
ap
tiv
e
f
ilter
in
g
h
as
b
ee
n
u
s
ed
wid
ely
in
th
e
v
ar
io
u
s
ap
p
licatio
n
s
s
u
ch
as c
h
an
n
el
eq
u
aliza
tio
n
,
s
y
s
tem
id
en
tific
atio
n
an
d
ec
h
o
ca
n
ce
llatio
n
.
Alth
o
u
g
h
,
MM
S
E
an
d
DW
T
s
h
o
w
s
im
p
r
o
v
em
e
n
t in
s
p
ee
ch
en
h
a
n
ce
m
en
t,
th
eir
p
e
r
f
o
r
m
an
ce
is
p
o
o
r
at
lo
w
SNR
co
n
d
itio
n
s
[
9
,
1
0
]
.
T
h
e
Du
al
T
r
ee
C
o
m
p
lex
W
av
elet
Pack
et
T
r
an
s
f
o
r
m
(
D
T
C
W
PT)
em
p
lo
y
ed
in
[
1
1
]
p
r
o
v
id
es
a
s
o
lu
tio
n
to
av
o
i
d
th
e
aliasin
g
an
d
o
s
cillatio
n
s
d
u
e
to
s
h
if
t
i
n
v
ar
ian
ce
in
DW
T
.
E
v
en
t
h
o
u
g
h
c
o
m
p
atib
ilit
y
ex
is
ts
b
etwe
en
MM
SE
b
ased
SP
P
an
d
DT
C
W
PT,
th
e
lo
s
s
o
f
in
tellig
ib
ilit
y
in
th
e
r
ec
o
n
s
tr
u
cted
s
p
ee
ch
d
u
e
to
th
e
f
in
ite
wa
v
elet
f
ilter
s
.
I
n
t
h
is
f
ield
,
C
h
o
i.
et.
al.
[
1
2
]
im
p
r
o
v
ed
th
e
ex
is
tin
g
ad
ap
tiv
e
f
ilter
in
g
an
d
d
ev
elo
p
ed
a
n
o
v
el
s
u
b
-
b
an
d
ad
ap
tiv
e
f
ilt
er
.
Fu
r
th
e
r
,
th
is
tech
n
iq
u
e
tak
es
th
e
ad
v
a
n
tag
e
o
f
n
o
r
m
-
o
p
ti
m
izatio
n
an
d
n
o
r
m
as
th
e
co
m
p
u
tatio
n
o
f
co
s
t
f
u
n
ctio
n
.
Au
t
h
o
r
s
h
a
v
e
claim
e
d
th
at
th
is
tech
n
iq
u
e
is
ca
p
a
b
le
to
im
p
r
o
v
e
th
e
s
y
s
tem
p
er
f
o
r
m
a
n
ce
an
d
r
o
b
u
s
tn
ess
in
im
p
u
ls
iv
e
n
o
is
e
s
ce
n
ar
io
s
.
T
h
e
o
p
tim
ally
m
o
d
if
ie
d
MM
SE
b
ased
lo
g
Sp
ec
tr
al
Am
p
li
tu
d
e
esti
m
ato
r
(
OM
-
L
SA)
im
p
r
o
v
es
th
e
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
th
e
MM
SE
in
ca
s
e
o
f
s
tatio
n
ar
y
a
n
d
s
o
m
e
s
o
r
t
o
f
n
o
n
-
s
tatio
n
ar
y
n
o
is
es
b
u
t
s
h
o
ws
s
ig
n
if
ican
tly
p
o
o
r
p
er
f
o
r
m
an
c
e
a
t
lo
w
an
d
h
ig
h
SNR
co
n
d
itio
n
s
[
1
3
,
1
4
]
.
SMPO
b
ased
MM
SE
tech
n
iq
u
e
f
o
r
s
p
ee
ch
en
h
an
ce
m
en
t
im
p
r
o
v
e
s
th
e
PESQ
in
all
k
in
d
s
o
f
n
o
is
y
en
v
ir
o
n
m
en
ts
.
B
in
ar
y
m
ask
in
g
r
esu
lts
g
o
o
d
s
p
ee
ch
q
u
ality
b
u
t
s
o
m
etim
es in
tellig
ib
ilit
y
o
f
th
e
s
p
ee
ch
m
a
y
b
e
lo
s
t
d
u
e
to
o
v
er
th
r
esh
o
l
d
s
in
m
ask
in
g
n
o
is
e
co
ef
f
icien
ts
.
SMPO
-
W
ein
er
i
s
g
o
in
g
to
p
r
o
v
id
e
b
etter
s
o
l
u
tio
n
in
th
ese
is
s
u
es
[
1
5
]
.
Hen
ce
co
m
b
in
atio
n
o
f
W
ein
er
f
ilter
with
MM
SE
b
a
s
ed
SP
P
th
r
o
u
g
h
s
o
f
t
d
ec
is
io
n
m
ay
im
p
r
o
v
e
th
e
s
p
ee
ch
in
tellig
ib
ilit
y
in
ad
d
itio
n
to
s
p
ee
ch
q
u
ality
.
−
C
h
allen
g
es
in
Sp
ee
ch
E
n
h
an
c
em
en
t:
Sp
ee
ch
is
co
n
s
id
er
e
d
a
s
p
o
wer
f
u
l
m
o
d
e
o
f
co
m
m
u
n
i
ca
tio
n
n
o
t
o
n
ly
f
o
r
h
u
m
a
n
s
b
u
t
also
f
o
r
h
u
m
a
n
m
ac
h
in
e
in
ter
f
ac
e.
So
m
etim
es
th
e
s
p
ee
ch
s
ig
n
al
tr
av
els
th
r
o
u
g
h
th
e
n
o
is
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
S
p
ee
ch
e
n
h
a
n
ce
men
t u
s
in
g
mo
d
ified
w
ien
er filter
b
a
s
ed
MM
S
E
a
n
d
s
p
ee
ch
…
(
V
.
V
ija
ya
s
r
i B
o
lis
etty
)
65
m
ed
iu
m
b
ef
o
r
e
r
ea
ch
i
n
g
t
o
th
e
lis
ten
er
(
r
ec
o
g
n
itio
n
s
y
s
tem
)
.
No
w
-
a
-
d
ay
s
,
au
t
o
m
ated
s
p
e
ec
h
p
r
o
ce
s
s
in
g
s
y
s
tem
s
h
av
e
g
ain
ed
lo
t
o
f
a
ttra
ctio
n
f
r
o
m
r
esear
c
h
er
s
an
d
h
av
e
b
ee
n
ad
o
p
ted
wid
ely
in
th
e
r
ea
l
-
life
s
ce
n
ar
io
s
.
State
-
of
-
ar
t
m
o
d
el
s
o
f
au
to
m
ated
s
p
ee
ch
p
r
o
c
ess
in
g
s
y
s
tem
s
wo
r
k
s
well
with
co
n
tr
o
lled
en
v
ir
o
n
m
en
t
b
u
t
r
ea
l
-
tim
e
s
y
s
tem
s
s
u
f
f
er
f
r
o
m
v
ar
io
u
s
b
a
ck
g
r
o
u
n
d
n
o
is
es,
r
ev
er
b
er
atio
n
an
d
s
p
ee
ch
f
r
o
m
o
th
er
s
p
ea
k
er
s
,
w
h
ich
c
au
s
es
p
ar
tial
lo
s
s
o
f
i
n
f
o
r
m
atio
n
to
co
m
p
lete
lo
s
s
esp
ec
ially
at
lo
w
SNR
.
No
is
e
r
em
o
v
al
ca
n
b
e
d
o
n
e
o
n
ly
b
y
id
en
tif
y
in
g
th
e
ch
a
r
ac
ter
is
tics
o
f
th
e
n
o
is
e
th
r
o
u
g
h
p
r
ep
r
o
ce
s
s
in
g
.
Ho
wev
er
,
h
u
g
e
n
u
m
b
er
o
f
r
esear
ch
es
h
av
e
b
ee
n
ca
r
r
ied
o
u
t
i
n
th
is
f
ield
b
u
t
d
u
e
to
co
m
p
u
tatio
n
al
co
m
p
lex
ities
,
s
p
ee
ch
en
h
a
n
c
em
en
t
s
till
r
em
ain
s
a
ch
allen
g
in
g
task
as
th
e
p
r
o
ce
s
s
its
elf
in
tr
o
d
u
ce
s
co
m
p
lex
ity
.
