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.
7
,
No
.
6
,
Dec
em
b
er
201
7
,
p
p
.
3
6
5
5
~
3
6
6
3
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v7
i
6
.
pp
3
6
5
5
-
3663
3655
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
jo
u
r
n
a
l.c
o
m/o
n
lin
e/in
d
ex
.
p
h
p
/I
JE
C
E
Tow
a
rds a
n O
pti
m
a
l Sp
ea
ker Mo
d
eling
in
Spea
ker
Verific
a
tion
Sy
ste
m
s
using
P
e
rso
na
liz
ed
Ba
ck
g
ro
und Mo
dels
Ay
o
ub
B
o
uzia
ne,
J
a
m
a
l K
ha
rr
o
ub
i,
Arsa
la
ne
Z
a
rg
hil
i
L
a
b
o
ra
to
ry
o
f
In
telli
g
e
n
t
S
y
ste
m
s
a
n
d
A
p
p
li
c
a
ti
o
n
s,
U
n
iv
e
rsity
S
id
i
M
o
h
a
m
e
d
Be
n
A
b
d
e
ll
a
h
F
e
z
,
M
o
r
o
c
c
o
.
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
9
,
2
0
1
7
R
ev
i
s
ed
J
u
n
2
9
,
2
0
1
7
A
cc
ep
ted
J
u
l
14
,
2
0
1
7
T
h
is
p
a
p
e
r
p
re
se
n
ts
a
n
o
v
e
l
sp
e
a
k
e
r
m
o
d
e
li
n
g
a
p
p
ro
a
c
h
f
o
r
sp
e
a
k
e
r
re
c
o
g
n
it
io
n
sy
ste
m
s.
T
h
e
b
a
sic
id
e
a
o
f
th
is
a
p
p
ro
a
c
h
c
o
n
sists
o
f
d
e
riv
in
g
th
e
targ
e
t
sp
e
a
k
e
r
m
o
d
e
l
f
ro
m
a
p
e
rs
o
n
a
li
z
e
d
b
a
c
k
g
ro
u
n
d
m
o
d
e
l,
c
o
m
p
o
se
d
o
n
ly
o
f
th
e
UBM
G
a
u
ss
ian
c
o
m
p
o
n
e
n
ts
w
h
ich
a
re
re
a
ll
y
p
re
se
n
t
in
th
e
sp
e
e
c
h
o
f
th
e
targ
e
t
sp
e
a
k
e
r.
T
h
e
m
o
ti
v
a
ti
o
n
b
e
h
in
d
th
e
d
e
riv
a
ti
o
n
o
f
sp
e
a
k
e
rs’
m
o
d
e
ls
f
ro
m
p
e
rso
n
a
li
z
e
d
b
a
c
k
g
ro
u
n
d
m
o
d
e
ls
is
to
e
x
p
lo
it
t
h
e
o
b
se
rv
e
d
d
if
f
e
re
n
c
e
in
so
m
e
a
c
o
u
stic
-
c
las
se
s
b
e
t
w
e
e
n
sp
e
a
k
e
rs,
in
o
rd
e
r
t
o
im
p
ro
v
e
th
e
p
e
rf
o
r
m
a
n
c
e
o
f
sp
e
a
k
e
r
re
c
o
g
n
it
io
n
sy
ste
m
s.
T
h
e
p
ro
p
o
se
d
a
p
p
r
o
a
c
h
w
a
s
e
v
a
lu
a
ted
f
o
r
sp
e
a
k
e
r
v
e
ri
f
ica
ti
o
n
tas
k
u
sin
g
v
a
rio
u
s
a
m
o
u
n
ts
o
f
train
in
g
a
n
d
t
e
stin
g
sp
e
e
c
h
d
a
ta.
T
h
e
e
x
p
e
ri
m
e
n
tal
re
su
lt
s
sh
o
w
e
d
th
a
t
th
e
p
ro
p
o
se
d
a
p
p
ro
a
c
h
is
e
ff
icie
n
ti
n
ter
m
so
f
b
o
th
v
e
rif
ic
a
ti
o
n
p
e
rf
o
r
m
a
n
c
e
a
n
d
c
o
m
p
u
tatio
n
a
l
c
o
st
d
u
rin
g
t
h
e
tes
ti
n
g
p
h
a
se
o
f
th
e
sy
ste
m
,
c
o
m
p
a
re
d
to
th
e
trad
it
io
n
a
l
UBM
b
a
se
d
sp
e
a
k
e
r
re
c
o
g
n
it
io
n
sy
ste
m
s.
K
ey
w
o
r
d
:
Sp
ea
k
er
m
o
d
elin
g
Sp
ea
k
er
v
er
i
f
icatio
n
Gau
s
s
ia
n
m
i
x
t
u
r
e
m
o
d
els
Un
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
Ma
x
i
m
u
m
a
p
o
s
ter
io
r
i
(
MA
P
)
P
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el
s
Co
p
y
rig
h
t
©
2
0
1
7
I
n
stit
u
te o
f
Ad
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
:
Ay
o
u
b
B
o
u
zian
e
,
L
ab
o
r
ato
r
y
o
f
I
n
telli
g
e
n
t S
y
s
t
e
m
s
a
n
d
A
p
p
licatio
n
s
,
Sid
i M
o
h
a
m
ed
B
en
A
b
d
ellah
Un
i
v
er
s
it
y
,
P
.
B
:
2
2
0
2
I
m
m
o
u
ze
r
R
o
ad
,
Fe
z,
Mo
r
o
cc
o
.
E
m
ail: a
y
o
u
b
.
b
o
u
zian
e
@
u
s
m
b
a.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
GM
M
-
UB
M
b
ased
s
p
ea
k
er
m
o
d
elin
g
ap
p
r
o
ac
h
w
a
s
f
ir
s
tl
y
i
n
tr
o
d
u
ce
d
to
t
h
e
s
p
ea
k
er
r
ec
o
g
n
itio
n
co
m
m
u
n
it
y
b
y
R
e
y
n
o
ld
s
,
i
n
2
0
0
0
[
1
]
–
[
3
]
.
Sin
ce
th
en
,
it
h
a
v
e
b
ec
o
m
e
th
e
p
r
ed
o
m
i
n
an
t
a
p
p
r
o
ac
h
f
o
r
s
p
ea
k
er
m
o
d
eli
n
g
i
n
tex
t
-
in
d
ep
en
d
en
t
s
p
ea
k
er
r
ec
o
g
n
i
tio
n
s
y
s
te
m
s
,
an
d
th
e
b
asi
s
o
f
t
h
e
m
o
s
t
s
u
cc
ess
f
u
l
ap
p
r
o
ac
h
es
th
at
h
a
v
e
b
ee
n
e
m
er
g
ed
in
t
h
e
last
d
ec
ad
e:
th
e
h
y
b
r
id
GM
M
-
SVM
ap
p
r
o
ac
h
[
4
]
,
[
5
]
,
th
e
jo
in
t
f
ac
to
r
an
al
y
s
is
ap
p
r
o
ac
h
[
6
]
–
[
8
]
,
an
d
th
e
r
ec
en
tl
y
i
n
tr
o
d
u
ce
d
i
-
v
ec
to
r
s
ap
p
r
o
ac
h
[
9
]
,
[
1
0
]
.
T
h
e
m
o
ti
v
atio
n
b
eh
i
n
d
t
h
e
u
s
e
o
f
ad
ap
ted
Gau
s
s
ia
n
m
i
x
tu
r
e
m
o
d
els
f
o
r
s
p
ea
k
er
m
o
d
eli
n
g
i
s
g
en
er
all
y
b
ased
o
n
th
e
ass
u
m
p
tio
n
th
at
Gau
s
s
ia
n
d
en
s
itie
s
m
a
y
m
o
d
el
a
s
et
o
f
h
id
d
en
ac
o
u
s
tical
cla
s
s
e
s
t
h
at
r
ef
lect
s
o
m
e
g
en
er
al
s
p
ea
k
er
d
ep
en
d
en
t
v
o
ca
l tr
ac
t
ch
ar
ac
ter
is
tic
s
.
T
h
e
m
ai
n
id
ea
o
f
th
e
UB
M
-
b
a
s
ed
s
p
ea
k
er
r
ec
o
g
n
itio
n
s
y
s
te
m
s
co
n
s
i
s
t
o
f
d
er
iv
i
n
g
t
h
e
tar
g
et
s
p
ea
k
er
m
o
d
el
f
r
o
m
a
u
n
iv
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
th
at
r
ep
r
esen
t
s
th
e
s
et
o
f
all
h
u
m
an
ac
o
u
s
tic
c
lass
es
(
e.
g
.
,
g
e
n
d
er
d
ep
en
d
en
t
ac
o
u
s
tic
cla
s
s
es,
a
g
e
-
d
ep
en
d
en
t
ac
o
u
s
tic
clas
s
e
s
,
ac
ce
n
t
d
ep
en
d
en
t
ac
o
u
s
tic
cla
s
s
es
…)
.
T
h
e
tar
g
et
s
p
ea
k
er
m
o
d
el
w
ill
t
h
er
ef
o
r
e
b
e
co
m
p
o
s
ed
o
f
an
ad
ap
ted
v
er
s
io
n
o
f
th
e
v
ar
io
u
s
ac
o
u
s
tic
cl
ass
es
d
ef
i
n
ed
in
t
h
e
u
n
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el.
Ho
w
e
v
er
,
as
it
s
o
u
n
d
s
to
t
h
e
h
u
m
a
n
ea
r
,
s
p
ea
k
er
s
d
o
n
’
t
s
h
ar
e
th
e
s
a
m
e
ac
o
u
s
t
ic
class
es.
