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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
,
p
p
.
4
8
6
9
~
4
8
8
2
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
13
.i
4
.
p
p
4
8
6
9
-
488
2
4869
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Integ
ra
ting g
a
it
a
nd speech
dyna
mi
cs method
o
lo
g
ies
for
enha
nced st
utt
eri
ng
det
ec
tion a
cro
ss
divers
e data
set
s
Ra
v
ik
ira
n R
edda
pp
a
Reddy
1,
2
,
Sa
ntho
s
h K
um
a
r
G
a
ng
a
dh
a
ra
ih
1
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s
a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
,
Ea
s
t
W
e
st
C
o
l
l
e
g
e
o
f
En
g
i
n
e
e
r
i
n
g
,
V
i
sv
e
sv
a
r
a
y
a
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
,
B
e
l
a
g
a
v
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
C
o
m
mu
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
S
J
C
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
C
h
i
c
k
b
a
l
l
a
p
u
r
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
2
2
,
2
0
2
3
R
ev
is
ed
Mar
6
,
2
0
2
4
Acc
ep
ted
Mar
15
,
2
0
2
4
S
tu
tt
e
ri
n
g
m
a
n
ifes
ts
a
s
i
n
v
o
l
u
n
ta
ry
i
n
terru
p
ti
o
n
s
in
t
h
e
fl
u
e
n
c
y
o
f
sp
e
e
c
h
,
o
ften
in
v
o
lv
i
n
g
re
p
e
ti
t
io
n
s,
p
ro
lo
n
g
a
ti
o
n
s,
o
r
b
lo
c
k
s
o
f
so
u
n
d
s
o
r
sy
ll
a
b
les
.
Th
e
se
d
isru
p
ti
o
n
s
c
a
n
sig
n
ifi
c
a
n
tl
y
imp
a
c
t
e
ffe
c
ti
v
e
c
o
m
m
u
n
ic
a
ti
o
n
a
n
d
p
sy
c
h
o
so
c
ial
we
ll
-
b
e
in
g
.
T
h
is
r
e
se
a
rc
h
in
tro
d
u
c
e
s
a
c
o
m
p
re
h
e
n
si
v
e
sy
ste
m
fo
r
sp
e
e
c
h
im
p
a
irme
n
t
d
e
tec
ti
o
n
a
n
d
g
a
it
a
n
a
ly
sis.
S
p
e
e
c
h
im
p
a
irme
n
t,
wit
h
a
p
rima
ry
fo
c
u
s
o
n
sta
m
m
e
r
re
c
o
g
n
it
io
n
,
p
re
se
n
ts
a
m
u
lt
ifac
e
ted
c
h
a
ll
e
n
g
e
in
th
e
fiel
d
o
f
s
p
e
e
c
h
p
r
o
c
e
ss
in
g
.
S
t
a
m
m
e
rs
c
a
n
m
a
n
ifes
t
in
v
a
rio
u
s
f
o
rm
s
a
n
d
d
e
tec
ti
n
g
t
h
e
m
a
c
c
u
ra
tely
is
a
c
o
m
p
lex
tas
k
.
Ou
r
p
ro
p
o
se
d
m
e
th
o
d
o
l
o
g
y
re
v
o
lv
e
s
a
ro
u
n
d
th
e
d
e
v
e
l
o
p
m
e
n
t
o
f
S
t
En
se
m
b
leN
e
t,
a
n
e
u
ra
l
n
e
two
rk
d
e
sig
n
e
d
to
lea
rn
s
p
e
c
tral
fe
a
tu
re
s
a
t
th
e
fra
m
e
lev
e
l,
e
n
a
b
li
n
g
p
re
c
ise
a
n
d
e
fficie
n
t
id
e
n
ti
fica
ti
o
n
o
f
s
p
e
e
c
h
imp
e
d
ime
n
ts.
Ad
d
it
io
n
a
ll
y
,
we
e
x
ten
d
o
u
r
sy
ste
m
'
s
c
a
p
a
b
il
it
ies
to
th
e
d
o
m
a
in
o
f
g
a
it
a
n
a
ly
sis,
lev
e
ra
g
i
n
g
a
n
o
v
e
l
a
d
a
p
ti
v
e
g
ra
p
h
to
p
o
l
o
g
y
c
o
n
v
o
lu
ti
o
n
n
e
two
rk
(AG
T
-
Co
n
v
Ne
t)
fo
r
sk
e
leta
l
m
o
ti
o
n
a
n
d
v
is
u
a
ll
y
e
n
h
a
n
c
e
d
to
p
o
l
o
g
ica
l
lea
rn
i
n
g
to
a
d
a
p
t
to
d
i
v
e
rse
v
isu
a
l
e
n
v
iro
n
m
e
n
ts
a
n
d
e
n
h
a
n
c
e
th
e
re
c
o
g
n
i
ti
o
n
o
f
g
a
it
p
a
tt
e
rn
s.
T
h
is
re
se
a
rc
h
n
o
t
o
n
l
y
c
o
n
tri
b
u
tes
to
t
h
e
field
o
f
sp
e
e
c
h
th
e
ra
p
y
b
u
t
a
ls
o
o
ffe
rs
p
o
ten
ti
a
l
a
p
p
li
c
a
ti
o
n
s in
h
e
a
lt
h
c
a
re
a
n
d
m
o
ti
o
n
a
n
a
ly
sis.
K
ey
w
o
r
d
s
:
Ad
ap
tiv
e
g
r
a
p
h
to
p
o
lo
g
y
co
n
v
o
l
u
tio
n
n
etwo
r
k
Hea
lth
ca
r
e
Neu
r
al
n
etwo
r
k
Sp
ee
ch
im
p
air
m
en
t
Sp
ee
ch
th
er
ap
y
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
:
R
av
ik
ir
an
R
ed
d
ap
p
a
R
ed
d
y
Dep
ar
tm
en
t o
f
E
lectr
o
n
ics
an
d
C
o
m
m
u
n
icatio
n
,
E
ast W
est
C
o
lleg
e
o
f
E
n
g
in
ee
r
i
n
g
Vis
v
esv
ar
ay
a
T
ec
h
n
o
l
o
g
ical
Un
iv
er
s
ity
B
an
g
alo
r
e
5
6
0
0
6
4
,
I
n
d
ia
E
m
ail: r
av
ik
ir
an
r
_
1
2
@
r
ed
if
f
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Stu
tter
in
g
is
a
s
p
ee
ch
d
is
o
r
d
er
th
at
ar
is
es
f
r
o
m
n
eu
r
o
d
ev
elo
p
m
en
tal
f
ac
to
r
s
an
d
is
ch
ar
ac
ter
ized
b
y
th
e
im
p
air
m
en
t
o
f
s
p
ee
c
h
s
en
s
o
r
i
m
o
to
r
f
u
n
ctio
n
s
.
T
h
e
co
n
d
itio
n
is
d
is
tin
g
u
is
h
ed
b
y
th
e
o
cc
u
r
r
en
ce
o
f
in
v
o
lu
n
ta
r
y
u
tter
an
ce
s
,
in
ter
jectio
n
s
(
s
u
ch
as
th
e
in
cl
u
s
io
n
o
f
s
o
u
n
d
s
li
k
e
"u
h
"
o
r
"u
h
m
")
,
an
d
c
o
r
e
b
eh
av
io
r
s
(
b
l
o
ck
s
,
r
ep
etitio
n
s
,
an
d
p
r
o
l
o
n
g
atio
n
s
,
wh
ich
ar
e
aty
p
ical
elo
n
g
atio
n
s
o
f
s
p
ee
ch
s
o
u
n
d
s
eg
m
en
ts
)
[
1
]
.