−
C
o
n
tr
ib
u
tio
n
o
f
th
e
wo
r
k
.
Ma
i
n
co
n
tr
ib
u
tio
n
s
o
f
th
is
wo
r
k
ar
e
as f
o
llo
ws:
a.
I
n
itially
,
th
e
f
ea
tu
r
es
o
f
th
e
s
p
ee
ch
s
ig
n
al,
lik
e
s
ilen
ce
is
esti
m
ated
f
r
o
m
th
e
n
o
is
y
s
ig
n
a
l
s
p
ec
tr
u
m
th
r
o
u
g
h
Sp
ee
ch
Pre
s
en
ce
Pr
o
b
ab
ilit
y
(
SP
P)
esti
m
atio
n
.
b.
E
x
tr
ac
tin
g
th
e
n
o
is
e
ch
ar
ac
ter
i
s
tics
f
r
o
m
th
e
s
ilen
ce
.
c.
Fin
ally
,
MM
SE
(
No
is
e
T
r
ac
k
er
)
b
ased
n
o
is
e
p
o
wer
esti
m
a
tio
n
to
s
u
p
p
r
ess
th
e
n
o
is
y
s
ig
n
al
u
s
in
g
n
o
is
e
p
o
wer
tr
ac
k
in
g
s
ch
em
e
.
3.
SI
NG
L
E
-
CH
ANN
E
L
SPE
E
CH
E
NH
ANC
E
M
E
N
T
Ass
u
m
in
g
th
at
th
e
m
ajo
r
ity
o
f
th
e
tr
an
s
m
is
s
io
n
n
o
is
e
is
ad
d
i
tiv
e,
th
e
s
p
ee
ch
en
h
an
ce
m
en
t f
o
r
s
in
g
le
-
ch
an
n
el
is
r
ep
r
esen
te
d
in
(
1
)
,
(
)
=
(
)
+
(
)
(
1
)
wh
er
e
x(
n
)
,
d
(
n
)
,
y(
n
)
co
r
r
esp
o
n
d
s
to
s
p
ee
ch
,
n
o
is
e
an
d
n
o
is
y
s
p
ee
ch
s
ig
n
als
an
d
n
in
d
i
ca
tes
d
is
cr
ete
tim
e
in
d
ex
.
B
o
th
x(
n
)
an
d
d
(
n
)
ar
e
i
n
d
ep
en
d
en
t
with
ze
r
o
m
ea
n
v
a
lu
e.
T
h
e
b
lo
ck
d
iag
r
am
o
f
s
in
g
le
ch
an
n
el
s
p
ee
ch
en
h
an
ce
m
e
n
t
is
r
ep
r
esen
ted
i
n
Fig
u
r
e
1
.
T
o
esti
m
ate
th
e
c
lean
s
p
ee
ch
s
ig
n
al
x
(
n
)
,
th
e
n
o
is
y
s
ig
n
al
y(
n
)
is
p
r
o
ce
s
s
ed
u
s
in
g
s
p
ee
ch
en
h
a
n
ce
m
en
t a
lg
o
r
ith
m
s
.
Fig
u
r
e
1
.
Gen
e
r
al
Sp
ee
ch
E
n
h
an
ce
m
en
t Sy
s
tem
s
An
an
aly
s
is
o
f
s
p
ee
ch
en
h
an
c
em
en
t
m
eth
o
d
f
o
r
a
n
o
is
y
s
p
ee
ch
s
ig
n
al
y(
n
)
is
illu
s
tr
ated
in
Fig
u
r
e
2
.
I
n
itially
,
th
e
n
o
is
y
s
ig
n
al
is
s
eg
m
en
ted
in
to
o
v
e
r
lap
p
in
g
f
r
am
es,
th
en
tr
an
s
f
o
r
m
ed
in
t
o
f
r
eq
u
en
cy
d
o
m
ain
u
s
in
g
DFT
o
r
STFT
.
T
o
ac
h
iev
e
g
o
o
d
q
u
ality
o
f
s
p
ee
c
h
s
ig
n
al,
DFT
tech
n
iq
u
e
is
ap
p
lied
as
it
ea
s
ily
u
n
d
er
s
tan
d
s
th
e
s
p
ec
tr
al
c
o
n
te
n
t o
f
th
e
s
ig
n
al.
Fig
u
r
e
2
.
An
al
y
s
is
–
s
y
n
th
esis
o
f
a
n
o
is
y
s
p
ee
ch
s
ig
n
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
9
,
No
.
2,
Au
g
u
s
t
20
20
:
63
–
7
2
66
Prio
r
to
th
e
s
p
ee
c
h
e
n
h
an
ce
m
en
t
tech
n
i
q
u
e,
th
e
n
o
is
e
c
h
ar
ac
ter
is
tics
ca
n
b
e
id
en
tifi
ed
b
y
th
e
s
tatis
t
ical
m
o
d
el.
T
h
e
g
ain
v
al
u
es a
r
e
ca
lcu
lated
b
ased
o
n
th
e
f
ea
tu
r
es e
x
tr
ac
ted
f
r
o
m
t
h
e
o
r
ig
in
al
in
p
u
t so
t
h
at
an
u
p
g
r
ad
ed
s
p
ee
ch
s
ig
n
al
is
ac
q
u
ir
ed
.
T
h
e
tar
g
e
t
ev
alu
at
io
n
ca
n
b
e
m
ad
e
th
r
o
u
g
h
th
e
p
o
wer
s
p
ec
tr
u
m
d
en
s
ity
o
f
th
e
n
o
is
y
s
p
ee
ch
s
eg
m
en
ts
an
d
n
o
is
e.
T
h
is
h
elp
s
t
o
g
en
er
ate
t
h
e
g
ain
c
o
ef
f
icien
t
s
o
f
th
e
s
ig
n
als.
W
ien
er
f
ilter
in
g
h
elp
s
to
r
ed
u
ce
th
e
Me
an
Sq
u
ar
e
E
r
r
o
r
(
MSE
)
am
o
n
g
th
e
ac
tu
al
s
ig
n
al
an
d
th
e
e
s
tim
ated
s
ig
n
al
an
d
to
b
o
o
s
t
u
p
th
e
o
r
ig
in
al
s
ig
n
al
f
r
o
m
th
e
n
o
is
y
s
ig
n
al.
Similar
l
y
,
th
e
n
o
is
e
ca
n
b
e
m
in
im
ized
th
r
o
u
g
h
h
ar
d
o
r
s
o
f
t
th
r
esh
o
ld
in
g
.
So
m
etim
es
t
h
e
im
p
o
r
tan
t
c
o
m
p
o
n
en
ts
o
f
th
e
o
r
ig
i
n
al
s
p
ee
ch
m
ig
h
t b
e
lo
s
t d
u
e
to
o
v
er
th
r
esh
o
ld
in
g
.
3
.
1
.
M
o
dified
wiener
f
ilte
rin
g
t
ec
hn
iqu
e
T
h
e
ad
v
an
ta
g
e
o
f
th
e
m
o
d
if
ie
d
ad
ap
tiv
e
wien
er
f
ilter
is
th
at
th
e
s
p
ee
ch
s
ig
n
al
is
p
r
o
ce
s
s
ed
th
r
o
u
g
h
f
ilter
b
y
v
ar
y
in
g
lo
ca
l
s
tatis
tics
s
u
ch
as
m
ea
n
,
v
a
r
ian
ce
.
Her
e,
th
e
m
ea
n
v
alu
e
o
f
th
e
a
d
d
itiv
e
n
o
is
e
d
(
n
)
is
co
n
s
id
er
ed
to
b
e
ze
r
o
an
d
h
o
ld
s
a
wh
ite
n
atu
r
e
with
v
ar
ia
n
ce
σ
d
2
.