Fo
r
ex
a
m
p
le,
t
h
e
ac
o
u
s
tic
clas
s
es
o
f
m
ale
s
p
ea
k
er
s
ar
e
d
if
f
er
en
t
f
r
o
m
t
h
o
s
e
o
f
f
e
m
ale
s
p
ea
k
er
s
a
n
d
th
e
ac
o
u
s
tic
clas
s
es
o
f
ad
u
lt
s
p
ea
k
er
s
ar
e
d
if
f
er
en
t
f
r
o
m
t
h
o
s
e
o
f
m
i
n
o
r
s
p
ea
k
er
s
,
as
well
as,
th
e
ac
o
u
s
tic
class
es
o
f
ti
m
id
s
p
ea
k
er
s
ar
e
d
if
f
er
en
t
f
r
o
m
th
o
s
e
o
f
b
o
ld
s
p
ea
k
er
s
,
etc.
T
h
er
eb
y
,
it
s
ee
m
s
t
h
at
it
i
s
n
o
t
lo
g
ical
to
m
o
d
el
a
s
p
ea
k
er
b
y
a
n
ad
a
p
ted
v
er
s
io
n
o
f
a
n
ac
o
u
s
tic
cla
s
s
w
h
ic
h
m
a
y
n
o
t
e
x
i
s
t
i
n
i
ts
v
o
ice.
F
u
r
th
er
m
o
r
e,
th
e
d
if
f
er
en
ce
o
f
ac
o
u
s
tic
clas
s
es
b
et
w
ee
n
s
p
ea
k
er
s
ca
n
b
e
ex
p
lo
ited
to
d
is
cr
im
i
n
ate
b
et
w
e
en
th
e
m
in
s
p
ea
k
er
r
ec
o
g
n
itio
n
s
y
s
te
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
6
5
5
–
3
6
6
3
3656
I
n
v
ie
w
o
f
th
e
s
e
o
b
s
er
v
atio
n
s
,
th
e
p
r
ese
n
t
s
t
u
d
y
atte
m
p
ts
,
f
ir
s
tl
y
,
to
e
x
p
er
i
m
e
n
tall
y
i
n
v
e
s
tig
ate
t
h
e
d
eg
r
ee
o
f
ac
o
u
s
t
ic
-
cla
s
s
e
s
’
d
i
f
f
er
en
ce
b
et
w
ee
n
s
p
ea
k
er
s
,
a
n
d
s
ec
o
n
d
l
y
,
to
p
r
o
p
o
s
e
a
s
p
ea
k
er
m
o
d
eli
n
g
ap
p
r
o
ac
h
th
at
tak
es
i
n
to
ac
co
u
n
t
th
e
s
e
o
b
s
er
v
atio
n
s
an
d
ex
p
lo
its
th
e
d
if
f
er
en
ce
s
i
n
ac
o
u
s
tic
class
e
s
b
et
w
ee
n
s
p
ea
k
er
s
to
i
m
p
r
o
v
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
s
p
ea
k
er
r
ec
o
g
n
itio
n
s
y
s
te
m
s
.
T
h
e
r
em
ai
n
d
er
o
f
t
h
i
s
p
ap
er
is
o
r
g
an
ized
a
s
f
o
llo
w
s
.
T
h
e
f
i
r
s
t
s
ec
t
io
n
g
i
v
e
s
a
b
r
ief
o
v
er
v
ie
w
o
f
t
h
e
tr
ad
itio
n
al
GM
M
-
UB
M
b
ased
Sp
ea
k
er
v
er
if
icatio
n
s
y
s
te
m
s
.
Nex
t,
th
e
p
r
o
p
o
s
ed
s
p
ea
k
er
m
o
d
eli
n
g
ap
p
r
o
ac
h
,
b
ased
o
n
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
els,
i
s
in
tr
o
d
u
ce
d
.
Af
ter
w
ar
d
,
ex
p
er
i
m
e
n
tal
r
es
u
lts
an
d
d
is
cu
s
s
io
n
s
ar
e
p
r
esen
ted
in
t
h
e
th
ir
d
s
ec
tio
n
.
Fin
all
y
,
co
n
cl
u
s
io
n
s
a
n
d
f
u
t
u
r
e
r
esear
ch
d
ir
ec
tio
n
s
ar
e
d
r
a
w
n
in
t
h
e
las
t sectio
n
.
2.
T
RAD
I
T
I
O
NAL
G
M
M
-
UB
M
B
ASE
D
SPEAK
E
R
VE
RI
F
I
CAT
I
O
N
SYS
T
E
M
S
T
h
e
GM
M
-
UB
M
b
ased
s
p
ea
k
er
v
er
i
f
icatio
n
s
y
s
te
m
w
as
o
r
ig
in
a
ll
y
p
r
o
p
o
s
ed
b
y
R
e
y
n
o
l
d
s
in
2
0
0
0
[
1
]
,
[
1
1
]
.
Sin
ce
th
en
,
it
h
a
s
b
ec
o
m
e
t
h
e
p
r
ed
o
m
i
n
an
t
ap
p
r
o
ac
h
f
o
r
s
p
ea
k
er
m
o
d
eli
n
g
i
n
tex
t
-
i
n
d
ep
en
d
en
t
s
p
ea
k
er
r
ec
o
g
n
i
tio
n
s
y
s
te
m
s
a
n
d
th
e
b
asi
s
o
f
t
h
e
m
o
s
t
s
u
cc
ess
f
u
l
ap
p
r
o
ac
h
es
t
h
at
h
av
e
b
ee
n
e
m
er
g
ed
in
t
h
e
last
d
ec
ad
e.
T
h
e
m
ai
n
id
ea
o
f
th
e
GM
M
-
UB
M
ap
p
r
o
ac
h
co
n
s
is
t
s
,
as
s
h
o
w
n
i
n
Fig
u
r
e
1
,
o
f
d
er
iv
in
g
s
p
ea
k
er
s
’
m
o
d
el
s
f
r
o
m
a
u
n
iv
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
u
s
i
n
g
m
a
x
i
m
u
m
a
p
o
s
ter
io
r
i
(
MA
P
)
ad
ap
tatio
n
[
1
]
.
T
h
e
Un
i
v
er
s
al
B
ac
k
g
r
o
u
n
d
Mo
d
el
(
UB
M)
is
t
y
p
icall
y
a
Ga
u
s
s
ian
m
i
x
tu
r
e
m
o
d
el
(
GM
M)
t
h
at
r
ep
r
esen
t
s
t
h
e
d
is
tr
ib
u
tio
n
o
f
th
e
e
n
tire
ac
o
u
s
tic
s
p
ac
e
o
f
s
p
ee
ch
[
2
]
,
[
3
]
.
Fig
u
r
e
1
.
A
M
A
P
ad
ap
tatio
n
o
f
a
GM
M
co
m
p
r
is
e
s
5
Gau
s
s
ia
n
d
en
s
i
ties
.
Or
ig
in
al
Ga
u
s
s
ia
n
d
en
s
ities
o
f
th
e
UB
M
ar
e
d
ep
icted
as u
n
f
illed
ellip
s
es (
d
o
tted
lin
e)
,
w
h
er
ea
s
th
e
ad
ap
ted
Gau
s
s
ia
n
d
en
s
itie
s
ar
e
d
en
o
ted
b
y
f
illed
ellip
s
e
s
,
an
d
th
e
o
b
s
er
v
e
d
f
ea
tu
r
e
v
ec
to
r
s
ar
e
d
ep
icted
as s
m
all
cir
cles
T
h
e
MA
P
ad
ap
tatio
n
p
r
o
ce
s
s
is
g
e
n
er
all
y
co
m
p
o
s
ed
o
f
t
w
o
s
tep
s
.
F
ir
s
t,
th
e
s
u
f
f
icie
n
t
s
tatis
t
ics
esti
m
ates
o
f
t
h
e
s
p
ea
k
er
’
s
tr
ain
i
n
g
d
ata
ar
e
co
m
p
u
ted
f
o
r
ea
ch
m
i
x
t
u
r
e
in
th
e
UB
M.
Nex
t,
th
e
co
m
p
u
ted
s
u
f
f
icie
n
t
s
tati
s
tic
s
esti
m
ate
s
ar
e
co
m
b
in
ed
w
i
th
t
h
e
o
l
d
s
u
f
f
icie
n
t
s
tatis
tics
f
r
o
m
th
e
UB
M
m
i
x
t
u
r
e
p
ar
am
eter
s
u
s
in
g
a
d
ata
-
d
ep
en
d
en
t
m
i
x
i
n
g
co
e
f
f
ic
ien
t.
T
h
e
s
p
ec
if
ics
o
f
t
h
e
ad
ap
tatio
n
ar
e
as
f
o
llo
w
s
.
Gi
v
en
th
at
t
h
e
u
n
iv
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
is
co
m
p
o
s
ed
o
f
M
Gau
s
s
ia
n
co
m
p
o
n
e
n
t
s
,
ea
ch
o
f
w
h
ic
h
is
p
ar
am
eter
ize
d
b
y
a
m
ea
n
v
ec
to
r
μ
i
,
v
ar
ian
ce
m
atr
ix
σ
2
i
an
d
its
w
ei
g
h
t
w
i
in
th
e
m
i
x
t
u
r
e
m
o
d
el.