B
ased
o
n
r
esear
ch
f
in
d
in
g
s
,
in
d
iv
id
u
als
wh
o
en
c
o
u
n
ter
s
tu
tter
in
g
,
co
m
m
o
n
l
y
k
n
o
wn
as
p
eo
p
le
wh
o
s
tam
m
er
(
PW
S),
en
co
u
n
ter
v
ar
io
u
s
d
if
f
icu
lties
in
th
eir
in
t
er
p
er
s
o
n
al
an
d
o
cc
u
p
atio
n
al
e
n
g
ag
em
e
n
ts
.
Stu
tter
in
g
d
etec
ti
o
n
ca
n
b
e
u
tili
ze
d
to
en
h
an
ce
th
e
e
f
f
icien
c
y
o
f
a
u
to
m
ated
s
p
ee
ch
r
ec
o
g
n
itio
n
(
ASR
)
s
y
s
tem
s
f
o
r
in
d
iv
id
u
als with
s
tu
tter
in
g
(
PW
S).
T
h
e
p
u
r
p
o
s
e
o
f
t
h
is
en
h
an
ce
m
en
t
is
to
im
p
r
o
v
e
t
h
e
u
s
ab
ilit
y
o
f
v
o
ice
ass
is
tan
ts
s
u
ch
as
Al
ex
a,
Siri,
an
d
o
th
er
s
im
ilar
p
latf
o
r
m
s
.
Ad
d
itio
n
ally
,
a
d
is
ce
r
n
ib
le
tr
en
d
h
as
b
ee
n
o
b
s
er
v
ed
wh
er
ein
u
s
er
s
ar
e
in
c
r
ea
s
in
g
ly
en
g
ag
in
g
with
v
o
ice
ass
is
tan
ts
.
Ho
wev
er
,
it is
c
o
m
m
o
n
f
o
r
th
ese
in
d
iv
id
u
als to
o
f
ten
o
v
e
r
lo
o
k
an
d
e
n
co
u
n
ter
ch
allen
g
es
wh
en
it c
o
m
es to
co
m
p
r
e
h
en
d
i
n
g
s
p
ee
ch
th
at
d
is
p
lay
s
s
tu
tter
in
g
.
T
h
e
d
iag
n
o
s
is
o
f
s
p
in
al
d
y
s
r
ap
h
is
m
ty
p
ically
in
v
o
lv
es
th
e
u
tili
za
tio
n
o
f
s
p
ec
if
ic
h
ea
r
i
n
g
a
n
d
b
r
ain
s
ca
n
test
s
.
T
h
e
im
p
lem
en
tatio
n
o
f
t
h
e
s
p
ee
ch
d
is
f
lu
en
c
y
a
p
p
r
o
ac
h
in
cu
r
s
s
ig
n
if
ican
t
c
o
s
ts
an
d
d
e
m
an
d
s
a
s
u
b
s
tan
tial
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
6
9
-
488
2
4870
co
m
m
itm
en
t
o
f
ef
f
o
r
t
f
r
o
m
s
p
ee
ch
th
er
ap
is
ts
[
2
]
.
Un
co
n
tr
o
lled
v
o
ca
lizatio
n
s
ar
e
p
er
ce
p
tib
le
m
an
if
estatio
n
s
th
at
co
n
tr
ib
u
te
to
th
e
d
if
f
e
r
en
t
iatio
n
o
f
v
ar
io
u
s
ty
p
es
o
f
s
tu
tt
er
in
g
.
A
m
ac
h
in
e
lear
n
in
g
m
o
d
el
was
em
p
lo
y
ed
b
y
s
ev
er
al
in
d
iv
id
u
als to
an
aly
ze
au
d
ito
r
y
s
ig
n
als p
r
esen
t in
s
p
ee
ch
af
f
ec
ted
b
y
s
tu
tter
in
g
.
T
h
e
p
r
im
ar
y
aim
o
f
r
esear
ch
o
n
ev
er
y
d
ay
b
e
h
av
io
r
is
to
d
is
t
in
g
u
is
h
an
d
class
if
y
d
is
tin
ct
b
eh
av
io
r
s
,
in
clu
d
in
g
b
u
t
n
o
t
lim
ited
to
walk
in
g
,
s
tan
d
i
n
g
,
s
itti
n
g
,
r
u
n
n
in
g
,
an
d
v
ar
i
o
u
s
o
t
h
er
s
[
3
]
.
T
h
e
ass
ess
m
en
t
o
f
t
h
e
im
p
ac
t
o
f
ac
tio
n
co
n
v
er
s
io
n
o
n
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
s
eg
m
en
tatio
n
r
ec
o
g
n
itio
n
s
y
s
tem
h
o
ld
s
s
ig
n
if
ican
ce
,
d
esp
ite
th
e
s
y
s
tem
n
o
t
b
ein
g
ex
p
licitly
d
esig
n
ed
to
h
an
d
le
tr
an
s
itio
n
al
ac
tiv
ities
.
Mo
r
eo
v
er
,
th
e
r
o
u
tin
e
ac
tiv
ities
p
er
f
o
r
m
ed
b
y
in
d
iv
i
d
u
als
o
f
ten
in
v
o
lv
e
m
u
ltip
le
s
h
if
ts
b
etwe
en
v
ar
i
o
u
s
s
tates
o
f
ac
t
io
n
,
p
r
esen
tin
g
a
co
m
b
in
atio
n
o
f
in
tr
icac
y
an
d
co
n
tin
u
ity
.
E
v
er
y
d
ay
task
s
o
f
ten
in
v
o
l
v
e
co
m
p
lex
p
r
o
ce
s
s
es,
wh
er
e
th
e
c
o
m
p
lex
ity
is
d
e
m
o
n
s
tr
ated
b
y
th
e
tr
an
s
itio
n
al
s
tep
s
th
at
o
cc
u
r
d
u
r
in
g
s
tate
ch
an
g
es.
T
h
e
p
r
o
ce
s
s
o
f
g
ait
d
etec
tio
n
is
p
r
ed
o
m
in
an
tly
d
ep
en
d
e
n
t
o
n
m
ac
h
in
e
lear
n
in
g
an
d
c
o
m
p
u
ter
v
is
io
n
tech
n
iq
u
es.
T
h
ese
tech
n
iq
u
es
ar
e
em
p
l
o
y
ed
to
p
r
ec
is
ely
id
en
tify
in
d
iv
id
u
als
b
y
a
n
aly
zin
g
th
ei
r
walk
in
g
p
atter
n
s
.
T
h
e
m
ain
o
b
jectiv
e
o
f
t
h
e
m
u
ltid
is
cip
lin
ar
y
f
ield
o
f
g
ait
r
ec
o
g
n
itio
n
is
to
o
b
tain
a
co
m
p
r
eh
en
s
iv
e
u
n
d
er
s
tan
d
in
g
o
f
an
in
d
i
v
id
u
al'
s
b
eh
av
io
r
b
y
an
aly
zin
g
t
h
e
g
ait
p
atter
n
s
[
4
]
.
T
h
is
co
m
b
in
es th
e
p
r
in
cip
les a
n
d
tech
n
i
q
u
es o
f
b
io
m
ec
h
an
ic
s
,
p
s
y
ch
o
lo
g
y
,
an
d
ad
v
an
ce
d
tech
n
o
lo
g
y
to
ac
h
iev
e
its
g
o
al.