T
h
u
s
,
th
e
p
o
wer
s
p
e
ctr
u
m
P
d
(
ω
)
ca
n
b
e
ex
p
r
ess
ed
in
(
2
)
,
(
)
=
2
(
2
)
Her
e
th
e
s
eg
m
en
ted
s
p
ee
c
h
s
ig
n
al
x
j
(
n
)
is
tr
ea
ted
as st
atio
n
ar
y
an
d
th
u
s
ca
n
b
e
m
o
d
eled
b
y
(
3
)
,
(
)
=
+
(
)
(
3
)
wh
er
e
m
xj
is
th
e
lo
ca
l
m
ea
n
an
d
σ
xj
is
th
e
s
tan
d
ar
d
d
ev
iatio
n
o
f
x
j
(
n
)
.
w
(
n
)
is
a
u
n
it
v
ar
ian
c
e
n
o
is
e
with
a
ze
r
o
m
ea
n
.
Fo
r
co
n
v
e
n
ien
ce
x
j
is
r
ep
r
esen
ted
as
x
,
th
e
m
ea
n
o
f
th
e
o
r
ig
in
al
s
ig
n
al
m
s
is
eq
u
al
to
av
er
ag
e
m
ea
n
v
alu
e
o
f
all
th
e
j
f
r
am
es,
m
x
.
T
h
e
W
ien
er
f
ilter
tr
an
s
f
er
f
u
n
cti
o
n
ca
n
b
e
esti
m
ated
b
y
(
4
)
,
(
)
=
(
)
(
)
+
(
)
=
2
2
+
2
(
4
)
T
h
e
im
p
u
ls
e
r
esp
o
n
s
e
o
f
th
e
wien
er
f
ilter
ca
n
b
e
ac
h
iev
ed
b
y
ap
p
ly
in
g
I
n
v
e
r
s
e
T
r
an
s
f
o
r
m
atio
n
to
H(
ω
)
an
d
is
g
i
v
en
in
(
5
)
,
ℎ
(
)
=
2
2
+
2
(
)
(
5
)
As
th
e
m
x
an
d
σ
x
ar
e
u
p
d
ated
at
ea
ch
s
am
p
le,
th
e
s
p
ee
ch
s
ig
n
al
ca
n
b
e
esti
m
ated
f
r
o
m
(
6
)
,
an
d
is
d
en
o
ted
as
̂
(
n
)
.
^
(
)
=
(
)
+
2
(
)
2
(
)
+
2
(
(
)
−
(
)
)
(
6
)
No
w
esti
m
atin
g
an
d
tr
ac
k
in
g
th
e
n
o
is
e
f
r
am
es
in
th
e
n
o
is
y
s
ig
n
al
is
th
e
cr
u
cial
task
in
th
e
p
r
o
ce
s
s
o
f
s
p
ee
ch
en
h
an
ce
m
e
n
t.
Usu
ally
,
th
e
v
o
ice
ac
tiv
ity
d
etec
to
r
is
em
p
lo
y
ed
f
o
r
f
in
d
in
g
th
e
p
r
esen
ce
o
f
n
o
is
e.
B
u
t
in
th
is
W
ein
er
f
ilter
in
g
m
eth
o
d
,
f
r
am
e
e
n
er
g
y
is
s
y
n
ch
r
o
n
ized
with
th
e
m
in
im
u
m
f
r
am
e
en
er
g
y
.
Min
im
u
m
en
er
g
y
v
ar
iatio
n
is
d
ir
ec
tly
p
r
o
p
o
r
tio
n
al
to
th
e
s
ig
n
al
co
n
d
it
io
n
s
.
T
h
er
ef
o
r
e,
b
y
u
s
in
g
s
m
o
o
th
ed
3
2
p
o
in
ts
o
f
th
e
s
p
ec
tr
u
m
t
h
e
s
p
ec
tr
al
d
e
v
ian
ce
is
ca
lcu
lated
.
T
o
g
ain
n
o
is
e
s
p
ec
tr
u
m
th
ese
p
o
in
ts
ar
e
v
er
if
ied
an
d
u
p
g
r
a
d
ed
a
d
d
itio
n
ally
.
T
h
en
t
h
e
lev
el
o
f
en
er
g
y
is
f
ix
e
d
to
1
0
d
B
.
So
b
y
an
aly
zi
n
g
th
e
en
er
g
y
lev
el
w
h
ich
is
m
o
r
e
th
an
1
0
d
B
an
d
R
MS
lev
el
m
o
r
e
th
an
8
d
B
,
th
e
p
r
ese
n
ce
o
f
s
p
ee
ch
ca
n
b
e
id
en
tifie
d
.
W
h
en
th
er
e
is
n
o
av
ailab
ilit
y
o
f
s
p
ee
ch
th
is
n
o
is
e
s
p
ec
tr
u
m
N
(
ω
,
m)
ca
n
b
e
m
e
asu
r
ed
f
o
r
v
ar
ied
tim
e
s
am
p
les
.
T
h
e
u
ltima
te
p
lan
is
to
im
p
lem
en
t
MM
SE
b
ase
d
n
o
is
e
p
o
wer
esti
m
atio
n
tec
h
n
iq
u
e.
I
n
th
is
p
r
o
ce
s
s
,
s
p
ee
c
h
&
n
o
is
e
s
p
ec
tr
al
co
ef
f
icien
ts
h
av
e
c
o
m
p
lex
Gau
s
s
ian
d
is
tr
ib
u
tio
n
wh
ich
ca
n
b
e
ex
p
r
ess
ed
in
(
7
)
.
(
)
=
1
2
(
−
|
|
2
2
)
(
7
)
Hen
ce
,
th
e
n
o
is
y
p
o
wer
s
p
ec
tr
al
co
ef
f
icien
ts
ca
n
b
e
ex
p
r
ess
e
d
as g
iv
en
in
(
8
)
.
(
)
=
1
2
(
1
+
)
(
−
|
|
2
2
(
1
+
)
)
(
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
S
p
ee
ch
e
n
h
a
n
ce
men
t u
s
in
g
mo
d
ified
w
ien
er filter
b
a
s
ed
MM
S
E
a
n
d
s
p
ee
ch
…
(
V
.
V
ija
ya
s
r
i B
o
lis
etty
)
67
3
.
2
.
No
is
e
P
SD e
s
t
im
a
t
i
o
n a
nd
t
ra
ck
ing
T
h
e
n
o
is
e
is
r
ep
r
esen
te
d
as
=
an
d
=
.
Sp
ec
tr
al
co
ef
f
icien
ts
ca
n
b
e
tr
an
s
f
o
r
m
e
d
in
to
p
o
lar
co
o
r
d
in
ates u
s
in
g
(
9
)
.
,
(
,
)
=
1
2
(
−
2
2
)
(
9
)
Pro
b
ab
ilit
y
d
is
tr
ib
u
tio
n
|
,
(
|
(
,
)
)
o
f
s
ig
n
al
is
g
iv
en
in
(
1
0
)
.
|
,
(
|
,
)
=
1
2
(
2
(
−
)
−
2
−
2
2
)
(
1
0
)
I
n
th
e
MM
SE
m
et
h
o
d
th
er
e
ar
e
s
ev
er
al
n
o
is
e
p
o
wer
esti
m
ato
r
s
p
r
esen
t
in
th
e
n
o
is
y
s
ig
n
al
p
er
io
d
o
g
r
am
esti
m
atio
n
.
L
et
u
s
co
n
s
id
er
a
p
r
i
o
r
i
SNR
ξ
an
d
esti
m
ated
n
o
is
e
p
o
we
r
is
2
.
T
h
e
n
o
is
e
p
er
io
d
o
g
r
am
ca
n
b
e
o
b
tain
ed
in
(
1
1
)
.
|
̂
|
2
=
(
|
|
2
)
=
(
1
1
+
̂
)
|
|
2
+
̂
1
+
̂
̂
2
(
1
1
)
Fro
m
th
e
(
11
)
it
is
id
en
tifie
d
th
at
th
e
n
o
is
e
p
er
i
o
d
o
g
r
am
ca
n
b
e
g
ain
ed
.
B
u
t
h
er
e
th
e
s
ig
n
al
g
ets
alter
ed
f
r
o
m
tim
e
to
ti
m
e,
th
e
r
ef
o
r
e
t
h
e
s
p
ec
tr
al
d
e
n
s
ity
h
as
to
b
e
r
ev
is
ed
f
r
e
q
u
en
tly
.
T
o
u
p
d
ate
th
e
s
p
ec
tr
al
d
en
s
ity
p
ar
am
eter
s
th
e
r
ec
u
r
s
i
v
e
s
m
o
o
th
in
g
is
ap
p
lied
as sh
o
wn
in
(
1
2
)
.