I
n
itiall
y
,
th
e
p
o
s
ter
io
r
i
p
r
o
b
ab
ilit
y
o
f
ea
c
h
UB
M
co
m
p
o
n
e
n
t
{
μ
i
,
σ
i
,
w
i
}
g
iv
e
n
t
h
e
f
e
atu
r
e
v
ec
to
r
x
t
is
co
m
p
u
ted
as f
o
llo
w
s
:
(
|
)
=
(
)
∑
(
)
=
1
(
1
)
w
h
er
e
p
i
(
x
t
)
is
th
e
d
en
s
i
t
y
o
f
t
h
e
p
r
o
b
ab
ilit
y
,
g
i
v
en
b
y
:
(
)
=
1
(
2
)
2
⁄
|
|
1
2
⁄
{
−
1
2
(
−
)
−
1
(
−
)
}
(
2
)
T
h
e
s
u
f
f
icie
n
t s
tatis
t
ics
f
o
r
th
e
w
ei
g
h
t,
m
ea
n
,
an
d
v
ar
ia
n
ce
p
ar
a
m
eter
s
ar
e
th
e
n
co
m
p
u
ted
as f
o
llo
w
s
:
Un
iv
e
rs
a
l
Ba
c
k
g
r
o
u
n
d
M
o
d
e
l
Un
i
v
e
rsa
l
Ba
c
k
g
ro
u
n
d
M
o
d
e
l
S
p
e
c
if
ic S
p
e
a
k
e
r
M
o
d
e
l
T
e
stin
g
p
h
a
se
M
A
P
A
d
a
p
tatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
To
w
a
r
d
s
a
n
Op
tima
l S
p
ea
ke
r
Mo
d
elin
g
in
S
p
ea
ke
r
V
erif
ica
tio
n
S
ystems
…
(
A
yo
u
b
B
o
u
z
ia
n
e
)
3657
=
∑
(
|
)
=
1
;
(
)
=
1
∑
(
|
)
=
1
;
(
2
)
=
1
∑
(
|
)
=
1
2
(
3
)
T
h
er
ea
f
ter
,
th
e
co
m
p
u
ted
s
u
f
f
i
cien
t
s
tatis
tics
ar
e
u
s
ed
f
o
r
est
i
m
ati
n
g
t
h
e
ad
ap
ted
m
i
x
t
u
r
e
weig
h
ts
w
i
̂
,
m
ea
n
s
μ
i
̂
an
d
v
ar
ian
ce
s
σ
i
̂
o
f
th
e
g
iv
en
s
p
ea
k
er
:
̂
=
[
(
)
+
(
1
−
)
]
(
4
)
̂
=
[
)
/
+
(
1
−
)
]
(
4
)
2
̂
=
[
2
(
2
)
+
(
1
−
2
)
(
2
−
2
)
−
̂
2
]
(
5
)
w
it
h
,
=
n
i
/
(
n
i
+
)
,
ρ
∈
{
w
,
,
2
}
(
6
)
Her
e,
γ
is
a
s
ca
le
f
ac
to
r
co
m
p
u
ted
o
v
er
all
ad
ap
te
d
m
i
x
tu
r
e
w
ei
g
h
ts
to
en
s
u
r
e
th
at
th
e
y
s
u
m
to
u
n
it
y
,
β
i
ρ
,
ρ
∈
{
w
,
μ
,
σ
2
}
ar
e
th
e
ad
ap
tatio
n
co
ef
f
icie
n
ts
th
at
co
n
tr
o
l
h
o
w
t
h
e
ad
a
p
ted
GM
M
p
ar
am
eter
s
w
ill
b
e
af
f
ec
ted
b
y
t
h
e
o
b
s
er
v
ed
s
p
ea
k
er
d
ata,
an
d
r
ρ
is
a
f
i
x
ed
r
elev
an
ce
f
ac
to
r
f
o
r
p
ar
am
eter
ρ
.
An
o
v
er
all
d
ia
g
r
a
m
o
f
t
h
e
G
MM
-
UB
M
b
ased
s
p
ea
k
er
v
er
i
f
icatio
n
s
y
s
te
m
is
s
h
o
w
n
i
n
F
i
g
u
r
e
2
.
T
h
e
b
asic o
p
er
atin
g
s
tr
u
c
tu
r
e
o
f
t
h
e
s
y
s
te
m
,
as
s
h
o
w
n
in
Fi
g
u
r
e
2
,
is
co
m
p
o
s
ed
o
f
th
r
ee
p
h
ases
: th
e
tr
ain
i
n
g
p
h
a
s
e,
th
e
e
n
r
o
ll
m
e
n
t
p
h
a
s
e
an
d
t
h
e
t
esti
n
g
p
h
a
s
e.
D
u
r
in
g
t
h
e
f
ir
s
t
p
h
ase,
i.e
.
th
e
tr
ai
n
i
n
g
p
h
ase,
a
lar
g
e
co
llectio
n
o
f
s
p
ee
ch
u
tter
an
ce
s
is
co
llec
ted
f
r
o
m
a
b
ac
k
g
r
o
u
n
d
p
o
p
u
l
atio
n
o
f
s
p
ea
k
er
s
,
th
e
ir
co
r
r
esp
o
n
d
in
g
f
ea
t
u
r
e
v
ec
to
r
s
ar
e
ex
tr
ac
ted
a
n
d
u
s
ed
to
tr
ai
n
t
h
e
u
n
iv
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el.
T
h
e
tr
ain
i
n
g
p
r
o
ce
s
s
o
f
t
h
e
UB
M
i
s
d
o
n
e
g
en
er
all
y
u
s
i
n
g
m
a
x
i
m
u
m
li
k
eli
h
o
o
d
esti
m
atio
n
v
ia
th
e
E
M
alg
o
r
ith
m
.
I
n
t
h
e
s
ec
o
n
d
p
h
ase,
i.e
.
th
e
en
r
o
ll
m
e
n
t
p
h
a
s
e,
s
p
ea
k
er
m
o
d
els
o
f
n
e
w
clie
n
t
s
p
ea
k
er
s
ar
e
d
er
iv
ed
f
r
o
m
t
h
e
u
n
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
th
r
o
u
g
h
M
A
P
ad
ap
tatio
n
u
s
in
g
t
h
e
s
p
ea
k
er
s
’
tr
ain
i
n
g
f
ea
tu
r
e
v
ec
to
r
s
[
1
3
]
.
W
h
ile
in
t
h
e
test
i
n
g
p
h
ase,
t
h
e
ex
tr
ac
ted
f
ea
t
u
r
e
v
ec
to
r
s
o
f
th
e
u
n
k
n
o
w
n
s
p
ea
k
er
’
s
u
t
ter
an
c
e
X
u
=
{
x
u
1
,
x
u
2
,
.
.
.
,
x
u
N
}
ar
e
co
m
p
ar
ed
ag
ai
n
s
t
b
o
th
th
e
clai
m
ed
tar
g
et
s
p
ea
k
er
m
o
d
el
an
d
t
h
e
b
a
c
k
g
r
o
u
n
d
m
o
d
el.
T
h
e
lo
g
lik
eli
h
o
o
d
r
atio
LLR
(
X
u
;
λ
s
p
k
,
λ
UB
M
)
b
et
w
ee
n
t
h
e
c
lai
m
ed
s
p
ea
k
er
m
o
d
el
an
d
t
h
e
u
n
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
is
t
h
e
n
ca
lcu
lated
a
n
d
u
s
ed
to
m
a
k
e
a
d
ec
is
io
n
ab
o
u
t
t
h
e
ac
ce
p
t
an
ce
/r
ej
ec
tio
n
o
f
t
h
e
clai
m
ed
id
en
tit
y
.
T
h
e
lo
g
lik
eli
h
o
o
d
r
atio
(
L
L
R
)
o
f
th
e
t
est u
tter
a
n
ce
X
u
b
etw
ee
n
t
h
e
s
p
e
ak
er
m
o
d
el
λ
j
an
d
th
e
UB
M
m
o
d
el
λ
UB
M
:
(
;
,
)
=
1
[
∑
(
|
)
=
1
−
∑
(
|
)
=
1
]
(
7
)
W
ith
,
(
|
)
=
∑
(
)
=
1
(
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
6
5
5
–
3
6
6
3
3658
Fig
u
r
e
2
.
B
lo
ck
d
iag
r
a
m
o
f
th
e
GM
M
-
UB
M
b
ased
s
p
ea
k
er
v
er
if
ica
tio
n
s
y
s
te
m
3.
SPEAK
E
R
RE
CO
G
N
I
T
I
O
N
USI
N
G
P
E
RSO
NA
L
I
Z
E
D
B
ACK
G
RO
UND
M
O
DE
L
S
(
P
B
M
S)
T
h
e
m
ai
n
id
ea
o
f
t
h
e
p
r
o
p
o
s
ed
P
B
M
-
b
ased
s
p
ea
k
er
m
o
d
elin
g
ap
p
r
o
ac
h
co
n
s
is
t
s
o
f
ad
ap
tin
g
th
e
tar
g
et
s
p
ea
k
er
m
o
d
el
f
r
o
m
a
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el
(
P
B
M)
,
co
m
p
o
s
ed
o
n
l
y
o
f
th
e
UB
M
Gau
s
s
ia
n
co
m
p
o
n
e
n
t
s
w
h
ich
ar
e
ac
tu
a
ll
y
p
r
ese
n
t
i
n
th
e
s
p
ea
k
er
’
s
s
p
ee
ch
.
T
h
e
MA
P
ad
ap
tatio
n
s
tep
o
f
tr
ad
itio
n
al
UB
M
b
ased
s
y
s
te
m
s
w
il
l,
th
er
ef
o
r
e,
b
e
p
r
ec
ed
e
d
b
y
a
s
elec
tio
n
s
tep
th
at
s
elec
t
s
t
h
e
b
ac
k
g
r
o
u
n
d
Gau
s
s
ia
n
co
m
p
o
n
e
n
t
s
w
h
ich
r
e
f
lect
t
h
e
g
en
er
al
f
o
r
m
o
f
t
h
e
ac
o
u
s
tic
cl
ass
es c
h
ar
ac
ter
izin
g
t
h
e
s
p
ea
k
er
,
s
ee
Fig
u
r
e
3.
Fig
u
r
e
3
.
A
M
A
P
ad
ap
tatio
n
o
f
a
GM
M
co
m
p
r
is
e
s
5
Gau
s
s
ia
n
d
en
s
i
ties
.
Or
ig
in
al
Ga
u
s
s
ia
n
d
en
s
ities
o
f
th
e
UB
M
ar
e
d
ep
icted
as u
n
f
illed
ellip
s
es (
d
o
tted
lin
e)
,
w
h
er
ea
s
th
e
ad
ap
ted
Gau
s
s
ia
n
d
en
s
itie
s
ar
e
d
en
o
ted
b
y
f
illed
ellip
s
e
s
,
an
d
th
e
o
b
s
er
v
e
d
f
ea
tu
r
e
v
ec
to
r
s
ar
e
d
ep
icted
as s
m
all
cir
cles.