T
h
e
ass
ess
m
en
t
o
f
an
in
d
iv
i
d
u
al'
s
s
tr
id
e
ca
n
p
r
o
v
id
e
v
alu
a
b
le
in
s
ig
h
ts
in
to
th
eir
co
g
n
itiv
e
an
d
em
o
tio
n
al
s
tate,
p
h
y
s
ical
well
-
b
ein
g
,
a
n
d
lev
el
o
f
f
atig
u
e
o
r
s
tr
ess
.
C
o
n
tem
p
o
r
a
r
y
co
m
p
u
ter
v
is
io
n
s
y
s
tem
s
an
d
m
ac
h
in
e
lear
n
i
n
g
alg
o
r
ith
m
s
d
em
o
n
s
tr
ate
th
e
ca
p
ab
ilit
y
to
p
r
ec
is
ely
d
etec
t
an
d
an
aly
z
e
ev
en
th
e
s
u
b
tles
t
in
tr
icac
ies
in
m
o
tio
n
[
5
]
.
T
h
e
ad
v
an
ce
m
e
n
ts
ac
h
iev
ed
in
th
i
s
f
ield
h
av
e
cr
ea
ted
n
ew
p
o
s
s
ib
ilit
ies
f
o
r
d
iv
er
s
e
ap
p
licatio
n
s
in
th
e
f
ield
s
o
f
s
ec
u
r
ity
,
h
ea
lth
ca
r
e,
an
d
p
s
y
ch
o
lo
g
ical
well
-
b
ein
g
.
T
h
e
tech
n
o
lo
g
y
,
h
o
wev
e
r
,
g
iv
es
r
is
e
to
p
r
iv
ac
y
c
o
n
ce
r
n
s
as
it
is
s
u
s
ce
p
tib
le
to
f
lu
ctu
ati
o
n
s
in
a
cc
u
r
ac
y
ca
u
s
ed
b
y
v
ar
io
u
s
en
v
ir
o
n
m
en
tal
f
ac
t
o
r
s
.
T
h
e
p
r
im
a
r
y
g
o
al
o
f
s
tu
d
y
in
g
g
ait
is
to
an
aly
ze
a
n
d
e
v
alu
ate
m
o
v
em
en
t
p
atter
n
s
.
I
t
is
im
p
o
r
tan
t
to
co
n
s
id
er
th
e
p
o
ten
tial
f
o
r
ex
p
lo
r
in
g
th
e
co
r
r
elatio
n
b
etwe
en
in
ter
c
o
n
n
ec
t
ed
m
o
t
o
r
p
atter
n
s
,
s
u
ch
as
g
ait
,
an
d
th
eir
ab
ilit
y
to
p
r
o
v
id
e
in
d
ir
ec
t
in
s
ig
h
ts
in
to
v
ar
io
u
s
co
n
d
itio
n
s
,
in
clu
d
i
n
g
s
tu
tter
in
g
.
T
h
is
ca
n
b
e
ac
co
m
p
li
s
h
ed
b
y
id
en
tify
in
g
ab
n
o
r
m
alities
with
in
th
e
m
o
t
o
r
s
y
s
tem
.
Gait
id
en
tific
atio
n
an
d
s
tu
tter
in
g
d
etec
tio
n
r
e
p
r
e
s
en
t
two
p
r
o
m
in
e
n
t
d
o
m
ain
s
k
n
o
wn
as
b
io
m
etr
ic
an
d
b
eh
av
io
r
al
a
n
aly
s
is
.
T
h
es
e
d
o
m
ai
n
s
o
f
f
er
v
alu
ab
le
in
s
ig
h
ts
in
to
t
h
e
in
tr
icate
in
ter
p
lay
b
etwe
en
p
h
y
s
io
lo
g
ical
an
d
n
eu
r
o
lo
g
ical
s
y
s
tem
s
.
Du
r
in
g
th
e
f
ir
s
t
ass
ess
m
en
t,
o
n
e
m
ig
h
t
o
b
s
er
v
e
a
d
is
ce
r
n
ib
le
d
if
f
e
r
en
ce
b
etwe
en
an
in
d
iv
id
u
al'
s
w
alk
in
g
s
ty
le
an
d
th
e
f
lu
en
c
y
o
f
th
eir
s
p
o
k
en
co
m
m
u
n
icatio
n
.
Up
o
n
co
n
d
u
ctin
g
a
m
o
r
e
d
et
ailed
an
aly
s
is
,
it
b
ec
o
m
es
ap
p
ar
en
t
th
at
th
er
e
e
x
is
t
p
o
ten
tial
in
ter
s
ec
tio
n
s
th
at
n
ec
ess
itate
ad
d
itio
n
al
in
v
esti
g
atio
n
[
6
]
.
T
h
e
o
cc
u
r
r
e
n
ce
o
f
s
tu
tter
in
g
g
o
es b
ey
o
n
d
m
er
e
s
p
ee
ch
er
r
o
r
s
,
as i
t c
an
p
r
o
v
id
e
v
alu
ab
le
in
s
ig
h
ts
in
to
an
in
d
iv
id
u
al'
s
b
r
o
ad
er
n
eu
r
o
l
o
g
ical
an
d
em
o
tio
n
al
s
tate.
T
h
e
o
cc
u
r
r
e
n
ce
o
f
th
e
c
o
n
d
itio
n
o
f
ten
o
cc
u
r
s
d
u
e
to
em
o
tio
n
al
tr
ig
g
er
s
,
s
u
ch
as
s
tr
ess
o
r
an
x
iet
y
.
J
u
s
t
as
em
o
tio
n
s
an
d
n
eu
r
o
lo
g
ical
d
is
o
r
d
er
s
h
av
e
th
e
p
o
ten
tial
to
im
p
ac
t
b
eh
av
io
r
,
g
ait
is
a
co
m
p
lex
m
o
to
r
ac
tio
n
with
m
u
ltip
l
e
f
ac
ets.
T
h
is
in
q
u
ir
y
s
ee
k
s
to
d
eter
m
in
e
if
aty
p
ical
walk
in
g
p
atter
n
s
ca
n
s
er
v
e
as
an
in
d
icato
r
o
r
p
r
e
d
icto
r
o
f
s
tu
tter
in
g
o
cc
u
r
r
en
ce
s
.
T
h
e
u
n
d
er
ly
in
g
f
o
u
n
d
atio
n
o
f
th
is
r
elatio
n
s
h
ip
is
g
r
o
u
n
d
ed
i
n
th
e
ac
k
n
o
wled
g
m
en
t
th
at
in
tr
icate
n
eu
r
al
n
etwo
r
k
s
co
n
tr
o
l
b
o
th
s
p
ee
ch
a
n
d
m
o
v
em
en
t.
Hen
ce
,
th
e
id
e
n
tific
atio
n
o
f
an
y
d
is
r
u
p
tio
n
o
r
d
e
v
iatio
n
with
in
a
n
y
o
f
th
ese
d
o
m
ain
s
m
ay
in
d
icate
po
ten
tial
is
s
u
es
with
in
th
e
r
em
ain
in
g
d
o
m
ain
s
[
2
]
.