̂
2
(
)
=
̂
2
(
−
1
)
+
(
1
−
)
|
(
)
|
2
(
1
2
)
W
h
er
e
α
=
0
.
0
.
W
h
en
e
v
er
ξ
<
1
,
th
e
n
o
is
e
p
er
i
o
d
o
g
r
am
is
u
p
d
ated
.
I
n
th
e
s
am
e
way
b
y
v
er
if
y
in
g
a
p
r
io
r
i
SNR
f
ac
to
r
s
th
e
s
p
ec
tr
al
n
o
is
e
p
o
wer
is
im
p
r
o
v
ed
.
Hen
ce
th
e
s
u
m
o
f
t
h
e
o
b
s
er
v
ed
n
o
is
y
s
ig
n
al
an
d
p
r
ev
io
u
s
esti
m
atio
n
o
f
s
p
ec
tr
al
n
o
is
e
p
o
wer
̂
2
is
r
ep
r
esen
ted
b
y
MM
SE.
Fro
m
th
is
esti
m
atio
n
th
e
p
r
io
r
i
SNR
v
alu
e
is
o
b
tain
ed
wh
ich
i
s
in
b
etwe
en
0
an
d
1
.
I
n
A
p
r
i
o
r
i SNR
ca
n
b
e
o
b
tain
e
d
b
y
f
o
llo
win
g
(
1
3
)
.
̂
=
(
0
,
̂
)
=
(
0
,
̂
−
1
)
(
1
3
)
wh
er
e
ml
d
en
o
tes
m
a
x
im
u
m
lik
elih
o
o
d
co
m
p
u
tatio
n
an
d
̂
(
)
=
|
(
)
|
2
2
(
−
1
)
.
T
h
e
(
1
3
)
is
u
n
b
ias
in
n
atu
r
e.
He
n
ce
,
a
p
r
io
r
i
SNR
in
f
o
r
m
atio
n
is
n
o
t
k
n
o
wn
.
So
,
th
e
esti
m
atio
n
o
f
th
is
u
n
k
n
o
wn
SNR
ca
n
g
en
e
r
ate
ab
ias
f
ac
to
r
as
ex
p
r
ess
ed
in
(
1
4
)
.
−
1
(
)
=
(
(
1
+
)
(
2
,
1
1
+
)
+
−
1
1
+
)
(
1
4
)
wh
er
e
r
ep
r
esen
ts
in
co
m
p
lete
g
am
m
a
f
u
n
ctio
n
.
Acc
o
r
d
in
g
t
o
th
e
n
atu
r
e
o
f
a
p
r
io
r
i
SNR
,
esti
m
ato
r
E
(
|d
|
2
x)
v
alu
e
is
u
n
b
iased
wh
e
r
ea
s
f
o
r
lo
w
SNR
v
alu
es it is u
n
d
er
-
b
i
ased
.
3
.
3
.
So
f
t
–
decisi
o
n t
ec
hn
iqu
e
f
o
r
no
is
e
presence
estim
a
t
io
n
Sp
ee
ch
p
r
esen
ce
o
r
ab
s
en
ce
is
to
b
e
id
e
n
tifie
d
,
to
id
en
tify
th
e
n
o
is
e
ch
ar
ac
ter
is
tics
.
Gen
er
a
lly
,
n
o
is
e
ch
ar
ac
ter
is
tics
ca
n
b
e
o
b
tain
ed
wh
en
th
e
s
p
ee
ch
is
ab
s
en
t.
T
h
e
s
p
ee
ch
o
r
s
ilen
ce
p
ar
t
class
if
icatio
n
ca
n
b
e
d
o
n
e
b
y
(
1
5
)
.
|
̂
(
)
|
2
=
(
|
̂
(
)
|
2
(
)
)
=
{
̂
2
(
−
1
)
,
̂
(
)
≥
1
|
(
)
|
2
,
̂
(
)
<
1
(
1
5
)
Ho
wev
er
,
p
r
o
p
o
s
ed
m
o
d
el
u
tili
ze
s
s
o
f
t
d
ec
is
io
n
-
b
ased
f
r
am
ewo
r
k
f
o
r
n
o
is
e
esti
m
ati
o
n
in
th
e
p
r
esen
ce
o
r
ab
s
en
ce
o
f
s
p
ee
ch
.
T
h
e
p
r
o
b
ab
ilit
y
o
f
s
p
ee
c
h
p
r
esen
ce
o
r
ab
s
en
ce
ca
n
b
e
e
s
tim
ated
f
r
o
m
(
1
6
)
an
d
(
1
7
)
.
(
1
|
)
|
|
(
1
|
)
>
>
|
(
0
|
)
=
1
(
1
6
)
(
1
|
)
|
|
(
1
|
)
<
<
|
(
0
|
)
=
0
(
1
7
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
9
,
No
.
2,
Au
g
u
s
t
20
20
:
63
–
7
2
68
wh
er
e,
|
0
(
)
=
1
(
)
−
1
(
−
)
(
1
8
)
|
1
(
)
=
1
(
)
(
1
+
1
)
−
1
(
−
1
+
1
)
(
1
9
)
T
h
e
(
1
8
)
an
d
(
1
9
)
d
escr
ib
es
s
p
ee
ch
p
r
esen
ce
an
d
s
p
ee
ch
a
b
s
en
ce
r
esp
ec
tiv
ely
.
No
is
e
p
er
io
d
o
g
r
am
u
n
d
er
s
p
ee
ch
p
r
esen
ce
co
n
d
itio
n
s
ca
n
b
e
ex
p
r
ess
ed
in
(
2
0
)
.
Fo
r
f
u
r
th
er
im
p
r
o
v
em
en
t
ch
i
-
s
q
u
ar
e
d
is
tr
ib
u
tio
n
-
b
ased
h
y
p
o
th
esis
an
aly
s
is
ca
n
b
e
ap
p
lied
.
(
|
̂
(
)
|
2
|
)
=
(
0
|
)
(
|
̂
(
)
|
2
|
,
0
)
+
(
1
|
)
(
|
|
2
|
,
1
)
(2
0)
T
h
er
e
ar
e
two
im
p
o
r
ta
n
t
co
m
p
o
n
e
n
ts
to
b
e
co
n
s
id
er
ed
in
esti
m
atin
g
s
p
ee
ch
p
r
esen
ce
p
r
o
b
ab
ilit
y
u
s
in
g
ch
i
-
s
q
u
ar
e
a
p
p
r
o
x
im
ati
o
n
:
o
b
s
er
v
atio
n
s
eq
u
en
ce
an
d
esti
m
ated
n
o
is
e
v
ar
ian
ce
as
s
h
o
wn
in
(
2
2
)
.
Sp
ec
tr
al
co
m
p
o
n
en
t
co
m
p
u
tat
io
n
is
ap
p
lied
r
esu
ltin
g
i
n
o
b
s
er
v
atio
n
s
eq
u
e
n
ce
a
n
d
esti
m
ated
n
o
is
e
v
ar
ian
ce
g
en
er
atio
n
w
h
er
e
to
tal
N
n
u
m
b
er
o
f
f
r
eq
u
en
cy
b
in
s
is
co
n
s
id
er
ed
g
i
v
en
in
(
2
1
)
.
=
[
1
,
2
,
.
.
.
,
,
.
.
.
,
]
(
2
1
)
=
[
1
,
2
,
.
.
.
,
,
.
.
.
,
]
I
n
n
ex
t
p
h
ase,
c
h
i
-
s
q
u
ar
e
ca
lc
u
latio
n
is
ap
p
lied
f
o
r
t
h
ese
f
r
e
q
u
en
cy
b
in
s
an
d
th
e
c
h
i
-
s
q
u
ar
e
s
tatis
tic
is
g
iv
en
b
y
(
2
2
)
.
2
=
∑
(
−
)
2
=
1
(
2
2
)
T
h
r
o
u
g
h
th
e
c
h
i
-
s
q
u
ar
e
ca
lcu
latio
n
th
e
o
b
tain
ed
v
alu
e
is
co
m
p
ar
ed
with
th
e
th
r
esh
o
ld
p
ar
am
eter
wh
er
e
(
N
-
1
)
to
tal
d
eg
r
ee
s
o
f
f
r
eq
u
en
cy
b
in
s
ar
e
a
v
ailab
le
.