Giv
e
n
a
UB
M
o
f
M
Gau
s
s
ia
n
co
m
p
o
n
en
t
λ
UB
M
=
{
μ
UB
M
i
,
σ
UB
M
i
,
w
UB
M
i
}
/
i
∈
{
1
,
2
,
…
,
N
}
,
an
d
M
tr
ain
i
n
g
f
ea
t
u
r
e
v
ec
to
r
s
X
=
{
x
1
,
x
2
,
…
,
x
N
}
,
ex
tr
ac
ted
f
r
o
m
t
h
e
tar
g
et
s
p
ea
k
er
’
s
s
p
ee
ch
.
T
h
e
P
B
MG
au
s
s
ia
n
co
m
p
o
n
e
n
t
s
ar
e
g
e
n
er
all
y
c
h
o
s
en
f
r
o
m
t
h
e
UB
M
u
s
i
n
g
a
win
n
er
-
ta
k
e
-
all
b
ased
s
tr
ate
g
y
.
A
UB
M
Ga
u
s
s
ia
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
To
w
a
r
d
s
a
n
Op
tima
l S
p
ea
ke
r
Mo
d
elin
g
in
S
p
ea
ke
r
V
erif
ica
tio
n
S
ystems
…
(
A
yo
u
b
B
o
u
z
ia
n
e
)
3659
co
m
p
o
n
e
n
t
θ
UB
M
=
(
μ
UB
M
,
σ
UB
M
,
w
UB
M
)
is
s
elec
ted
to
b
elo
n
g
to
t
h
e
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el
λ
PB
M
o
f
th
e
tar
g
et
s
p
ea
k
er
,
i
f
t
h
er
e
is
at
lea
s
t
o
n
e
f
ea
tu
r
e
v
ec
to
r
x
n
∈
X
,
w
h
er
e
t
h
e
UB
M
Ga
u
s
s
ian
co
m
p
o
n
en
t
θ
UB
M
i
ac
h
iev
e
s
t
h
e
m
a
x
i
m
u
m
p
o
s
ter
i
o
r
p
r
o
b
ab
ilit
y
o
f
x
n
b
elo
n
g
i
n
g
n
e
s
s
:
(
θ
|
n
)
≥
(
θ
|
n
)
,
∀
∈
{
1
,
2
,
…
,
N
}
(
9
)
On
ce
th
e
P
B
M
Gau
s
s
ia
n
co
m
p
o
n
en
ts
ar
e
s
elec
ted
,
th
e
w
ei
g
h
ts
w
j
o
f
th
e
s
elec
ted
co
m
p
o
n
e
n
t
s
ar
e
d
iv
id
ed
b
y
t
h
eir
s
u
m
s
o
th
at
t
h
e
to
tal
w
ei
g
h
t is eq
u
al
to
u
n
it
y
.
A
b
lo
ck
d
ia
g
r
a
m
o
f
th
e
s
p
ea
k
er
m
o
d
elin
g
p
r
o
ce
s
s
u
s
i
n
g
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
els
is
s
h
o
w
n
in
Fig
u
r
e
4
.
Firstl
y
,
t
h
e
tr
ai
n
in
g
f
ea
t
u
r
e
v
ec
to
r
s
o
f
t
h
e
t
ar
g
et
s
p
ea
k
er
ar
e
e
x
tr
ac
ted
f
r
o
m
its
e
n
r
o
ll
m
en
t
u
tter
an
ce
s
.
Nex
t,
th
e
e
x
tr
ac
te
d
f
ea
tu
r
e
v
ec
to
r
s
a
r
e
u
s
ed
to
s
elec
t
th
e
UB
M
Gau
s
s
ia
n
co
m
p
o
n
en
ts
w
h
ich
w
il
l
co
m
p
o
s
e
th
e
s
p
ea
k
er
’
s
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el.
Af
t
er
w
ar
d
s
,
t
h
e
co
m
p
o
s
ed
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el
i
s
u
tili
ze
d
d
o
d
er
iv
e
t
h
e
s
p
ea
k
er
m
o
d
el
u
s
in
g
t
h
e
MA
P
ad
ap
tatio
n
p
r
o
ce
d
u
r
e.
Fin
all
y
,
t
h
e
ad
a
p
ted
m
o
d
el
i
s
s
to
r
ed
to
g
eth
er
w
it
h
t
h
e
co
r
r
esp
o
n
d
in
g
i
n
d
ices o
f
t
h
e
P
B
M
Gau
s
s
ian
co
m
p
o
n
en
t
s
in
th
e
UB
M.
An
e
x
a
m
p
le
o
f
a
t
w
o
-
d
i
m
en
s
io
n
al
p
r
o
j
ec
tio
n
o
f
t
w
o
s
p
ea
k
er
s
’
f
ea
tu
r
es
an
d
th
e
m
ea
n
s
o
f
t
h
eir
co
r
r
esp
o
n
d
in
g
UB
M
an
d
P
B
M
ad
ap
ted
m
o
d
elsi
s
s
h
o
w
n
i
n
Fig
u
r
e
5.
A
s
it
ca
n
b
e
s
ee
n
f
r
o
m
t
h
is
f
ig
u
r
e,
th
e
m
ea
n
s
o
f
t
h
e
P
B
M
ad
ap
ted
m
o
d
el
s
f
it
t
h
e
s
p
ea
k
er
s
’
f
ea
t
u
r
es
b
etter
th
a
n
t
h
e
m
ea
n
s
o
f
t
h
e
tr
ad
itio
n
al
UB
M
ad
ap
ted
m
o
d
els.
Mo
r
eo
v
er
,
it
s
ee
m
s
t
h
at
th
e
ad
ap
tatio
n
o
f
t
h
e
UB
M
Gau
s
s
ia
n
co
m
p
o
n
e
n
t
s
w
h
ic
h
h
av
e
n
'
t
an
y
r
elatio
n
s
h
ip
w
it
h
th
e
tar
g
et
s
p
ea
k
er
in
f
l
u
en
ce
o
n
th
e
f
ea
tu
r
e
v
ec
to
r
s
b
elo
n
g
i
n
g
n
e
s
s
to
th
e
a
p
p
r
o
p
r
iate
Gau
s
s
ian
co
m
p
o
n
e
n
t
s
,
w
h
ic
h
t
h
er
ef
o
r
e
af
f
ec
t n
e
g
ati
v
el
y
th
e
ad
ap
ted
m
o
d
el.
Fig
u
r
e
4
.
B
lo
ck
d
iag
r
a
m
o
f
s
p
ea
k
er
m
o
d
elin
g
p
r
o
ce
s
s
u
s
i
n
g
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
els
T
h
e
sp
e
a
k
e
r’s
f
e
a
tu
re
v
e
c
to
rs
Th
e
U
n
iv
e
rsa
l
Ba
c
k
g
ro
u
n
d
M
o
d
e
l
Th
e
U
BM
a
d
a
p
t
e
d
m
o
d
e
l
Th
e
P
BM
a
d
a
p
ted
m
o
d
e
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
6
5
5
–
3
6
6
3
3660
Fig
u
r
e
5
.
A
t
w
o
-
d
i
m
e
n
s
io
n
al
p
r
o
j
ec
tio
n
o
f
t
w
o
s
p
ea
k
er
s
’
f
e
atu
r
es (
b
lu
e
p
o
in
t
s
)
,
th
e
m
ea
n
s
o
f
th
eir
co
r
r
esp
o
n
d
in
g
UB
M
an
d
P
B
M
ad
ap
ted
m
o
d
els (
in
g
r
ee
n
a
n
d
y
ello
w
p
o
in
t
s
,
r
esp
ec
tiv
el
y
)
an
d
th
e
m
ea
n
s
o
f
th
e
UB
M
m
o
d
el
(
r
ed
p
o
in
ts
)
Du
r
in
g
t
h
e
test
p
h
a
s
e,
th
e
lo
g
-
lik
e
lih
o
o
d
r
atio
LLR
(
X
u
;
λ
s
p
k
,
λ
PB
M
)
b
etw
ee
n
t
h
e
clai
m
ed
s
p
ea
k
er
m
o
d
el
a
n
d
th
e
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
el
is
u
s
ed
to
m
a
k
e
a
d
ec
is
io
n
ab
o
u
t
th
e
ac
ce
p
tan
ce
o
r
th
e
r
ej
ec
tio
n
o
f
th
e
clai
m
ed
id
en
tit
y
:
T
h
e
m
o
ti
v
atio
n
b
eh
in
d
th
e
u
s
e
o
f
t
h
e
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
eli
n
s
tead
o
f
t
h
e
u
n
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
is
to
p
en
al
ize
th
e
d
ec
is
io
n
s
co
r
e
o
f
i
m
p
o
s
to
r
s
p
ea
k
er
s
w
h
o
d
o
n
’
t
s
h
ar
e
th
e
s
a
m
e
ac
o
u
s
t
ic
class
es
w
i
th
t
h
e
tar
g
e
t sp
ea
k
er
.
4.
E
XP
E
R
I
M
E
NT
S,
R
E
SU
L
T
S AN
D
DI
SCUS
SI
O
N
4
.
1
.
T
he
E
x
peri
m
ent
a
l P
ro
t
o
co
l
T
h
e
p
er
f
o
r
m
ed
e
x
p
er
i
m
e
n
t
s
i
n
t
h
is
s
t
u
d
y
w
er
e
co
n
d
u
cted
o
n
t
h
e
T
HUYG
-
2
0
S
R
E
d
ata
b
ase
[
1
4
]
.
T
h
is
d
atab
ase
is
g
en
er
all
y
co
m
p
o
s
ed
o
f
3
5
3
s
p
ea
k
er
s
,
co
ll
ec
ted
in
a
co
n
tr
o
lled
en
v
ir
o
n
m
en
t
(
s
ilen
t
o
f
f
ice
b
y
th
e
s
a
m
ec
ar
b
o
n
M
icr
o
p
h
o
n
e)
.