T
h
e
ap
p
licatio
n
o
f
a
co
m
p
r
eh
e
n
s
iv
e
m
eth
o
d
o
lo
g
y
th
at
in
teg
r
ates
d
ata
f
r
o
m
s
p
ee
ch
p
atter
n
s
an
d
g
ait
an
al
y
s
is
allo
ws
r
esear
ch
er
s
to
u
n
co
v
er
c
o
n
ce
aled
p
atter
n
s
th
at
m
ay
n
o
t
b
e
im
m
e
d
iately
ap
p
a
r
en
t
wh
en
a
n
aly
zin
g
ea
c
h
d
o
m
a
in
in
d
ep
e
n
d
en
tly
.
T
h
e
i
n
teg
r
ati
o
n
o
f
a
m
u
ltim
o
d
al
s
tr
ateg
y
s
h
o
ws
p
o
ten
tial
in
f
ac
ilit
atin
g
th
e
p
r
o
g
r
ess
o
f
in
n
o
v
ativ
e
m
ed
ical
ap
p
r
o
ac
h
es,
co
m
p
r
eh
en
s
iv
e
p
atien
t
m
o
n
ito
r
in
g
,
a
n
d
ea
r
ly
d
etec
tio
n
s
y
s
tem
s
.
T
h
e
in
teg
r
atio
n
o
f
g
ait
id
en
tific
atio
n
an
d
s
tu
tter
i
n
g
d
etec
tio
n
h
o
ld
s
s
ig
n
if
ican
t
p
o
ten
tial
f
o
r
ad
v
a
n
cin
g
th
e
u
n
d
e
r
s
tan
d
in
g
o
f
h
u
m
an
b
e
h
av
io
r
an
d
th
e
u
n
d
e
r
ly
in
g
n
e
u
r
o
l
o
g
ical
p
r
o
ce
s
s
es
th
at
g
o
v
er
n
it.
T
h
e
o
n
g
o
in
g
m
u
ltid
is
cip
lin
ar
y
in
v
esti
g
atio
n
is
an
ticip
ated
to
g
en
er
ate
s
ig
n
if
ican
t
in
s
ig
h
ts
in
to
th
ese
p
ar
ticu
lar
d
o
m
ain
s
.
T
h
e
ex
am
in
atio
n
o
f
b
o
d
ily
m
o
v
em
en
ts
,
s
p
ec
if
ically
g
ait,
o
f
f
e
r
s
s
ig
n
if
ican
t
in
s
ig
h
ts
in
to
an
in
d
iv
id
u
al'
s
em
o
tio
n
al
an
d
p
s
y
ch
o
lo
g
ical
co
n
d
itio
n
s
.
T
h
e
co
n
v
er
g
e
n
ce
o
f
tech
n
o
lo
g
y
an
d
p
s
y
c
h
o
lo
g
y
in
th
e
f
ield
o
f
g
ait
r
ec
o
g
n
itio
n
p
r
o
v
id
es
a
u
n
i
q
u
e
p
o
in
t
f
o
r
u
n
d
er
s
tan
d
in
g
h
u
m
an
b
eh
av
i
o
r
.
Stu
tter
in
g
,
r
elate
d
to
o
th
er
m
ed
ical
co
n
d
itio
n
s
,
ca
n
b
e
i
d
en
tifie
d
th
r
o
u
g
h
t
h
e
a
n
aly
s
is
o
f
s
p
ee
c
h
p
atter
n
s
.
On
e
p
o
s
s
ib
le
m
eth
o
d
f
o
r
e
n
h
an
ci
n
g
t
h
e
u
n
d
er
s
tan
d
i
n
g
o
f
s
tu
tter
in
g
in
v
o
lv
es
in
teg
r
atin
g
tech
n
o
lo
g
i
es
th
at
d
etec
t
s
p
ee
ch
an
d
g
ait.
B
y
an
aly
zin
g
t
h
e
u
n
d
er
ly
i
n
g
f
ac
to
r
s
t
h
at
co
n
tr
ib
u
te
to
v
o
ca
l
ab
n
o
r
m
alities
an
d
th
e
b
r
o
ad
er
s
p
ec
tr
u
m
o
f
m
o
to
r
s
y
s
tem
b
eh
a
v
io
r
s
,
it
b
ec
o
m
es
f
ea
s
ib
le
to
ac
q
u
i
r
e
a
m
o
r
e
co
m
p
r
eh
en
s
iv
e
co
m
p
r
eh
en
s
io
n
o
f
th
is
p
a
r
ticu
lar
co
n
d
itio
n
.
T
h
e
u
n
i
o
n
p
lace
s
s
ig
n
if
ican
t
im
p
o
r
ta
n
ce
o
n
th
e
f
u
n
d
am
en
tal
in
ter
c
o
n
n
e
ctio
n
b
etwe
en
th
e
m
in
d
an
d
b
o
d
y
.
I
t
p
r
o
v
id
es
n
o
t
o
n
ly
tech
n
o
lo
g
ical
a
d
v
an
ce
m
e
n
ts
b
u
t a
ls
o
ad
o
p
ts
a
c
o
m
p
r
e
h
en
s
iv
e
ap
p
r
o
ac
h
to
e
n
h
an
ci
n
g
h
u
m
an
well
-
b
ei
n
g
.
‒
C
o
m
p
r
eh
en
s
iv
e
s
p
ee
ch
im
p
ai
r
m
en
t
d
etec
tio
n
:
T
h
e
s
y
s
tem
is
d
esig
n
ed
to
d
etec
t
v
ar
io
u
s
ty
p
es
o
f
s
p
ee
ch
im
p
ed
im
en
ts
,
in
clu
d
i
n
g
s
tam
m
er
s
,
ef
f
ec
tiv
ely
ad
d
r
ess
in
g
t
h
e
co
m
p
lex
it
y
o
f
m
u
ltip
le
s
tam
m
er
ty
p
es
an
d
p
r
o
v
id
i
n
g
co
m
p
r
e
h
en
s
iv
e
s
p
ee
ch
im
p
air
m
en
t
d
etec
tio
n
.
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
I
n
teg
r
a
tin
g
g
a
it a
n
d
s
p
ee
ch
d
y
n
a
mics me
th
o
d
o
lo
g
ies fo
r
en
h
a
n
ce
d
… (
R
a
vikira
n
R
ed
d
a
p
p
a
R
ed
d
y
)
4871
‒
Sp
ec
tr
al
lear
n
in
g
f
o
r
ac
cu
r
ate
d
etec
tio
n
:
T
h
e
m
eth
o
d
o
lo
g
y
e
m
p
lo
y
s
StEn
s
em
b
leNe
t,
a
n
eu
r
al
n
etwo
r
k
th
at
lear
n
s
s
p
ec
tr
al
f
ea
tu
r
es
at
t
h
e
f
r
am
e
lev
el,
en
h
a
n
cin
g
th
e
s
y
s
tem
's
ab
ilit
y
to
ac
c
u
r
ately
i
d
en
tify
s
p
ee
ch
im
p
ed
im
en
ts
th
r
o
u
g
h
s
p
ec
tr
al
r
ep
r
esen
tatio
n
s
.
‒
Gait
an
aly
s
is
ad
ap
ted
to
v
is
u
al
en
v
ir
o
n
m
e
n
ts
: T
h
e
m
eth
o
d
o
lo
g
y
in
tr
o
d
u
ce
s
en
h
a
n
ce
d
to
p
o
l
o
g
ical
lear
n
in
g
,
wh
ich
en
ab
les
th
e
m
o
d
el
to
a
d
ap
t
to
d
if
f
e
r
en
t
v
is
u
al
en
v
i
r
o
n
m
en
ts
an
d
en
h
an
ce
s
its
ca
p
ac
ity
to
r
ec
o
g
n
ize
an
d
a
n
aly
ze
g
ait
p
atter
n
s
ac
r
o
s
s
v
ar
y
in
g
c
o
n
d
itio
n
s
,
im
p
r
o
v
in
g
g
ait
an
aly
s
is
b
ased
o
n
s
k
eleta
l m
o
tio
n
d
ata.