Co
nd
it
io
n:
If
(
N
s
2
>Th
)
//
th
r
esh
o
ld
v
alu
e
i
s
d
en
o
ted
b
y
Th
I
(
λ,
k)
=1
// a
cc
ep
t c
o
n
d
itio
n
H
1
E
ls
e
I
(
λ,
k)
=0
// a
cc
ep
t c
o
n
d
itio
n
H
0
E
n
d
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
n
o
is
e
ch
ar
ac
ter
is
tics
ar
e
esti
m
ated
f
r
o
m
th
e
n
o
is
y
s
p
ee
ch
s
ig
n
al
d
u
r
in
g
th
e
ab
s
en
ce
o
f
s
p
ee
ch
an
d
th
e
s
p
ee
ch
clea
r
n
ess
is
d
eter
m
in
ed
f
r
o
m
Per
f
o
r
m
a
n
ce
E
v
alu
atio
n
o
f
Sp
ee
ch
Qu
ality
(
PESQ)
an
d
s
eg
m
en
ted
SNR
(
s
eg
SNR
)
.
I
n
th
e
asp
ec
t
o
f
m
o
b
ile
a
p
p
licatio
n
,
a
s
am
p
le
-
r
ate
o
f
1
6
KHz
,
8
KHz
T
I
MI
T
d
at
a
b
ase
s
ig
n
als
ar
e
u
s
ed
as
in
p
u
t
to
m
e
asu
r
e
th
e
q
u
ality
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
.
I
n
itially
,
th
e
s
ig
n
als
ar
e
T
F
-
d
ec
o
m
p
o
s
ed
an
d
Han
n
-
win
d
o
wed
f
r
am
es
with
a
len
g
th
o
f
2
5
6
s
am
p
les.
B
u
t
to
ca
lcu
late
th
e
o
b
jectiv
e
in
tellig
ib
ilit
y
,
ea
ch
f
r
am
e
is
ze
r
o
-
p
a
d
d
ed
u
p
to
5
1
2
s
am
p
les.
As
th
e
PESQ
is
h
ig
h
ly
co
r
r
elate
d
with
th
e
s
u
b
jectiv
e
m
ea
s
u
r
es,
it
is
r
ec
o
m
m
en
d
e
d
b
y
I
T
U
-
T
f
o
r
m
ea
s
u
r
in
g
th
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
p
r
o
ce
s
s
in
g
tech
n
iq
u
e.
T
h
e
s
im
u
latio
n
h
as b
ee
n
d
o
n
e
in
MA
T
L
AB
2
0
1
3
b
to
o
l.
T
h
e
n
o
is
y
s
ig
n
als
s
u
ch
as
b
ab
b
le,
air
p
o
r
t,
ca
r
,
tr
ain
n
o
is
e
ar
e
ch
o
s
en
at
d
if
f
e
r
en
t
SNR
0
d
B
,
5
d
B
,
1
0
d
B
with
3
0
s
am
p
les
o
f
m
ale
an
d
f
em
ale
s
p
ea
k
er
s
f
r
o
m
NOI
Z
E
US
d
atab
ase
ar
e
ch
o
s
en
f
o
r
p
er
f
o
r
m
a
n
ce
ev
alu
atio
n
o
f
t
h
e
p
r
o
p
o
s
ed
t
ec
h
n
iq
u
e.
B
y
u
s
in
g
t
h
e
o
v
e
r
lap
-
an
d
-
ad
d
p
r
o
ce
d
u
r
e
th
e
s
ig
n
al
is
s
y
n
th
esized
.
Gen
er
al
s
im
u
latio
n
p
a
r
am
eter
s
lik
e
d
o
wn
-
s
am
p
lin
g
,
Nu
m
b
er
o
f
FF
T
p
o
in
ts
,
win
d
o
w
le
n
g
th
a
n
d
win
d
o
w
o
v
er
lap
a
r
e
d
escr
ib
ed
in
T
ab
le
1
.
T
ab
le
1
.
s
im
u
latio
n
p
ar
am
eter
s
co
n
s
id
er
ed
S
i
mu
l
a
t
i
o
n
P
a
r
me
t
e
r
N
a
me
S
i
mu
l
a
t
i
o
n
P
a
r
a
m
e
t
e
r
V
l
a
u
e
S
i
g
n
a
l
D
o
w
n
sa
mp
l
i
n
g
8
k
H
z
N
F
F
T
2
^
1
4
W
i
n
d
o
w
Le
n
g
t
h
(
ms)
3
2
ms
W
i
n
d
o
w
o
v
e
r
l
a
p
1
6
ms
N
u
mb
e
r
o
f
a
v
e
r
a
g
i
n
g
w
i
n
d
o
w
10
P
S
D
(
P
o
w
e
r
S
p
e
c
t
r
a
l
d
e
n
s
i
t
y
)
e
st
i
ma
t
i
o
n
p
a
r
a
met
e
r
0
.
8
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
S
p
ee
ch
e
n
h
a
n
ce
men
t u
s
in
g
mo
d
ified
w
ien
er filter
b
a
s
ed
MM
S
E
a
n
d
s
p
ee
ch
…
(
V
.
V
ija
ya
s
r
i B
o
lis
etty
)
69
Fig
u
r
e
3
.
Or
ig
i
n
al
s
p
ee
ch
s
ig
n
al
Fig
u
r
e
4
.
No
is
y
s
ig
n
al
Fig
u
r
e
5
.
Po
wer
s
p
ec
tr
a
Fig
u
r
e
6
.
Fil
ter
ed
Sig
n
al
a
n
d
n
o
is
e
s
u
p
p
r
ess
ed
s
ig
n
al
T
ab
le
2
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
f
o
r
b
ab
b
le
n
o
is
e
I
n
p
u
t
S
N
R
P
a
r
a
me
t
e
r
N
a
me
OM
-
LSA
[
1
3
]
S
M
P
O
[
1
1
]
M
M
S
E
-
SPP
[
1
0
]
B
W
T
[
1
6
]
W
a
v
e
l
e
t
P
a
c
k
e
t
[
1
1
]
P
r
o
p
o
se
d
0
d
B
P
ESQ
2
.
0
6
2
.
0
1
2
.
0
8
2
.
2
5
2
.
1
1
2
.
3
7
seg
S
N
R
3
.
6
2
4
.
0
0
2
.
2
2
0
.
9
6
5
.
2
5
5
.
5
1
5
d
B
P
ESQ
2
.
4
2
2
.
3
8
2
.
4
7
2
.
2
9
2
.
4
4
2
.
5
2
seg
S
N
R
3
.
5
9
3
.
6
2
1
.
5
0
2
.
5
5
4
.
6
5
4
.
7
3
1
0
d
B
P
ESQ
2
.
8
6
2
.
8
3
2
.
8
7
2
.
5
0
2
.
8
7
2
.
9
9
seg
S
N
R
3
.
1
0
4
.
1
1
0
.
7
6
3
.
2
7
4
.
2
3
4
.
7
T
h
e
o
r
ig
in
al
s
p
ee
ch
,
n
o
is
y
s
ig
n
als
ar
e
co
n
s
id
er
ed
f
o
r
s
im
u
la
tio
n
in
.
wav
f
o
r
m
an
d
is
r
ep
r
e
s
en
ted
in
Fig
u
r
e
3
an
d
Fig
u
r
e
4
r
esp
ec
tiv
ely
.
T
h
r
o
u
g
h
th
ese
d
iag
r
a
m
m
atica
l
p
r
esen
tatio
n
s
th
e
p
r
esen
ce
an
d
ab
s
en
ce
ca
n
b
e
id
en
tifie
d
.
T
h
e
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
u
tili
ze
s
th
e
p
h
ase
a
n
aly
s
is
m
eth
o
d
an
d
Fig
u
r
e
5
.
s
h
o
ws
th
e
Po
wer
s
p
ec
tr
a
o
f
o
r
ig
i
n
al,
n
o
i
s
y
s
ig
n
als.