T
h
e
en
tire
s
p
ee
c
h
co
r
p
u
s
was
d
iv
id
ed
i
n
to
t
h
r
ee
d
ata
s
e
ts
:
t
h
e
f
ir
s
t
d
atase
t
co
n
s
is
ts
o
f
2
0
0
g
e
n
d
er
b
alan
c
ed
s
p
ea
k
er
s
(
1
0
0
Ma
le
a
n
d
1
0
0
Fe
m
ale)
a
n
d
d
ev
o
ted
to
t
r
ain
t
h
e
U
n
i
v
er
s
al
B
ac
k
g
r
o
u
n
d
Mo
d
el
(
UB
M)
,
t
h
e
s
ec
o
n
d
a
n
d
th
e
t
h
ir
d
d
ata
s
ets
ar
e
co
m
p
o
s
ed
o
f
t
h
e
s
a
m
e
s
et
o
f
1
5
3
clien
t
s
p
ea
k
er
s
.
T
h
e
f
ir
s
t
d
ata
s
et
co
m
p
r
i
s
es
t
h
e
tr
ai
n
in
g
s
p
ee
c
h
d
ata,
w
h
er
ea
s
th
e
s
ec
o
n
d
co
m
p
r
i
s
es
t
h
e
tes
tin
g
s
p
ee
ch
d
ata
o
f
th
e
1
5
3
s
p
ea
k
er
s
.
D
u
r
in
g
th
e
te
s
ti
n
g
p
h
ase
o
f
th
e
s
y
s
te
m
,
t
h
e
clien
t
s
p
ea
k
er
s
w
er
e
test
ed
ag
ain
s
t e
ac
h
o
th
er
,
r
esu
l
tin
g
i
n
to
tal
o
f
8
9
6
,
8
8
6
tr
ials
o
f
4
s
ec
o
n
d
s
an
d
4
4
1
,
5
5
8
tr
ials
o
f
8
s
ec
o
n
d
s
.
T
h
e
f
ea
tu
r
e
v
ec
to
r
s
o
f
t
h
e
o
v
er
all
s
p
ee
ch
u
tter
a
n
ce
s
w
er
e
e
x
tr
ac
ted
u
s
i
n
g
th
e
MF
C
C
ap
p
r
o
ac
h
:
th
e
d
ig
itized
s
p
ee
c
h
i
s
f
ir
s
tl
y
e
m
p
h
asized
u
s
in
g
a
s
i
m
p
le
f
ir
s
t o
r
d
er
d
ig
ital
f
ilter
w
it
h
tr
a
n
s
f
er
f
u
n
ct
io
n
H
(
z)
=
1
–
0
.
9
5
z.
Nex
t,
th
e
e
m
p
h
a
s
ized
s
p
ee
ch
s
ig
n
al
is
b
lo
ck
ed
in
to
Ham
m
i
n
g
-
w
in
d
o
w
ed
f
r
a
m
es
o
f
2
5
m
s
(
4
0
0
s
a
m
p
les)
in
len
g
t
h
w
i
th
1
0
m
s
(
1
6
0
s
a
m
p
les)
o
v
er
lap
b
e
t
w
ee
n
a
n
y
t
w
o
ad
j
ac
en
t
f
r
a
m
es.Fin
a
ll
y
,
1
9
Me
l
-
Fre
q
u
en
c
y
C
ep
s
tr
al
C
o
ef
f
icie
n
ts
w
er
e
ex
tr
ac
ted
f
r
o
m
ea
c
h
f
r
a
m
e
[
1
5
]
.
Du
r
in
g
t
h
e
tr
ai
n
in
g
p
h
a
s
e,
a
u
n
i
v
er
s
a
l
b
ac
k
g
r
o
u
n
d
m
o
d
el
(
UB
M)
o
f
1
0
2
4
Gau
s
s
ian
co
m
p
o
n
e
n
ts
was
tr
ain
ed
o
n
t
h
e
o
v
er
all
tr
ai
n
i
n
g
d
ata
(
7
.
5
h
o
u
r
s
o
f
s
p
ee
ch
)
u
s
i
n
g
t
h
e
E
M
alg
o
r
ith
m
.
4
.
2
.
I
nv
estig
a
t
io
n
o
n t
he
Deg
re
e
o
f
Dif
f
er
ence
in
Aco
us
t
ic
Cla
s
s
e
s
bet
w
ee
n Spea
k
er
s
T
h
ea
i
m
o
f
th
e
p
er
f
o
r
m
ed
ex
p
er
im
e
n
t
s
in
th
i
s
s
ec
tio
n
is
to
in
v
e
s
ti
g
ate
th
e
d
eg
r
ee
o
f
d
if
f
er
en
ce
i
n
ac
o
u
s
tic
clas
s
esb
et
w
ee
n
s
p
e
ak
er
s
.
Fo
r
th
is
p
u
r
p
o
s
e,
w
e
h
av
e
ca
r
r
ied
o
u
t
s
ev
er
al
s
p
ea
k
er
v
er
if
icat
io
n
ex
p
er
i
m
e
n
ts
in
w
h
ic
h
w
e
h
a
v
e
b
ased
o
n
l
y
o
n
th
e
d
if
f
er
en
ce
b
et
w
ee
n
t
h
e
ac
o
u
s
tic
clas
s
es
p
r
esen
t
in
th
e
tar
g
et
s
p
ea
k
er
’
s
s
p
ee
ch
an
d
th
o
s
e
p
r
esen
t
in
t
h
e
clai
m
ed
s
p
ea
k
er
’
s
s
p
ee
ch
.
T
o
p
r
o
ce
ed
,
w
e
h
a
v
e
r
ep
r
esen
ted
th
e
tr
ain
i
n
g
s
p
ee
c
h
u
tter
a
n
ce
(
s
)
o
f
th
e
tar
g
et
s
p
ea
k
er
a
n
d
th
e
t
est
s
p
ee
ch
u
tter
an
ce
o
f
t
h
e
cl
ai
m
ed
s
p
ea
k
er
b
y
h
is
to
g
r
a
m
s
.
E
ac
h
h
is
to
g
r
a
m
i
s
co
m
p
o
s
ed
o
f
1
0
2
4
b
in
s
,
w
h
er
ee
ac
h
b
in
r
ep
r
esen
t
s
aUB
M
Gau
s
s
ian
co
m
p
o
n
en
t.
(
;
,
)
=
1
[
∑
(
|
)
=
1
−
∑
(
|
)
=
1
]
(
10
)
W
ith
,
(
|
)
=
∑
(
)
=
1
(
11
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
To
w
a
r
d
s
a
n
Op
tima
l S
p
ea
ke
r
Mo
d
elin
g
in
S
p
ea
ke
r
V
erif
ica
tio
n
S
ystems
…
(
A
yo
u
b
B
o
u
z
ia
n
e
)
3661
T
h
e
v
alu
e
o
f
ea
c
h
b
in
is
d
ef
i
n
ed
as
th
e
n
u
m
b
e
r
o
f
ti
m
es
t
h
at
th
e
co
r
r
esp
o
n
d
in
g
UB
M
Gau
s
s
ia
n
co
m
p
o
n
en
t
h
as
t
h
e
m
a
x
i
m
u
m
p
o
s
ter
io
r
p
r
o
b
ab
ilit
y
o
v
er
t
h
e
f
ea
tu
r
e
v
ec
t
o
r
s
o
f
th
e
s
p
ea
k
er
.
T
h
e
s
p
ec
i
f
i
cs
o
f
t
h
e
h
is
to
g
r
a
m
co
n
s
tr
u
ct
io
n
p
r
o
ce
s
s
ar
e
s
h
o
wn
in
Alg
o
r
it
h
m
1
.
Alg
o
rit
h
m
1
T
h
e
h
is
to
g
r
a
m
co
n
s
tr
u
ctio
n
p
r
o
ce
s
s
I
np
ut:
T
h
e
f
ea
t
u
r
e
v
ec
to
r
s
X
=
{
x
1
,
x
2
,
…
,
x
N
}
o
f
th
e
s
p
ee
ch
u
tter
an
ce
,
T
h
e
u
n
i
v
er
s
al
b
ac
k
g
r
o
u
n
d
m
o
d
el
λ
UB
M
=
{
μ
UB
M
i
,
σ
UB
M
i
,
w
UB
M
i
}
/
i
∈
{
1
,
2
,
…
,
N
}
.
O
utput
:
T
h
e
co
r
r
esp
o
n
d
in
g
h
i
s
to
g
r
a
m
ℋ
o
f
th
e
s
p
ee
ch
u
tter
an
c
e.
ℋ
=
(
1
,
1024
)
=
(
|
)
ℋ
(
)
=
ℋ
(
)
+
1
E
ND
On
ce
th
e
h
i
s
to
g
r
a
m
o
f
th
e
ta
r
g
et
an
d
th
e
clai
m
ed
s
p
ea
k
er
s
ar
e
co
n
s
tr
u
cted
,
ℋ
T
an
d
ℋ
C
r
esp
ec
tiv
el
y
,
th
e
co
m
p
ar
is
o
n
b
et
w
ee
n
t
h
e
m
is
d
o
n
e
u
s
i
n
g
th
e
B
h
attac
h
ar
y
y
a
d
is
tan
ce
:
D
(
ℋ
T
,
ℋ
C
)
=
−
l
og
∑
√
ℋ
T
(
)
.
ℋ
C
(
)
1024
=
1
(
12
)
T
h
e
co
m
p
u
ted
d
is
ta
n
ce
D
(
ℋ
,
ℋ
C
)
is
t
h
en
u
s
ed
to
m
ak
e
a
d
ec
is
io
n
ab
o
u
t
th
e
ac
ce
p
ta
n
ce
o
r
t
h
e
r
ej
ec
tio
n
o
f
t
h
e
clai
m
ed
s
p
ea
k
er
.
T
h
e
o
b
tain
ed
r
esu
lts
o
f
th
e
p
er
f
o
r
m
ed
e
x
p
er
i
m
e
n
ts
,
w
h
i
le
v
ar
y
i
n
g
t
h
e
a
m
o
u
n
t o
f
tr
ai
n
i
n
g
a
n
d
test
i
n
g
s
p
ee
ch
d
ata,
ar
e
s
h
o
w
n
i
n
T
a
b
le1
.