2.
RE
L
AT
E
D
WO
RK
T
h
e
in
itial
in
v
esti
g
atio
n
s
in
t
h
is
f
ield
f
o
cu
s
ed
p
r
im
ar
ily
o
n
v
ar
io
u
s
s
p
ec
tr
o
g
r
am
-
b
ased
f
ea
t
u
r
es,
s
u
ch
as
lin
ea
r
p
r
e
d
ictiv
e
ce
p
s
tr
al
c
o
ef
f
icien
ts
(
L
PC
C
s
)
,
Me
l
-
f
r
eq
u
en
cy
ce
p
s
tr
al
co
ef
f
icien
ts
(
M
FC
C
s
)
,
an
d
s
im
ilar
ch
ar
ac
ter
is
tics
[
7
]
.
Fu
r
th
er
m
o
r
e,
th
er
e
h
as
b
ee
n
a
n
o
tab
le
r
is
e
in
in
ter
est
r
eg
ar
d
i
n
g
th
e
ap
p
licatio
n
o
f
W
T
(
wav
elet
t
r
an
s
f
o
r
m
)
im
a
g
es
f
o
r
p
h
y
s
io
lo
g
ical
s
ig
n
al
p
r
o
ce
s
s
in
g
a
p
p
licatio
n
s
,
as
e
v
id
en
ce
d
b
y
r
ec
e
n
t
r
esear
c
h
.
T
h
e
im
ag
es
p
r
o
v
id
e
s
ig
n
if
ica
n
t
in
s
ig
h
ts
in
to
th
e
d
y
n
a
m
ics
o
f
tim
e
an
d
f
r
eq
u
e
n
cy
.
T
h
e
u
tili
za
tio
n
o
f
th
ese
m
eth
o
d
s
allo
ws
f
o
r
th
e
ac
q
u
is
itio
n
o
f
in
f
o
r
m
atio
n
r
e
g
ar
d
in
g
th
e
f
u
n
d
a
m
en
tal
f
ac
to
r
s
co
n
tr
i
b
u
tin
g
to
s
tu
tter
in
g
,
wh
ile
also
o
f
f
er
in
g
s
ig
n
if
ican
t
o
b
s
er
v
atio
n
s
r
eg
ar
d
in
g
th
e
p
r
o
g
r
ess
iv
e
ch
ar
ac
ter
is
tics
o
f
th
e
c
o
n
d
itio
n
as
it
d
ev
elo
p
s
o
v
er
tim
e.
I
n
2
0
1
9
,
Ku
p
p
ev
elt
et
a
l.
[
8
]
co
n
d
u
c
ted
a
s
tu
d
y
in
wh
ich
th
e
y
tr
ain
ed
a
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
u
s
in
g
1
0
-
d
im
en
s
io
n
al
r
esp
ir
ato
r
y
c
h
ar
ac
ter
is
tics
,
f
o
c
u
s
in
g
s
p
ec
if
ically
o
n
th
e
b
lo
ck
SD
[
9
]
.
T
h
e
r
esear
ch
er
s
u
tili
ze
d
a
d
ataset
co
n
s
is
tin
g
o
f
6
8
in
d
i
v
id
u
als
wh
o
wer
e
Sp
an
is
h
s
p
ea
k
e
r
s
an
d
o
r
i
g
in
ated
f
r
o
m
v
ar
io
u
s
co
u
n
tr
ies in
latin
Am
e
r
ica
to
ca
r
r
y
o
u
t th
ei
r
ca
s
e
s
tu
d
y
.
T
h
e
b
in
ar
y
d
ee
p
lear
n
in
g
class
i
f
ier
in
tr
o
d
u
ce
d
b
y
[
1
0
]
in
th
eir
s
tu
d
y
em
p
lo
y
s
a
co
m
b
i
n
atio
n
o
f
r
esid
u
al
n
etwo
r
k
an
d
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
R
esNet+BiLSTM
)
to
ac
cu
r
ately
d
etec
t
an
d
class
if
y
s
ix
d
is
tin
ct
ty
p
es
o
f
d
is
f
lu
en
cies.
Dis
f
lu
en
cies
ca
n
m
an
if
est
i
n
d
if
f
er
en
t
f
o
r
m
s
,
in
clu
d
in
g
p
r
o
lo
n
g
atio
n
,
w
o
r
d
r
ep
etitio
n
,
s
o
u
n
d
r
ep
etitio
n
,
p
h
r
ase
r
ep
etitio
n
,
an
d
f
alse
s
tar
ts
.
Pro
m
is
in
g
r
esu
lts
wer
e
o
b
s
er
v
ed
i
n
a
r
estricte
d
s
u
b
s
et
o
f
th
e
Un
i
v
er
s
ity
C
o
lleg
e
L
o
n
d
o
n
’
s
Ar
ch
iv
e
o
f
Stu
tte
r
ed
Sp
ee
ch
(
UC
L
ASS
)
d
ataset
,
wh
ich
co
n
s
is
ted
o
f
2
5
s
p
ea
k
e
r
s
.
T
h
ese
r
esu
lts
wer
e
ac
h
iev
ed
b
y
u
tili
zin
g
s
p
ec
tr
o
g
r
am
s
as
in
p
u
t
f
ea
tu
r
es
[
1
1
]
.
I
n
a
s
p
ec
if
ic
s
tu
d
y
,
W
an
g
et
a
l.
[
1
2
]
em
p
lo
y
ed
th
e
co
n
v
o
l
u
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
C
o
n
v
L
STM
)
m
o
d
el
to
id
en
tify
a
n
d
d
etec
t
s
ix
s
ep
ar
ate
ca
teg
o
r
ies
o
f
s
tu
tter
in
g
.
T
h
e
ca
teg
o
r
ies
m
en
tio
n
ed
co
n
s
is
t
o
f
b
lo
ck
s
,
p
r
o
lo
n
g
atio
n
s
,
s
o
u
n
d
r
ep
etitio
n
s
,
wo
r
d
r
ep
etitio
n
s
,
an
d
in
ter
jectio
n
s
.
T
h
e
in
p
u
t
o
f
th
e
m
o
d
el
co
n
s
is
ts
o
f
4
1
-
d
im
en
s
io
n
al
p
h
o
n
e
m
e
f
ea
t
u
r
e
v
ec
t
o
r
s
,
th
r
ee
-
d
im
en
s
io
n
al
p
itch
ch
ar
ac
ter
is
tics
,
eig
h
t
-
d
im
en
s
io
n
al
ar
ticu
lato
r
y
f
ea
tu
r
es,
a
n
d
4
0
-
d
im
en
s
io
n
al
MFC
C
s
.
T
h
e
Stu
t
ter
Net
class
if
ier
was in
tr
o
d
u
ce
d
[
1
3
]
.
T
h
e
cl
ass
i
f
ie
r
u
til
izes
a
s
i
n
g
u
l
ar
m
u
lti
-
cl
ass
ti
m
e
d
e
la
y
n
eu
r
al
n
etw
o
r
k
(
T
DNN
)
.