Fin
ally
,
f
ilter
ed
an
d
n
o
is
e
s
u
p
p
r
ess
ed
s
ig
n
al
an
a
ly
s
is
is
r
ep
r
esen
ted
in
Fig
u
r
e
6
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
co
m
p
ar
ed
with
th
e
o
th
e
r
ex
is
tin
g
s
tate
o
f
ar
t
(
Pen
g
f
ei
Su
n
,
J
u
n
Qin
[
1
1
]
)
.
PESQ
an
d
s
eg
SN
R
ar
e
th
e
c
h
ar
ac
ter
is
tics
f
ea
tu
r
es
wh
ich
ar
e
u
s
ed
to
ev
alu
ate
th
e
p
r
o
c
ess
in
g
ap
p
r
o
ac
h
.
B
y
v
ar
y
in
g
th
e
n
o
is
y
i
n
p
u
t
SNR
v
alu
es
f
r
o
m
0
d
B
to
1
0
d
B
th
e
av
er
ag
e
v
alu
es
o
f
PESQ
an
d
s
eg
SNR
v
alu
es
ar
e
r
ep
r
esen
ted
in
th
e
tab
le
f
o
r
all
s
am
p
les.
T
h
e
No
izeu
s
d
atab
ase
with
3
0
s
p
ea
k
er
s
in
ea
ch
s
et
co
n
tain
s
b
ab
b
le
n
o
is
e,
air
p
o
r
t
n
o
is
e,
ca
r
n
o
is
e
an
d
tr
ain
n
o
is
e
wh
ich
is
also
t
r
ea
ted
as
im
p
o
r
ta
n
t
ch
ar
ac
ter
i
s
tics
to
id
en
tify
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
eth
o
d
.
T
h
e
T
ab
le
2
,
T
a
b
le
3
,
T
a
b
le
4
,
an
d
T
ab
le
5
d
escr
ib
e
th
e
p
er
f
o
r
m
an
ce
an
aly
s
is
f
o
r
b
ab
b
le,
air
p
o
r
t,
ca
r
an
d
tr
ain
n
o
is
es r
esp
ec
tiv
ely
.
T
h
e
ca
lcu
lated
PESQ
an
d
s
e
g
m
en
tal
SNR
ar
e
r
elate
d
with
b
en
c
h
m
ar
k
alg
o
r
ith
m
s
to
e
s
t
im
ate
th
e
ef
f
icien
cy
o
f
th
e
s
y
s
tem
.
I
n
ter
m
s
o
f
b
ab
b
le
a
n
d
tr
ain
n
o
is
es
o
f
0
d
B
,
5
d
B
an
d
1
0
d
B
SNR
s
ig
n
als
o
f
ex
is
tin
g
s
tate
-
of
-
ar
t
m
eth
o
d
th
e
p
r
o
p
o
s
ed
m
eth
o
d
ac
h
iev
es
e
n
h
an
ce
d
r
esu
lts
b
y
m
ea
n
s
o
f
b
ab
b
le
an
d
tr
ain
n
o
is
es
0
d
B
,
5
d
B
an
d
1
0
d
B
SNR
s
ig
n
als.
Als
o
,
b
etter
r
esu
lts
ar
e
o
b
tain
ed
in
p
r
esen
ce
o
f
air
p
o
r
t
an
d
ca
r
n
o
is
es
f
o
r
all
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
9
,
No
.
2,
Au
g
u
s
t
20
20
:
63
–
7
2
70
s
ig
n
als
wh
ich
ar
e
s
h
o
wn
in
T
a
b
le
3
an
d
4
.
Fro
m
th
e
r
esu
lts
i
t
is
clea
r
ly
i
d
en
tifie
d
th
at
n
o
is
e
d
etec
tio
n
is
m
o
r
e
ef
f
icien
t
th
an
t
h
e
o
th
e
r
ex
is
tin
g
ap
p
r
o
ac
h
es.
T
h
e
s
tatio
n
ar
y
n
o
is
e
is
r
ec
o
g
n
ized
ea
s
ily
wh
er
ea
s
in
s
o
m
e
ca
s
e
th
e
n
o
n
-
s
tatio
n
ar
y
n
o
is
e
lik
e
b
ab
b
le
n
o
is
e
an
d
tr
ain
n
o
is
e
tr
ac
k
ed
with
t
h
e
h
elp
o
f
s
o
f
t
d
ec
is
io
n
b
ased
esti
m
atio
n
.
T
ab
le
3
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
f
o
r
air
p
o
r
t n
o
is
e
I
n
p
u
t
S
N
R
P
a
r
a
me
t
e
r
N
a
me
OM
-
LSA
[
1
3
]
S
M
P
O
[
1
1
]
M
M
S
E
-
SPP
[
1
0
]
B
W
T
[
1
6
]
W
a
v
e
l
e
t
P
a
c
k
e
t
[
1
1
]
P
r
o
p
o
se
d
0
d
B
P
ESQ
2
.
1
3
1
.
9
8
2
.
1
3
2
.
1
0
2
.
2
2
2
.
3
5
seg
S
N
R
3
.
3
4
4
.
7
1
4
.
4
5
1
.
0
7
5
.
1
5
5
.
3
6
5
d
B
P
ESQ
2
.
7
8
2
.
5
8
2
.
7
2
2
.
4
1
2
.
7
2
2
.
7
seg
S
N
R
2
.
1
0
4
.
1
2
3
.
6
7
2
.
7
1
4
.
9
1
4
.
8
6
1
0
d
B
P
ESQ
2
.
8
8
2
.
8
0
2
.
7
9
2
.
6
5
2
.
8
6
2
.
9
8
seg
S
N
R
1
.
1
0
3
.
7
1
2
.
9
6
2
.
8
2
4
.
3
0
4
.
3
6
T
ab
le
4
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
f
o
r
ca
r
n
o
is
e
I
n
p
u
t
S
N
R
P
a
r
a
me
t
e
r
N
a
me
OM
-
LSA
[
1
3
]
S
M
P
O
[
1
1
]
M
M
S
E
-
SPP
[
1
0
]
B
W
T
[
1
6
]
W
a
v
e
l
e
t
P
a
c
k
e
t
[
11
]
P
r
o
p
o
se
d
0
d
B
P
ESQ
2
.
3
9
2
.
1
9
2
.
2
5
2
.
2
2
2
.
2
7
2
.
3
6
seg
S
N
R
4
.
7
8
5
.
4
3
5
.
5
2
.
8
6
6
.
0
3
5
.
5
5
d
B
P
ESQ
2
.
5
9
2
.
5
4
2
.
6
4
2
.
2
6
2
.
7
1
2
.
6
9
seg
S
N
R
3
.
4
8
5
.
1
2
4
.
9
6
3
.
6
9
5
.
5
2
4
.
8
9
1
0
d
B
P
ESQ
2
.
9
0
2
.
8
9
2
.
9
2
2
.
7
0
2
.
8
8
2
.
9
5
seg
S
N
R
1
.
4
5
4
.
1
2
4
.
6
4
4
.
7
8
4
.
9
0
4
.
7
7
T
ab
le
5
.
C
o
m
p
a
r
ativ
e
p
er
f
o
r
m
an
ce
f
o
r
tr
ain
n
o
is
e
I
n
p
u
t
S
N
R
P
a
r
a
me
t
e
r
N
a
me
OM
-
LSA
[
1
3
]
S
M
P
O
[
11
]
M
M
S
E
-
SPP
[
1
0
]
B
W
T
[
1
6
]
W
a
v
e
l
e
t
P
a
c
k
e
t
[
11
]
P
r
o
p
o
se
d
0
d
B
P
ESQ
2
.
1
5
2
.
1
7
2
.
0
7
2
.
0
8
2
.
2
6
2
.
2
9
seg
S
N
R
4
.
8
5
3
.
8
4
4
.
7
7
3
.
4
5
4
.
9
0
5
.
5
0
5
d
B
P
ESQ
2
.
5
1
2
.
2
2
2
.
4
8
2
.
2
8
2
.
5
3
2
.
5
9
seg
S
N
R
4
.
5
2
.
4
3
4
.
2
8
3
.
8
9
4
.
0
6
4
.
7
9
1
0
d
B
P
ESQ
2
.
7
5
2
.
6
7
2
.
7
6
2
.
5
9
2
.
8
9
2
.
9
4
seg
S
N
R
4
.
0
6
1
.