T
ab
le
1
.
T
h
e
o
b
tain
ed
E
E
R
s
u
s
in
g
s
e
v
er
al
a
m
o
u
n
t o
f
tr
ain
in
g
an
d
test
in
g
s
p
ee
ch
d
ata.
A
mo
u
n
t
o
f
E
n
r
o
l
l
m
e
n
t
S
p
e
e
c
h
D
a
t
a
2
0
se
c
o
n
d
s
4
0
se
c
o
n
d
s
6
0
se
c
o
n
d
s
A
mo
u
n
t
o
f
T
e
st
i
n
g
S
p
e
e
c
h
D
a
t
a
4
se
c
o
n
d
s
6
.
3
8
5
.
9
0
4
.
6
4
8
se
c
o
n
d
s
5
.
8
6
5
.
3
1
3
.
9
8
First
a
n
d
f
o
r
e
m
o
s
t,
as
it
ca
n
b
e
s
ee
n
f
r
o
m
T
ab
le
1
,
th
e
o
b
tai
n
ed
r
es
u
lts
ar
e
h
i
g
h
l
y
e
n
co
u
r
a
g
in
g
.
T
h
e
lo
w
er
o
b
tain
ed
eq
u
al
er
r
o
r
r
ates,
b
ased
o
n
l
y
o
n
t
h
e
d
i
f
f
er
en
ce
i
n
h
id
d
en
ac
o
u
s
tic
cla
s
s
e
s
b
et
w
ee
n
s
p
ea
k
er
s
,
r
ef
lect
a
g
r
ea
t
d
i
f
f
er
en
ce
i
n
t
h
o
s
e
h
id
d
en
ac
o
u
s
tic
cla
s
s
e
s
b
e
t
w
ee
n
s
p
ea
k
er
s
.
A
d
d
itio
n
all
y
,
it
ap
p
ea
r
s
th
at
ea
ch
in
cr
ea
s
e
i
n
th
e
a
m
o
u
n
t o
f
tr
ain
in
g
o
r
test
in
g
s
p
ee
c
h
d
ata
is
tr
an
s
lated
i
n
to
b
etter
v
er
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
T
h
is
p
r
o
p
o
r
tio
n
al
r
elatio
n
s
h
ip
b
et
wee
n
t
h
e
a
m
o
u
n
ts
o
f
s
p
ee
ch
d
a
ta
an
d
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
s
y
s
te
m
r
ef
lects
t
h
e
f
ac
t
t
h
at
t
h
e
o
v
er
all
ac
o
u
s
tic
class
es
o
f
a
s
p
ea
k
er
ca
n
n
o
t
b
e
ass
e
m
b
led
i
n
it
s
p
r
o
n
u
n
ciat
io
n
o
f
o
n
e
o
r
t
w
o
u
tter
an
ce
s
.
4
.
3
.
Ass
ess
m
ent
o
f
t
he
P
ro
po
s
ed
G
M
M
-
P
B
M
Appro
a
ch
Co
m
pa
re
d
t
o
t
he
T
ra
di
t
io
na
l
G
M
M
-
UB
M
Appro
a
ch
T
h
e
p
er
f
o
r
m
ed
e
x
p
er
i
m
e
n
ts
i
n
th
i
s
s
ec
tio
n
at
te
m
p
t
to
a
s
s
e
s
s
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
ed
GM
M
-
P
B
M
b
ased
s
p
ea
k
er
m
o
d
elin
g
ap
p
r
o
ac
h
,
co
m
p
ar
ed
to
th
e
tr
ad
itio
n
al
GM
M
-
UB
M
b
ase
d
ap
p
r
o
ac
h
.
Hen
ce
,
v
ar
io
u
s
ex
p
er
i
m
e
n
ts
w
er
e
ca
r
r
ied
o
u
t
u
s
i
n
g
t
h
e
t
w
o
ap
p
r
o
ac
h
es
w
h
ile
v
ar
y
i
n
g
t
h
e
a
m
o
u
n
t
o
f
tr
ai
n
in
g
an
d
test
i
n
g
s
p
ee
c
h
d
ata.
T
h
e
o
b
tai
n
ed
r
esu
lt
s
ar
e
illu
s
tr
ated
in
Fi
g
u
r
e
6.
T
h
e
ex
p
er
i
m
e
n
tal
r
e
s
u
l
ts
s
h
o
w
,
ac
r
o
s
s
t
h
e
v
ar
io
u
s
a
m
o
u
n
t
s
o
f
tr
ain
in
g
a
n
d
te
s
ti
n
g
s
p
ee
ch
d
ata
t
h
at
o
u
r
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
h
as
a
ch
iev
ed
a
b
etter
v
er
if
icatio
n
p
er
f
o
r
m
a
n
ce
c
o
m
p
ar
ed
to
th
e
tr
ad
itio
n
al
UB
M
b
ased
ap
p
r
o
ac
h
.
E
v
en
in
litt
le
a
m
o
u
n
ts
o
f
tr
ai
n
i
n
g
s
p
ee
ch
d
ata,
w
h
er
e
t
h
e
s
p
ea
k
er
’
s
ac
o
u
s
tic
clas
s
es
m
a
y
n
o
t
b
e
all
p
r
esen
t,
o
u
r
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
d
em
o
n
s
tr
ate
s
its
en
h
a
n
ce
d
v
er
if
icat
io
n
p
er
f
o
r
m
a
n
ce
as
co
m
p
ar
ed
w
it
h
th
e
UB
M
b
a
s
ed
ap
p
r
o
ac
h
.
F
u
r
th
er
m
o
r
e,
w
e
ca
n
s
ee
th
at
t
h
e
o
b
tain
ed
p
er
f
o
r
m
an
ce
u
n
d
er
th
e
UB
M
b
ased
s
y
s
te
m
u
s
i
n
g
test
u
tter
a
n
ce
s
o
f
8
s
ec
o
n
d
s
w
as
o
b
tain
ed
u
n
d
er
o
u
r
p
r
o
p
o
s
ed
PB
M
b
ase
d
s
y
s
te
m
u
s
in
g
test
u
tter
an
ce
s
o
f
o
n
l
y
4
s
ec
o
n
d
s
.
Mo
r
eo
v
er
,
it
ca
n
b
e
s
ee
n
t
h
at
th
e
r
e
lati
v
e
er
r
o
r
r
ed
u
ctio
n
w
a
s
d
o
u
b
led
w
h
en
we
h
av
e
d
o
u
b
led
th
e
a
m
o
u
n
t o
f
te
s
tin
g
u
tter
an
ce
s
.
I
n
ad
d
itio
n
to
its
p
er
f
o
r
m
a
n
ce
ad
v
an
ta
g
e,
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
ca
n
s
i
g
n
i
f
ica
n
tl
y
r
ed
u
c
e
th
e
C
P
U
ti
m
e
r
eq
u
ir
ed
f
o
r
s
p
ea
k
er
v
e
r
if
icatio
n
d
u
r
i
n
g
th
e
test
i
n
g
p
h
ase
o
f
th
e
s
y
s
te
m
co
m
p
ar
ed
to
th
e
tr
ad
itio
n
a
l
s
y
s
te
m
,
s
ee
Fi
g
u
r
e
7
.
I
n
f
ac
t,
th
e
d
er
iv
atio
n
o
f
s
p
ea
k
er
s
’
m
o
d
els
f
r
o
m
p
er
s
o
n
alize
d
b
ac
k
g
r
o
u
n
d
m
o
d
els
r
ed
u
ce
s
th
e
o
r
d
er
o
f
th
eir
ad
ap
ted
m
o
d
els,
w
h
ic
h
co
n
s
eq
u
en
t
l
y
,
r
ed
u
ce
s
t
h
eir
s
to
r
ag
e
a
n
d
co
m
p
u
tat
io
n
al
co
s
t
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
6
5
5
–
3
6
6
3
3662
Fig
u
r
e
6
.
T
h
e
o
b
tain
ed
E
q
u
al
E
r
r
o
r
R
ates u
s
i
n
g
te
s
t u
t
ter
an
c
es o
f
4
s
ec
o
n
d
s
(
lef
t f
i
g
u
r
e)
an
d
8
s
ec
o
n
d
s
(
r
ig
h
t
f
i
g
u
r
e)
Fig
u
r
e
7
.
C
P
U
ti
m
e
r
eq
u
ir
ed
f
o
r
s
p
ea
k
er
v
er
if
icatio
n
u
s
in
g
t
h
e
UB
M
b
ased
ap
p
r
o
ac
h
an
d
t
h
e
P
B
M
b
ased
ap
p
r
o
ac
h
5.
CO
NCLU
SI
O
N
AND
F
U
T
U
RE
RE
SE
ARCH
DIR
E
C
T
I
O
NS
T
h
e
aim
o
f
t
h
e
p
r
esen
t
s
t
u
d
y
w
a
s
t
w
o
-
f
o
ld
.
T
h
e
f
ir
s
t
ai
m
was
to
in
v
es
tig
a
te
th
e
d
eg
r
ee
o
f
d
if
f
er
en
ce
in
h
id
d
en
ac
o
u
s
t
ic
-
cla
s
s
e
s
b
et
w
ee
n
s
p
ea
k
er
s
.
T
h
e
s
ec
o
n
d
ai
m
w
a
s
to
p
r
o
p
o
s
e
a
n
o
v
el
s
p
ea
k
er
m
o
d
elin
g
ap
p
r
o
ac
h
th
at
e
x
p
lo
its
t
h
is
d
if
f
er
en
ce
to
i
m
p
r
o
v
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
s
p
ea
k
er
r
ec
o
g
n
i
tio
n
s
y
s
te
m
s
.
T
h
e
f
i
n
d
in
g
s
o
f
t
h
e
s
t
u
d
y
r
e
v
ea
le
d
th
at
t
h
er
e
i
s
a
g
r
ea
t
d
i
f
f
er
e
n
ce
i
n
h
id
d
en
ac
o
u
s
tic
clas
s
e
s
b
et
w
ee
n
s
p
ea
k
er
s
.