T
h
e
cl
as
s
if
i
er
ca
n
ac
c
u
r
at
el
y
i
d
en
tif
y
a
n
d
d
if
f
e
r
e
n
ti
ate
b
etw
ee
n
co
r
e
b
e
h
a
v
i
o
r
s
an
d
f
l
u
e
n
t
s
p
ee
ch
s
e
g
m
e
n
ts
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
cl
ass
if
ie
r
d
e
m
o
n
s
t
r
a
tes
p
r
o
m
is
i
n
g
d
ete
cti
o
n
c
ap
a
b
ili
ties
o
n
a
s
i
g
n
i
f
ic
a
n
t
s
u
b
s
et
o
f
UC
L
ASS
s
p
e
a
k
e
r
s
w
h
e
n
co
m
p
a
r
e
d
t
o
s
t
ate
-
of
-
t
h
e
-
ar
t
cl
ass
if
ie
r
s
.
T
h
e
i
n
p
u
t
f
o
r
m
at
ac
ce
p
te
d
b
y
t
h
e
m
o
d
el
is
li
m
it
e
d
t
o
2
0
-
d
i
m
e
n
s
i
o
n
al
MFC
C
.
Sm
i
th
an
d
W
e
b
e
r
[
1
4
]
i
n
tr
o
d
u
ce
"
T
r
a
n
St
u
tt
e
r
,
"
a
n
i
n
n
o
v
a
ti
v
e
d
ee
p
-
le
a
r
n
in
g
m
o
d
el
t
h
a
t
le
v
e
r
a
g
es
tr
a
n
s
f
o
r
m
e
r
s
.
T
h
e
m
o
d
el
b
ei
n
g
p
r
ese
n
t
ed
eli
m
i
n
a
tes
t
h
e
n
ee
d
f
o
r
co
n
v
o
l
u
ti
o
n
a
n
d
h
as
b
ee
n
s
p
e
ci
f
ic
all
y
d
esi
g
n
e
d
to
s
h
o
wc
ase
o
u
ts
t
an
d
i
n
g
p
e
r
f
o
r
m
a
n
c
e
i
n
v
o
ic
e
d
is
f
lu
en
c
y
c
lass
i
f
i
ca
t
io
n
.
T
h
e
T
r
a
n
St
u
tte
r
s
y
s
te
m
u
til
izes
m
u
lti
-
h
ea
d
s
el
f
-
att
e
n
ti
o
n
a
n
d
p
o
s
iti
o
n
al
e
n
c
o
d
i
n
g
te
c
h
n
iq
u
e
s
to
e
f
f
ec
t
iv
el
y
c
a
p
t
u
r
e
i
n
t
r
i
ca
t
e
te
m
p
o
r
al
p
at
te
r
n
s
,
d
is
t
in
g
u
is
h
i
n
g
i
t
f
r
o
m
c
o
n
v
en
ti
o
n
al
m
et
h
o
d
s
.
C
o
n
s
i
d
e
r
i
n
g
th
e
ad
o
p
ti
o
n
o
f
g
a
it
,
t
h
er
e
h
a
v
e
b
e
en
n
u
m
e
r
o
u
s
w
o
r
k
s
s
ep
ar
ate
ly
o
n
g
ai
t
r
ec
o
g
n
iti
o
n
,
we
ar
a
b
le
s
en
s
o
r
s
a
r
e
em
p
l
o
y
e
d
t
o
m
o
n
i
to
r
d
ai
ly
ac
ti
v
iti
e
s
a
n
d
li
f
est
y
le
[
1
5
]
.
T
as
k
-
s
p
e
ci
f
i
c
m
ac
h
i
n
e
le
a
r
n
in
g
t
ec
h
n
i
q
u
es
ar
e
e
m
p
l
o
y
e
d
t
o
a
ch
i
e
v
e
ac
cu
r
a
te
s
e
g
m
en
tat
io
n
o
f
r
ec
o
r
d
e
d
ac
ti
v
i
ties
.
T
h
e
p
r
i
o
r
iti
za
ti
o
n
o
f
id
e
n
ti
f
y
i
n
g
a
n
d
a
n
al
y
z
in
g
ess
e
n
ti
al
a
cti
o
n
s
is
a
c
r
i
tic
al
p
r
o
c
ess
.
T
h
e
p
u
r
p
o
s
e
f
u
l
d
ev
el
o
p
m
en
t
o
f
t
h
e
s
e
g
m
e
n
t
ati
o
n
s
y
s
t
em
e
n
a
b
l
es
i
t
t
o
a
cc
u
r
at
el
y
i
d
en
ti
f
y
an
d
d
if
f
er
en
tia
te
s
i
x
p
r
im
ar
y
ac
t
io
n
s
:
w
al
k
i
n
g
,
asce
n
d
i
n
g
s
t
ai
r
s
,
d
esc
en
d
i
n
g
s
tai
r
s
,
s
itt
in
g
,
s
t
a
n
d
in
g
,
an
d
l
y
i
n
g
d
o
wn
.
T
h
e
ac
ti
o
n
s
m
e
n
t
io
n
e
d
ar
e
p
r
e
cise
ly
d
et
ec
te
d
a
n
d
s
e
p
ar
ate
d
f
r
o
m
a
co
n
ti
n
u
o
u
s
s
i
g
n
a
l.
T
h
e
i
n
i
tia
l
s
ta
g
e
o
f
t
h
e
p
r
o
c
ess
in
v
o
l
v
es
t
h
e
s
e
g
m
e
n
ta
ti
o
n
o
f
co
n
ti
n
u
o
u
s
s
i
g
n
a
ls
.
T
h
e
n
e
x
t
s
t
ep
i
n
v
o
lv
es
th
e
i
d
e
n
ti
f
i
ca
t
io
n
o
f
tr
an
s
iti
o
n
al
ac
ti
v
i
ties
.
T
h
e
r
es
ea
r
c
h
e
r
s
em
p
l
o
y
e
d
th
e
g
r
ee
d
y
G
au
s
s
ia
n
s
e
g
m
e
n
t
at
io
n
(
GGS
)
tec
h
n
i
q
u
e
to
d
et
ec
t
d
is
c
o
n
ti
n
u
it
ies
th
at
o
cc
u
r
d
u
r
i
n
g
ac
ti
o
n
tr
an
s
iti
o
n
s
.
T
h
e
X
GB
o
o
s
t
m
o
d
e
l w
as
u
t
ili
ze
d
t
o
p
r
ed
ict
h
u
m
a
n
a
cti
v
it
ies
i
n
ea
c
h
f
r
am
e.
T
h
e
au
th
o
r
s
d
id
n
o
t
ad
d
r
ess
th
e
id
en
tific
atio
n
o
f
tr
an
s
itio
n
al
ac
ts
.
T
h
e
au
th
o
r
s
o
f
th
e
s
tu
d
y
h
av
e
in
tr
o
d
u
ce
d
a
n
o
v
el
ap
p
r
o
ac
h
[
1
6
]
to
s
eg
m
e
n
t
an
d
id
e
n
tify
c
o
n
tin
u
o
u
s
ac
tio
n
s
eq
u
en
ce
s
.
T
h
e
s
eq
u
en
ce
ca
n
b
e
class
if
ied
in
to
th
r
ee
p
r
im
ar
y
ca
teg
o
r
ies:
tr
an
s
itio
n
al
ac
tiv
it
ies,
s
tatic
ac
ts
,
an
d
d
y
n
am
ic
ac
tio
n
s
.