0
7
3
.
7
1
3
.
5
3
.
6
4
4
.
4
7
T
h
e
Fig
u
r
e
7
,
Fig
u
r
e
8
,
Fig
u
r
e
9
,
Fig
u
r
e
1
0
s
h
o
ws
th
at
co
m
p
ar
ativ
e
an
aly
s
is
r
ep
r
esen
tatio
n
o
f
p
r
o
p
o
s
ed
ap
p
r
o
a
ch
in
ter
m
s
o
f
PESQ
f
o
r
b
ab
b
le,
air
p
o
r
t,
ca
r
an
d
tr
ain
n
o
is
es
f
o
r
0
d
B
,
5
d
B
an
d
1
0
d
B
s
ig
n
als,
co
n
s
id
er
in
g
t
h
e
f
iv
e
s
tates
o
f
ar
t
tech
n
iq
u
es.
T
h
e
P
E
SQ
m
ea
s
u
r
em
en
t
an
d
im
p
r
o
v
em
en
t
in
t
h
e
T
ab
le
2
,
3
,
4
an
d
5
s
h
o
ws
th
at
p
r
o
p
o
s
ed
m
eth
o
d
d
ev
elo
p
s
th
e
q
u
ality
o
f
th
e
en
h
an
ce
d
s
ig
n
al
at
a
r
ate
o
f
7
.
7
6
%
av
er
ag
e
im
p
r
o
v
em
e
n
t f
o
r
0
d
B
,
2
.
6
5
% im
p
r
o
v
em
e
n
t f
o
r
5
d
B
an
d
f
in
ally
3
.
7
8
% im
p
r
o
v
e
m
en
t is m
ar
k
ed
at
1
0
d
B
SNR
s
ig
n
als.
Fig
u
r
e
7
.
PESQ
p
er
f
o
r
m
a
n
ce
an
aly
s
is
f
o
r
b
ab
b
le
n
o
is
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
S
p
ee
ch
e
n
h
a
n
ce
men
t u
s
in
g
mo
d
ified
w
ien
er filter
b
a
s
ed
MM
S
E
a
n
d
s
p
ee
ch
…
(
V
.
V
ija
ya
s
r
i B
o
lis
etty
)
71
Fig
u
r
e
8
.
PESQ
p
er
f
o
r
m
a
n
ce
an
aly
s
is
f
o
r
air
p
o
r
t n
o
is
e
Fig
u
r
e
9
.
PESQ
p
er
f
o
r
m
a
n
ce
an
aly
s
is
f
o
r
ca
r
n
o
is
e
Fig
u
r
e
1
0
.
PESQ
p
er
f
o
r
m
an
ce
an
aly
s
is
f
o
r
tr
ain
n
o
is
e
5.
CO
NCLU
SI
O
N
T
h
e
o
b
jectiv
e
in
tellig
ib
ilit
y
m
ea
s
u
r
ed
f
o
r
v
ar
i
o
u
s
v
o
iced
an
d
u
n
v
o
ice
d
s
ig
n
als
is
f
o
u
n
d
to
b
e
b
etter
in
th
e
p
r
o
p
o
s
ed
m
eth
o
d
.
W
ien
er
f
ilter
in
g
is
im
p
lem
en
ted
alo
n
g
with
s
p
ec
tr
al
co
ef
f
icien
t,
an
d
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
m
o
d
els,
to
im
p
r
o
v
e
th
e
p
er
f
o
r
m
an
ce
o
f
s
p
ee
ch
q
u
ality
in
ter
m
s
o
f
PESQ
an
d
s
eg
SNR
.
W
h
ile
ex
tr
ac
tin
g
th
e
n
o
is
e
ch
ar
ac
ter
is
tics
p
h
ase
co
ef
f
icien
ts
ar
e
p
r
eser
v
ed
an
d
r
eu
s
ed
af
ter
t
h
e
r
ec
o
n
s
tr
u
ctio
n
.
I
t
p
er
f
o
r
m
s
b
etter
in
ca
s
e
o
f
l
o
w
SNR
wh
er
e
n
o
is
e
p
o
wer
is
ca
lcu
lated
lo
ca
lly
.
Fu
r
th
er
m
o
r
e,
SP
P
i
s
also
u
s
ed
to
m
ak
e
a
s
o
f
t
-
d
ec
is
io
n
i
n
d
etec
tin
g
th
e
n
o
is
e
s
tatis
tics
in
th
e
ab
s
en
ce
o
f
th
e
s
p
ee
ch
an
d
MM
SE
b
ased
n
o
is
e
tr
ac
k
in
g
is
u
s
ed
in
th
e
s
p
ee
c
h
p
r
esen
ce
.
T
h
is
g
iv
es
b
etter
n
o
is
e
esti
m
atio
n
an
d
im
p
r
o
v
es
th
e
p
er
f
o
r
m
a
n
ce
at
h
ig
h
SNR
.
Fin
ally
,
r
ec
u
r
s
iv
e
s
m
o
o
th
in
g
is
ap
p
lied
r
es
u
ltin
g
in
th
e
ef
f
icien
t
s
in
g
l
e
-
ch
an
n
el
s
p
ee
ch
en
h
an
ce
m
e
n
t.
T
h
e
s
p
ee
c
h
en
h
an
ce
m
en
t
u
s
in
g
ad
ap
tiv
e
f
ilter
in
g
m
ay
r
ed
u
ce
th
e
ce
p
s
tr
al
s
m
o
o
th
in
g
,
also
ec
h
o
ca
n
ce
latio
n
an
d
im
p
r
o
v
e
th
e
r
esu
lts
f
u
r
th
er
.
T
h
e
r
esu
lts
in
d
icate
th
at
th
e
p
r
o
p
o
s
ed
wo
r
k
is
im
p
r
o
v
in
g
th
e
s
p
ee
ch
q
u
ality
at
0
d
B
,
5
d
B
an
d
1
0
d
B
an
d
s
tan
d
s
m
o
s
tly
g
o
o
d
,
an
d
c
o
m
p
ar
ativ
e
i
n
s
o
m
e
p
ar
ticu
lar
ca
s
es.
Sp
ee
ch
in
tellig
ib
ilit
y
is
also
g
o
o
d
ac
c
o
r
d
in
g
to
th
e
Me
an
Op
in
io
n
Sco
r
e
(
MO
S)
o
f
t
h
e
s
u
b
jects.
Als
o
,
it
im
p
r
o
v
es
th
e
q
u
ality
o
f
th
e
lo
wer
en
er
g
y
co
n
ce
n
tr
ated
s
ig
n
a
ls
ef
f
icien
tly
wh
en
co
m
p
a
r
ed
t
o
th
e
o
th
er
s
tate
o
f
ar
t te
ch
n
iq
u
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
9
,
No
.
2,
Au
g
u
s
t
20
20
:
63
–
7
2
72
RE
F
E
R
E
NC
E
S
[1
]
R.
E.
Ka
lma
n
,
“
A
Ne
w
Ap
p
r
o
a
c
h
t
o
Li
n
e
a
r
F
il
terin
g
a
n
d
P
re
d
i
c
ti
o
n
P
ro
b
lem
s,”
J
o
u
rn
a
l
o
f
Ba
s
ic
En
g
in
e
e
rin
g
,
v
o
l.
8
2
,
n
o
.
1
,
p
p
.
3
5
-
4
5
,
1
9
6
0
.
[2
]
Bo
ll
,
S
.
F
,
“
S
u
p
p
re
ss
io
n
o
f
a
c
o
u
stic
n
o
ise
in
sp
e
e
c
h
u
sin
g
sp
e
c
tr
a
l
su
b
trac
ti
o
n
,
”
IEE
E
Tran
sa
c
ti
o
n
s
o
n
Ac
o
u
stics
,
S
p
e
e
c
h
,
a
n
d
S
i
g
n
a
l
P
ro
c
e
ss
in
g
,
p
p
.
1
1
3
–
1
2
0
,
1
9
7
9
.
[3
]
M
.
Be
ro
u
t
i,
R.
S
c
h
wa
rtz,
a
n
d
J.