A
d
d
itio
n
al
l
y
,
th
e
e
v
alu
a
tio
n
o
f
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
d
em
o
n
s
tr
ates
its
e
f
f
icien
c
y
in
ter
m
s
o
f
b
o
th
v
er
if
ica
tio
n
p
er
f
o
r
m
a
n
ce
a
n
d
co
m
p
u
tatio
n
al
co
s
t
d
u
r
i
n
g
t
h
e
v
er
if
ica
tio
n
p
h
ase
o
f
th
e
s
y
s
te
m
,
co
m
p
ar
ed
to
t
h
e
tr
ad
itio
n
al
ap
p
r
o
ac
h
.
Fu
t
u
r
e
r
esear
ch
w
ill
co
n
ce
n
tr
ate
o
n
ap
p
ly
in
g
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
w
it
h
i
n
t
h
e
h
y
b
r
id
GM
M
-
S
VM
an
d
t
h
e
i
-
v
ec
t
o
r
s
b
ased
s
y
s
te
m
s
.
RE
F
E
R
E
NC
E
S
[1
]
D.
A
.
Re
y
n
o
ld
s,
T
.
F
.
Qu
a
ti
e
ri,
a
n
d
R.
B.
Du
n
n
,
“
S
p
e
a
k
e
r
V
e
rif
ic
a
ti
o
n
Us
in
g
A
d
a
p
ted
G
a
u
ss
ian
M
ix
tu
re
M
o
d
e
ls
”
,
Dig
it
.
S
ig
n
a
l
Pro
c
e
ss
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
9
–
4
1
,
Ja
n
.
2
0
0
0
.
[2
]
D.
Re
y
n
o
ld
s,
“
Un
iv
e
rsa
l
Ba
c
k
g
r
o
u
n
d
M
o
d
e
ls
”
,
in
En
c
y
c
l
o
p
e
d
ia
o
f
B
io
me
trics
,
S
.
Z.
L
i
a
n
d
A
.
K.
Ja
in
,
Ed
s
.
S
p
rin
g
e
r
US,
2
0
1
5
,
p
p
.
1
5
4
7
–
1
5
5
0
.
[3
]
T
.
Kin
n
u
n
e
n
a
n
d
H.
L
i,
“
A
n
o
v
e
rv
ie
w
o
f
tex
t
-
in
d
e
p
e
n
d
e
n
t
s
p
e
a
k
e
r
re
c
o
g
n
it
io
n
:
F
r
o
m
f
e
a
tu
re
s
t
o
su
p
e
rv
e
c
to
rs
”
,
S
p
e
e
c
h
Co
mm
u
n
.
,
v
o
l.
5
2
,
n
o
.
1
,
p
p
.
1
2
–
4
0
,
Ja
n
.
2
0
1
0
.
[4
]
N.
De
h
a
k
a
n
d
G
.
Ch
o
ll
e
t,
“
S
u
p
p
o
rt
v
e
c
to
r
G
M
M
s
f
o
r
sp
e
a
k
e
r
v
e
rif
ica
ti
o
n
”
,
in
2
0
0
6
IEE
E
Od
y
ss
e
y
-
T
h
e
S
p
e
a
k
e
r
a
n
d
L
a
n
g
u
a
g
e
Rec
o
g
n
it
io
n
W
o
rk
sh
o
p
,
2
0
0
6
,
p
p
.
1
–
4.
2
2.
3
2.
6
2.
9
3.
2
3.
5
3.
8
20
s
e
c
onds
40
s
e
c
onds
60
s
e
c
onds
A
m
o
u
n
t
o
f
En
ro
ll
m
e
n
t
sp
e
e
c
h
d
a
ta
GMM
-
U
BM
sy
ste
m
GMM
-
P
BM
sy
ste
m
Eq
u
a
l
Err
o
r
Ra
te (E
ER)
2
2.
3
2.
6
2.
9
3.
2
3.
5
3.
8
20
s
e
c
onds
40
s
e
c
onds
60
s
e
c
onds
A
m
o
u
n
t
o
f
En
ro
ll
m
e
n
t
sp
e
e
c
h
d
a
ta
GMM
-
U
BM
sy
ste
m
GMM
-
P
BM
sy
ste
m
Eq
u
a
l
Erro
r
Ra
te
(EE
R)
0
0.
05
0.
1
0.
15
0.
2
0.
25
2
0
sec
o
n
d
s
4
0
sec
o
n
d
s
6
0
sec
o
n
d
s
A
m
o
u
n
t
o
f
En
ro
ll
m
e
n
t
sp
e
e
c
h
d
a
ta
GMM
-
U
BM
b
a
se
d
sy
ste
m
GMM
-
P
UBM
b
a
se
d
sy
ste
m
CP
U
ti
m
e
(in
se
c
o
n
d
s)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
To
w
a
r
d
s
a
n
Op
tima
l S
p
ea
ke
r
Mo
d
elin
g
in
S
p
ea
ke
r
V
erif
ica
tio
n
S
ystems
…
(
A
yo
u
b
B
o
u
z
ia
n
e
)
3663
[5
]
W
.
M
.
Ca
m
p
b
e
ll
,
D.
E.
S
tu
rim
,
a
n
d
D.
A
.
Re
y
n
o
ld
s,
“
S
u
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
s
u
sin
g
G
M
M
su
p
e
rv
e
c
to
rs
f
o
r
s
p
e
a
k
e
r
v
e
rif
i
c
a
ti
o
n
”
,
IEE
E
S
i
g
n
a
l
Pro
c
e
ss
.
L
e
tt
.
,
v
o
l.
1
3
,
n
o
.
5
,
p
p
.
3
0
8
–
3
1
1
,
m
a
i
2
0
0
6
.
[6
]
N.
De
h
a
k
e
t
a
l.
,
“
S
u
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
s
a
n
d
Jo
in
t
F
a
c
to
r
A
n
a
l
y
sis
f
o
r
sp
e
a
k
e
r
v
e
rif
i
c
a
ti
o
n
”
,
in
2
0
0
9
IE
E
E
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ac
o
u
stics
,
S
p
e
e
c
h
a
n
d
S
ig
n
a
l
Pr
o
c
e
ss
in
g
,
2
0
0
9
,
p
p
.
4
2
3
7
–
4
2
4
0
.
[7
]
N.
De
h
a
k
,
R.
De
h
a
k
,
P
.
Ke
n
n
y
,
a
n
d
P
.
D
u
m
o
u
c
h
e
l,
“
Co
m
p
a
riso
n
b
e
twe
e
n
f
a
c
to
r
a
n
a
l
y
sis
a
n
d
G
M
M
su
p
p
o
rt
v
e
c
to
r
mac
h
in
e
s f
o
r
sp
e
a
k
e
r
v
e
ri
f
ica
ti
o
n
”
,
in
Od
y
ss
e
y
,
2
0
0
8
,
p
.
9
.
[8
]
P
.
Ke
n
n
y
,
“
Jo
in
t
f
a
c
to
r
a
n
a
l
y
sis
o
f
sp
e
a
k
e
r
a
n
d
se
ss
io
n
v
a
r
i
a
b
il
it
y
:
T
h
e
o
r
y
a
n
d
a
lg
o
rit
h
m
s
”
,
CRIM
M
o
n
tre
a
lR
e
p
o
rt
CRIM
-
0
6
0
8
-
13
,
2
0
0
5
.
[9
]
P
.
Ke
n
n
y
,
“
A
s
m
a
ll
f
o
o
tp
rin
t
i
-
v
e
c
to
r
e
x
trac
to
r
”
,
in
O
d
y
ss
e
y
,
2
0
1
2
,
p
p
.
1
–
6.
[1
0
]
D.
W
u
,
J.
Ca
o
,
a
n
d
H.
W
a
n
g
,
“
S
p
e
a
k
e
r
Re
c
o
g
n
it
io
n
Ba
se
d
o
n
i
-
v
e
c
to
r
a
n
d
Im
p
ro
v
e
d
L
o
c
a
l
P
re
se
r
v
in
g
P
r
o
je
c
ti
o
n
”
,
In
d
o
n
e
s.
J
.
E
lec
tr.
En
g
.
C
o
mp
u
t.
S
c
i.
,
v
o
l
.
1
2
,
n
o
.
6
,
p
p
.
4
2
9
9
–
4
3
0
5
,
Ju
n
.
2
0
1
4
.
[1
1
]
H.
Be
ig
i,
Fu
n
d
a
me
n
ta
ls
o
f
S
p
e
a
k
e
r R
e
c
o
g
n
it
i
o
n
.
S
p
rin
g
e
r,
2
0
1
1
.
[1
2
]
D.
A
.
Re
y
n
o
ld
s,
“
Co
m
p
a
riso
n
o
f
b
a
c
k
g
ro
u
n
d
n
o
rm
a
li
z
a
ti
o
n
m
e
th
o
d
s
f
o
r
tex
t
-
in
d
e
p
e
n
d
e
n
t
sp
e
a
k
e
r
v
e
ri
f
ica
ti
o
n”
,
i
n
Eu
ro
sp
e
e
c
h
,
1
9
9
7
.
[1
3
]
D.
Re
y
n
o
ld
s,
“
Ga
u
ss
ian
M
ix
tu
re
M
o
d
e
ls
”
,
in
En
c
y
c
lo
p
e
d
ia
o
f
Bi
o
me
trics
,
S
.
Z.
L
i
a
n
d
A
.
K.
Ja
i
n
,
Ed
s.
Bo
st
o
n
,
M
A
:
S
p
rin
g
e
r
US,
2
0
1
5
,
p
p
.
8
2
7
–
8
3
2
.
[1
4
]
A
.
Ro
z
i,
D.
W
a
n
g
,
Z.
Zh
a
n
g
,
a
n
d
T
.
F
.