T
h
is
s
tu
d
y
in
tr
o
d
u
ce
s
a
n
o
v
el
m
et
h
o
d
o
lo
g
y
f
o
r
th
e
id
e
n
tific
atio
n
o
f
th
e
g
ait
cy
cle
u
tili
zin
g
b
io
m
etr
i
c
d
ata.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
th
is
m
eth
o
d
is
to
ac
cu
r
ately
id
en
tify
an
d
ca
teg
o
r
ize
all
d
y
n
am
ic
ac
tiv
ities
th
at
tak
e
p
lace
in
a
s
p
ec
if
ied
s
eq
u
en
ce
.
T
h
e
ac
h
i
ev
em
en
t
was
ac
co
m
p
lis
h
ed
b
y
an
aly
zin
g
t
h
e
ch
ar
ac
ter
is
tic
s
o
f
th
e
g
ait
s
ig
n
al
d
u
r
in
g
d
y
n
am
ic
ac
tiv
ities
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
t
h
e
an
a
ly
s
is
was
to
ass
ess
th
e
d
is
tr
ib
u
tio
n
o
f
f
lu
ctu
atio
n
p
o
in
ts
in
two
d
is
tin
ct
ty
p
es
o
f
ac
tiv
ities
,
tak
in
g
in
to
ac
co
u
n
t
th
e
d
is
p
ar
ity
in
s
ig
n
al
co
n
v
er
s
io
n
f
r
eq
u
en
cy
b
etwe
en
s
tatic
ac
tio
n
s
an
d
t
r
an
s
itio
n
ac
tio
n
s
.
T
h
e
GaitPar
s
in
g
f
r
am
ewo
r
k
is
a
s
p
ec
i
alize
d
m
eth
o
d
o
lo
g
y
d
ev
elo
p
e
d
to
s
tu
d
y
h
u
m
a
n
s
em
an
tic
p
ar
s
in
g
an
d
g
ait
d
etec
ti
o
n
[
1
7
]
.
T
h
e
m
ain
g
o
al
is
to
m
eth
o
d
ically
ex
am
i
n
e
th
e
h
u
m
an
b
o
d
y
an
d
ca
teg
o
r
iz
e
it
in
to
s
ep
ar
ate
an
ato
m
ical
el
em
en
ts
th
at
h
av
e
well
-
d
ef
in
e
d
an
d
co
m
p
r
eh
e
n
s
iv
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
6
9
-
488
2
4872
s
tr
u
ctu
r
es.
T
o
f
u
lf
ill
th
is
r
eq
u
i
r
em
en
t,
a
d
u
al
-
b
r
a
n
ch
f
ea
tu
r
e
ex
tr
ac
tio
n
n
etwo
r
k
is
em
p
l
o
y
e
d
.
T
h
e
n
etwo
r
k
h
as
b
ee
n
in
ten
tio
n
ally
d
esig
n
e
d
to
ef
f
icien
tly
h
a
n
d
le
b
o
th
wh
o
le
-
b
o
d
y
g
ait
an
d
in
d
i
v
id
u
al
b
o
d
y
p
ar
ts
.
On
e
p
o
ten
tial
m
eth
o
d
f
o
r
ass
ess
in
g
s
elf
-
o
cc
lu
s
i
o
n
in
a
g
ait
s
eq
u
en
ce
i
n
v
o
lv
es
th
e
u
s
e
o
f
a
tech
n
i
q
u
e
ca
lled
s
elf
-
o
cc
lu
s
io
n
f
r
am
e
ev
alu
atio
n
.
T
h
e
m
ain
o
b
jectiv
e
o
f
th
is
ev
alu
atio
n
is
to
en
h
an
ce
th
e
ef
f
icien
c
y
o
f
g
a
it f
r
am
es th
at
ex
h
ib
it
s
ig
n
if
ican
t v
ar
iatio
n
s
.
T
h
is
p
ap
er
p
r
esen
ts
a
n
o
v
el
m
eth
o
d
o
l
o
g
y
f
o
r
im
p
r
o
v
in
g
th
e
f
u
n
ctio
n
ality
o
f
h
u
m
an
g
e
s
tu
r
e
an
d
ac
tiv
ity
r
ec
o
g
n
itio
n
(
HGAR)
[
1
8
]
,
[
1
9
]
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
in
v
o
l
v
es
u
s
in
g
a
n
ewly
d
ev
elo
p
ed
s
m
ar
t
b
elt
to
co
llect
p
o
s
tu
r
e
d
ata.
I
n
ad
d
itio
n
,
th
e
p
r
o
ce
s
s
em
p
lo
y
s
an
ad
v
an
ce
d
s
tack
in
g
d
esig
n
.
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
in
tellig
en
t
f
r
am
ewo
r
k
o
u
tlin
ed
in
th
is
p
ap
er
was
ev
alu
ated
th
r
o
u
g
h
a
s
er
i
es
o
f
ex
p
er
im
e
n
ts
.
T
h
e
ex
p
er
im
en
ts
in
v
o
lv
ed
th
e
in
teg
r
ati
o
n
o
f
m
u
ltip
le
s
en
s
o
r
s
to
d
ev
elo
p
a
h
o
li
s
tic
s
o
lu
tio
n
f
o
r
th
e
HGAR p
r
o
b
lem
.
3.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
W
e
d
ev
elo
p
an
d
d
esig
n
a
s
y
s
tem
th
at
is
u
tili
ze
d
t
o
d
etec
t
a
n
d
u
n
d
e
r
s
tan
d
p
eo
p
le
h
av
i
n
g
ch
allen
g
es
wh
ile
s
p
ea
k
in
g
,
s
p
ec
if
ically
v
ar
io
u
s
k
in
d
s
o
f
s
tam
m
er
s
.
T
h
e
tr
ad
itio
n
al
m
eth
o
d
to
r
eso
lv
e
th
is
is
s
u
e
is
th
e
d
ev
elo
p
m
e
n
t
o
f
v
ar
io
u
s
task
class
if
ier
s
,
th
i
s
f
u
r
t
h
er
b
ec
o
m
es
co
m
p
lex
as
f
o
r
ev
er
y
in
p
u
t
th
e
s
o
u
n
d
s
n
ip
p
et
co
u
ld
h
av
e
v
ar
io
u
s
k
in
d
s
o
f
s
t
am
m
er
s
.
C
o
n
s
id
er
in
g
th
e
s
ce
n
ar
io
an
d
th
e
s
ca
r
city
o
f
s
tam
m
er
d
ata
in
f
o
r
m
atio
n
,
s
p
ee
ch
d
is
r
u
p
tio
n
a
n
aly
ze
r
s
ar
e
u
s
ed
to
aid
b
y
d
ec
r
ea
s
in
g
th
e
d
if
f
icu
lty
o
f
ev
er
y
b
in
ar
y
w
o
r
k
.