M
a
k
h
o
u
l,
“
E
n
h
a
n
c
e
m
e
n
t
o
f
sp
e
e
c
h
c
o
rru
p
ted
b
y
a
c
o
u
stic
n
o
ise
,
”
in
th
e
Pro
c
e
e
d
in
g
s
o
f
IEE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Aco
u
st
ics
,
S
p
e
e
c
h
a
n
d
S
ig
n
a
l
Pro
c
e
ss
in
g
.
,
p
p
.
2
0
8
–
2
1
1
,
1
9
7
9
.
[4
]
K.
T.
Tala
le,
a
n
d
S
.
T.
G
a
n
d
h
e
,
“
S
p
e
e
c
h
Co
m
p
re
ss
io
n
u
si
n
g
AD
P
CM
,
”
IJ
CA
-
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
mp
u
ter
Ap
p
li
c
a
ti
o
n
s I
n
t
h
e
Pro
c
e
e
d
in
g
s o
n
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
in
Co
mp
u
ta
ti
o
n
a
l
In
tell
ig
e
n
c
e
,
n
o
.
8
,
2
0
1
2
.
[5
]
L.
R.
Ra
b
i
n
e
r,
a
n
d
R
.
W.
S
c
h
a
fe
r,
“
Dig
it
a
l
P
r
o
c
e
ss
in
g
o
f
S
p
e
e
c
h
S
ig
n
a
l
,
”
T
h
e
J
o
u
rn
a
l
o
f
t
h
e
Aco
u
stica
l
S
o
c
iety
o
f
Ame
r
ica
,
v
o
l.
6
7
,
n
o
.
4
,
p
p
.
1
4
0
6
-
1
4
0
7
,
1
9
8
0
.
[6
]
H.
J.
M
.
S
tee
n
e
k
e
n
a
n
d
T.
H
o
u
t
g
a
st,
“
A
p
h
y
sic
a
l
m
e
th
o
d
f
o
r
m
e
a
su
rin
g
sp
e
e
c
h
-
tran
sm
issio
n
q
u
a
li
ty
,
”
J
o
u
rn
a
l
o
f
t
h
e
Aco
u
stica
l
S
o
c
iety
o
f
Ame
ric
a
,
v
o
l.
6
7
,
n
o
.
1
,
p
p
.
3
1
8
–
3
2
6
,
1
9
8
0
.
[7
]
T.
G
e
rk
m
a
n
n
a
n
d
R.
C
.
He
n
d
ri
k
s,
“
No
ise
P
o
we
r
Esti
m
a
ti
o
n
Ba
se
d
o
n
Th
e
P
ro
b
a
b
il
it
y
o
f
S
p
e
e
c
h
P
re
se
n
c
e
,
”
IEE
E
W
o
rk
sh
o
p
o
n
Ap
p
li
c
a
ti
o
n
s
o
f
S
ig
n
a
l
Pro
c
e
ss
in
g
t
o
A
u
d
io
a
n
d
Aco
u
stics
,
2
0
1
1
.
[8
]
P
a
rk
,
Je
o
n
g
-
S
i
k
,
G
il
-
Jin
Ja
n
g
,
a
n
d
Ji
-
Hw
a
n
Kim
.
"
N
o
ise
Re
d
u
c
ti
o
n
S
c
h
e
m
e
fo
r
S
p
e
e
c
h
Re
c
o
g
n
it
io
n
in
M
o
b
il
e
De
v
ice
s,"
In
fo
rm
a
ti
o
n
T
e
c
h
n
o
lo
g
y
Co
n
v
e
rg
e
n
c
e
.
S
p
rin
g
e
r,
Do
r
d
re
c
h
t,
p
p
.
9
4
9
-
9
5
5
,
2
0
1
3
.
[9
]
Zh
a
o
,
Hu
a
n
,
e
t
a
l.
"
A
n
e
w
so
ft
m
a
sk
in
g
m
e
th
o
d
fo
r
sp
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
in
th
e
fre
q
u
e
n
c
y
d
o
m
a
i
n
.
"
El
e
k
tro
n
ika
ir
El
e
k
tro
tec
h
n
ika
,
v
o
l.
20
,
n
o
.
2
,
p
p
.
58
-
63
,
2
0
1
4
.
[1
0
]
J.
Je
n
se
n
,
R.
C.
He
n
d
ri
k
s,
a
n
d
T.
G
e
rk
m
a
n
,
“
DFT
d
o
m
a
in
b
a
se
d
sin
g
le
-
m
icro
p
h
o
n
e
n
o
ise
re
d
u
c
ti
o
n
fo
r
s
p
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t,
”
A
su
rv
e
y
o
f
th
e
sta
te
-
of
-
th
e
-
a
rt
,
2
0
1
3
.
[1
1
]
P
.
S
u
n
,
J
u
n
Qi
n
,
“
S
p
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
v
ia
two
-
sta
g
e
d
u
a
l
tree
c
o
m
p
lex
wa
v
e
let
p
a
c
k
e
t
tran
sfo
r
m
with
a
sp
e
e
c
h
p
re
se
n
c
e
p
ro
b
a
b
il
it
y
e
stim
a
to
r”
,
T
h
e
J
o
u
rn
a
l
o
f
Ac
o
u
stic
S
o
c
iety
o
f
Ame
ric
a
,
v
o
l.
1
4
1
,
no.
2
,
p
p
.
8
0
8
-
8
1
7
,
Ja
n
2
0
1
7
.
[1
2
]
Yo
u
n
g
-
S
e
o
k
Ch
o
i,
“
A
Ne
w
S
u
b
b
a
n
d
Ad
a
p
t
iv
e
F
i
lt
e
rin
g
Alg
o
rit
h
m
fo
r
S
p
a
rse
S
y
ste
m
Id
e
n
ti
fica
ti
o
n
with
Im
p
u
lsiv
e
No
ise
”
,
J
o
u
rn
a
l
o
f
A
p
p
l
ied
M
a
t
h
e
ma
ti
c
s
,
v
o
l.
2
0
1
4
.
[1
3
]
Tran
,
T
ien
Du
n
g
,
Qu
o
c
Cu
o
n
g
Ng
u
y
e
n
,
a
n
d
Da
n
g
Kh
o
a
N
g
u
y
e
n
.
"
S
p
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
u
sin
g
m
o
d
if
ied
IM
CRA
a
n
d
OML
S
A m
e
th
o
d
s
,
"
I
n
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Co
mm
u
n
ica
t
io
n
s a
n
d
E
lec
tro
n
ics
2
0
1
0
.
,
IEE
E,
2
0
1
0
.
[1
4
]
I.
Co
h
e
n
,
“
No
ise
sp
e
c
tru
m
e
stim
a
ti
o
n
in
a
d
v
e
rse
e
n
v
ir
o
n
m
e
n
ts:
I
m
p
ro
v
e
d
m
in
ima
c
o
n
tr
o
ll
e
d
re
c
u
rsiv
e
a
v
e
ra
g
i
n
g
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
in
S
p
e
e
c
h
A
u
d
i
o
Pro
c
e
ss
in
g
,
v
o
l.
1
1
,
p
p
.
4
6
6
-
4
7
5
,
2
0
0
3
.
[1
5
]
Zh
a
o
,
Hu
a
n
,
e
t
a
l.
"
A
n
e
w
so
ft
m
a
sk
in
g
m
e
th
o
d
fo
r
sp
e
e
c
h
e
n
h
a
n
c
e
m
e
n
t
in
th
e
fre
q
u
e
n
c
y
d
o
m
a
in
.
"
El
e
k
tro
n
ik
a
i
r
El
e
k
tro
tec
h
n
ika
,
v
o
l
.
2
0
.
n
o
.
2
,
p
p
.
58
-
63
,
2
0
1
4
.
[1
6
]
S
o
n
u
Ba
la,
a
n
d
M
o
h
a
m
m
a
d
Arif
,
“
P
e
rfo
rm
a
n
c
e
c
o
m
p
a
riso
n
o
f
d
is
c
re
te
tran
sfo
rm
s
o
n
sp
e
e
c
h
c
o
m
p
re
ss
e
d
se
n
sin
g
,
”
In
t
h
e
Pro
c
e
e
d
in
g
s
o
f
IE
EE
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
C
o
mp
u
ta
ti
o
n
a
l
I
n
telli
g
e
n
c
e
&
Co
mm
u
n
ica
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
p
p
.
6
3
2
-
6
3
7
,
2
0
1
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.