Z
h
e
n
g
,
“
A
n
o
p
e
n
/f
re
e
d
a
tab
a
se
a
n
d
Be
n
c
h
m
a
rk
f
o
r
U
y
g
h
u
r
sp
e
a
k
e
r
re
c
o
g
n
it
io
n
”
,
in
Or
ien
ta
l
COCO
S
DA
h
e
ld
jo
i
n
tl
y
wit
h
2
0
1
5
C
o
n
fe
re
n
c
e
o
n
Asi
a
n
S
p
o
k
e
n
L
a
n
g
u
a
g
e
Res
e
a
rc
h
a
n
d
Eva
lu
a
ti
o
n
(
O
-
COCO
S
DA/CA
S
L
RE
),
2
0
1
5
I
n
ter
n
a
t
io
n
a
l
Co
n
fer
e
n
c
e
,
2
0
1
5
,
p
p
.
8
1
–
8
5
.
[1
5
]
B.
Ay
o
u
b
,
K.
Ja
m
a
l,
a
n
d
Z.
A
rs
a
lan
e
,
“
A
n
a
n
a
l
y
sis
a
n
d
c
o
m
p
a
r
a
ti
v
e
e
v
a
lu
a
ti
o
n
o
f
M
F
CC
v
a
ria
n
ts
f
o
r
sp
e
a
k
e
r
id
e
n
ti
f
ica
ti
o
n
o
v
e
r
Vo
I
P
n
e
tw
o
rk
s”
,
in
2
0
1
5
W
o
rl
d
Co
n
g
re
ss
o
n
In
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
a
n
d
Co
m
p
u
te
r
Ap
p
li
c
a
ti
o
n
s C
o
n
g
re
ss
(W
CIT
CA)
,
2
0
1
5
,
p
p
.
1
–
6.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
Ay
o
u
b
Bo
u
z
ian
e
re
c
e
iv
e
d
th
e
M
.
S
c
.
d
e
g
re
e
in
in
telli
g
e
n
t
s
y
ste
m
s
a
n
d
n
e
tw
o
rk
s
f
ro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s
a
n
d
T
e
c
h
n
o
l
o
g
ies
,
F
e
z
,
M
o
r
o
c
c
o
,
i
n
2
0
1
2
.
He
is
c
u
rre
n
tl
y
p
u
rsu
in
g
t
h
e
P
h
.
D.
d
e
g
re
e
in
t
h
e
i
n
telli
g
e
n
t
sy
ste
m
s
a
n
d
a
p
p
li
c
a
ti
o
n
s
lab
o
ra
t
o
ry
,
S
id
i
M
o
h
a
m
m
e
d
b
e
n
A
b
d
u
ll
a
h
Un
iv
e
rsity
,
M
o
ro
c
c
o
.
His
re
se
a
rc
h
in
tere
sts
c
o
v
e
r
sig
n
a
l
p
ro
c
e
ss
in
g
a
n
d
m
a
c
h
in
e
lea
rn
in
g
,
m
o
stl
y
w
it
h
a
p
p
li
c
a
ti
o
n
s
in
a
u
to
m
a
ti
c
sp
e
a
k
e
r
re
c
o
g
n
it
io
n
.
In
p
a
ra
ll
e
l
w
it
h
h
is
re
se
a
rc
h
a
c
ti
v
it
ies
,
h
e
tea
c
h
e
s
u
n
d
e
rg
ra
d
u
a
te
lev
e
l
c
o
u
rse
s
in
c
o
m
p
u
ter
sc
ien
c
e
,
a
t
th
e
F
a
c
u
lt
y
o
f
S
c
ie
n
c
e
s
a
n
d
Tec
h
n
o
lo
g
ies
,
F
e
z
.
A
d
d
it
io
n
a
ll
y
,
He
is
a
m
e
m
b
e
r
o
f
IEE
E
S
ig
n
a
l
P
r
o
c
e
ss
in
g
S
o
c
iety
,
IEE
E
Co
m
p
u
tatio
n
a
l
In
telli
g
e
n
c
e
S
o
c
iety
,
IEE
E
Co
m
p
u
ter
S
o
c
iety
a
n
d
th
e
In
tern
a
ti
o
n
a
l
S
p
e
e
c
h
a
n
d
Co
m
m
u
n
ica
ti
o
n
A
s
so
c
iatio
n
(IS
CA
).
Ja
m
a
l
Kh
a
rro
u
b
i
h
a
s
h
is
B.
S
c
.
in
Co
m
p
u
ter
S
c
ien
c
e
f
ro
m
S
id
i
M
o
h
a
m
e
d
Be
n
A
b
d
e
ll
a
h
Un
iv
e
rsit
y
(F
e
z
-
M
o
ro
c
c
o
)
in
1
9
9
6
.
T
w
o
y
e
a
rs
a
f
ter,
h
e
g
o
t
h
is
p
o
stg
ra
d
u
a
te
d
e
g
re
e
in
th
e
d
o
m
a
in
o
f
A
rti
f
icia
l
In
telli
g
e
n
c
e
f
ro
m
G
a
li
lee
’s
In
stit
u
te
-
P
a
r
is
X
III
Un
iv
e
rsity
.
In
2
0
0
2
,
He
re
c
e
i
v
e
d
h
is
P
h
.
D.
d
e
g
re
e
in
a
u
to
m
a
ti
c
sp
e
a
k
e
r
r
e
c
o
g
n
it
io
n
sy
st
e
m
s
f
ro
m
Tele
c
o
m
P
a
ris
T
e
c
h
(Eco
le
Na
ti
o
n
a
le
S
u
p
é
rieu
re
d
e
s
T
é
lé
c
o
m
m
u
n
ica
ti
o
n
s
d
e
P
a
ris
-
F
ra
n
c
e
)”
.
S
in
c
e
Ja
n
u
a
r
y
2
0
0
3
,
h
e
isa
n
a
ss
o
c
iate
p
ro
f
e
ss
o
r
in
th
e
D
e
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
t
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
lo
g
y
.
In
2
0
0
8
,
h
e
re
c
e
iv
e
d
h
is
HD
R
d
ip
l
o
m
a
(Ha
b
il
it
a
ti
o
n
to
c
o
n
d
u
c
t
re
se
a
rc
h
).
A
d
d
it
io
n
a
ll
y
,
He
isc
u
rre
n
tl
y
th
e
c
o
o
r
d
in
a
t
o
r
o
f
th
e
M
a
ste
r
o
f
In
tell
ig
e
n
t
S
y
ste
m
s
a
n
d
Ne
t
w
o
rk
s.
M
o
re
o
v
e
r,
h
e
is
th
e
a
u
th
o
r
o
f
m
o
r
e
th
a
n
th
ir
ty
p
u
b
li
c
a
ti
o
n
s in
p
e
e
r
-
re
v
ie
w
e
d
sc
ien
ti
f
ic j
o
u
rn
a
ls
&
c
o
n
f
e
re
n
c
e
p
ro
c
e
e
d
in
g
s.
His res
e
a
rc
h
in
tere
sts a
re
f
o
c
u
se
d
o
n
sig
n
a
l
a
n
d
im
a
g
e
p
ro
c
e
ss
in
g
,
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
e
tc.
A
rs
a
lan
e
Zarg
h
il
i
is
a
Do
c
to
r
o
f
S
c
ien
c
e
f
ro
m
S
id
i
M
o
h
a
m
e
d
Be
n
A
b
d
e
ll
a
h
Un
iv
e
rsity
(F
e
z
-
M
o
ro
c
c
o
)
.
He
re
c
e
iv
e
d
h
is
P
h
.
D.
in
2
0
0
1
a
n
d
j
o
in
e
d
t
h
e
sa
m
e
Un
iv
e
rsit
y
in
2
0
0
2
a
s
P
r
o
f
e
ss
o
r
a
t
th
e
c
o
m
p
u
ter
sc
ien
c
e
d
e
p
a
rt
m
e
n
t
o
f
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
(F
S
T
).
In
2
0
0
7
h
e
w
a
s
h
e
a
d
o
f
th
e
c
o
m
p
u
ter
sc
ien
c
e
s
d
e
p
a
rtm
e
n
t
a
n
d
c
h
a
ir
o
f
th
e
S
o
f
twa
re
Qu
a
li
t
y
M
a
s
ter
in
t
h
e
FST
-
F
e
z
.
He
lec
tu
re
s
P
ro
g
ra
m
m
i
n
g
,
Distrib
u
ted
,
c
o
m
p
il
a
ti
o
n
a
n
d
In
f
o
rm
a
ti
o
n
p
r
o
c
e
ss
in
g
,
f
o
r
b
o
t
h
u
n
d
e
r
g
ra
d
u
a
te
a
n
d
m
a
ste
rle
v
e
ls.
In
2
0
0
8
h
e
o
b
tain
e
d
h
is
HD
R
in
in
f
o
rm
a
ti
o
n
p
ro
c
e
ss
in
g
.
In
2
0
1
1
,
h
e
is
th
e
c
o
-
f
o
u
n
d
e
r
a
n
d
th
e
h
e
a
d
o
f
th
e
Lab
o
ra
to
r
y
o
f
In
telli
g
e
n
t
S
y
ste
m
s
a
n
d
A
p
p
li
c
a
ti
o
n
s
in
th
e
F
S
T
o
f
F
e
z
.
He
is
a
m
e
m
b
e
r
o
f
th
e
ste
e
rin
g
c
o
m
m
it
tee
o
f
th
e
d
e
p
a
rtme
n
t
o
f
c
o
m
p
u
ter
sc
ien
c
e
s
a
n
d
w
a
s
a
m
e
m
b
e
r
o
f
th
e
fa
c
u
lt
y
b
o
a
rd
.
He
is
a
lso
IEE
E
m
e
m
b
e
r
sin
c
e
2
0
1
1
.
His
m
a
in
re
se
a
rc
h
is
a
b
o
u
t
p
a
tt
e
rn
re
c
o
g
n
it
io
n
,
im
a
g
e
in
d
e
x
in
g
a
n
d
re
tri
e
v
a
l
s
y
ste
m
s
in
c
u
lt
u
ra
l
h
e
rit
a
g
e
,
b
io
m
e
tri
c
,
e
tc.
Evaluation Warning : The document was created with Spire.PDF for Python.