T
h
e
m
ain
f
o
cu
s
is
to
d
ev
el
o
p
a
s
p
ec
if
ic
s
y
s
te
m
ar
ch
itectu
r
e
th
at
is
tr
ain
ed
s
p
ec
if
ically
f
o
r
th
e
r
ec
o
g
n
itio
n
an
d
d
etec
tio
n
o
f
v
ar
io
u
s
k
in
d
s
o
f
s
tam
m
er
s
f
o
r
ev
er
y
tr
ain
ed
s
itu
atio
n
,
wh
er
e
in
co
llectiv
ely
it
ca
n
r
ec
o
g
n
iz
e
v
ar
io
u
s
s
tam
m
er
s
an
d
s
tu
tter
s
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
t
em
(
PS
)
aim
s
to
d
e
v
elo
p
a
n
et
wo
r
k
th
at
h
as
th
e
a
b
ilit
y
o
f
s
p
e
ctr
al
lear
n
in
g
at
th
e
f
r
am
e
s
tag
e
e
x
p
r
ess
io
n
s
an
d
te
m
p
o
r
al
lin
k
s
.
Ho
wev
e
r
,
th
e
s
y
s
tem
h
as
to
em
p
h
asize
p
r
o
m
in
en
t
r
eg
io
n
s
o
f
in
p
u
t
f
o
r
ef
f
icien
t
lear
n
in
g
o
f
s
p
ee
ch
im
p
ed
im
en
ts
as
w
ell
as
ac
cu
r
ate
p
er
f
o
r
m
an
ce
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
also
f
o
cu
s
es
o
n
g
ait
d
etec
tio
n
b
ased
o
n
s
k
eleta
l
m
o
v
em
e
n
ts
.
I
t
r
ef
er
s
t
o
a
tech
n
iq
u
e
t
h
at
in
v
o
lv
es
s
tu
d
y
i
n
g
t
h
e
m
an
n
er
an
d
p
atter
n
s
o
f
an
in
d
iv
id
u
al’
s
walk
in
g
.
T
h
e
p
r
o
p
o
s
ed
ad
a
p
tiv
e
g
r
ap
h
to
p
o
lo
g
y
c
o
n
v
o
lu
tio
n
n
etwo
r
k
(
AGT
-
C
o
n
v
Net)
f
o
r
s
k
eleta
l
m
o
tio
n
with
t
h
e
jo
in
t
-
f
o
cu
s
ed
lear
n
in
g
f
ilter
en
h
an
ce
s
t
h
e
g
ait
p
r
o
f
ile
b
y
co
n
s
id
er
in
g
v
ar
i
o
u
s
v
is
u
al
en
v
ir
o
n
m
en
ts
.
T
h
e
p
r
o
p
o
s
ed
en
h
a
n
ce
d
to
p
o
lo
g
ical
lear
n
in
g
im
p
r
o
v
es th
e
g
r
ap
h
ica
l
s
tr
u
ctu
r
es
th
at
in
clu
d
e
lin
k
s
b
e
twee
n
th
e
jo
in
ts
,
it
is
d
ev
elo
p
e
d
to
aid
a
c
o
m
b
in
e
d
ag
g
r
eg
ati
o
n
f
o
r
g
r
ap
h
-
b
ased
co
n
v
o
l
u
tio
n
s
in
th
e
n
ex
t p
h
ase
o
f
th
e
a
r
ch
itectu
r
e.
3
.
1
.
StEn
s
em
bleNet
:
Stut
t
er
ing
det
ec
t
io
n t
hro
ug
h e
ns
em
ble dee
p lea
rni
ng
a
rc
hite
ct
ure
T
h
e
s
tu
d
y
p
r
o
p
o
s
es
a
n
etwo
r
k
th
at
u
tili
ze
s
s
p
ec
tr
o
g
r
am
s
(
MF
C
C
)
o
f
s
o
u
n
d
s
n
ip
p
et
in
p
u
ts
,
th
is
n
etwo
r
k
is
ter
m
ed
as
StEn
s
em
b
leNe
t.
Sp
ec
tr
o
g
r
am
s
ar
e
u
s
ed
in
p
r
ev
io
u
s
m
eth
o
d
s
f
o
r
th
e
s
p
ee
c
h
p
r
o
ce
s
s
,
an
d
R
esNet
is
u
s
ed
f
o
r
lear
n
in
g
s
p
ec
tr
al
f
r
am
e
s
tag
e
ex
p
r
ess
i
o
n
s
.
A
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
te
r
m
m
e
m
o
r
y
is
u
tili
ze
d
wh
en
s
tam
m
er
v
ar
iatio
n
s
h
av
e
s
p
ec
i
f
ic
tem
p
o
r
al
as
well
as
s
p
ec
tr
a
l
f
ea
tu
r
es.
Usi
n
g
class
if
icatio
n
o
f
s
tam
m
er
in
g
as
well
as a
u
to
m
atic
s
p
ee
ch
d
ete
ctio
n
is
u
s
ed
to
b
o
o
s
t th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
n
etwo
r
k
.
I
n
th
e
s
p
ec
tr
o
g
r
am
,
co
l
o
r
s
d
ep
ict
am
p
litu
d
e
f
o
r
e
v
er
y
f
ilter
at
a
p
ar
ticu
lar
f
r
a
m
e,
f
ewe
r
am
p
litu
d
es
ar
e
d
ep
icted
with
b
l
u
e
an
d
th
e
co
lo
r
s
y
ello
w
as
well
as
g
r
ee
n
s
h
o
w
in
cr
ea
s
ed
am
p
litu
d
es.
T
h
e
p
r
o
p
o
s
ed
StEn
s
em
b
leNe
t
aim
s
m
o
d
el
to
lear
n
ef
f
icien
t
d
e
p
ictio
n
s
f
o
r
ev
er
y
s
p
ec
tr
o
g
r
am
as
in
p
u
t.
I
n
th
is
ca
s
e,
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
ar
e
u
s
ed
alo
n
g
with
ex
citatio
n
an
d
s
q
u
ee
z
e
m
o
d
els
th
at
en
h
an
ce
p
r
io
r
m
eth
o
d
s
in
v
ar
io
u
s
d
o
m
ain
s
.
T
h
e
ar
ch
it
ec
tu
r
e
o
f
th
e
R
esNet
is
s
h
o
wn
to
h
av
e
a
r
eso
lu
tio
n
t
o
r
is
in
g
g
r
ad
ien
t
i
s
s
u
es,
th
ese
ar
e
ch
allen
g
es
th
at
ar
is
e
d
u
r
in
g
b
ac
k
p
r
o
p
ag
atio
n
.
No
r
m
ally
,
t
h
e
r
is
e
in
th
e
d
ep
th
o
f
th
e
m
o
d
el
ca
u
s
es
th
e
weig
h
t
g
r
ad
ien
t
to
r
ed
u
ce
f
o
r
e
v
er
y
la
y
er
.
E
v
e
n
tu
ally
,
th
is
co
u
ld
s
to
p
th
e
g
r
a
d
ien
ts
f
r
o
m
alter
in
g
o
r
en
lar
g
in
g
weig
h
ts
.
T
h
er
ef
o
r
e,
th
is
p
r
ev
en
ts
m
o
d
if
icatio
n
s
in
th
e
m
o
d
el.
T
h
e
R
esNet
m
o
d
el
r
eso
lv
es
th
is
i
s
s
u
e
b
y
u
s
in
g
s
h
o
r
tcu
ts
in
th
e
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
eg
m
en
ts
th
at
lead
to
th
e
p
r
eser
v
atio
n
o
f
s
eg
m
en
ts
th
at
ca
n
m
o
v
e
th
e
g
r
ad
ien
ts
f
o
r
d
ee
p
lear
n
in
g
m
o
d
els.
Pre
v
io
u
s
tech
n
o
lo
g
ies
u
s
ed
in
th
is
f
ield
s
u
ch
as
ex
citatio
n
an
d
s
q
u
ee
ze
ar
e
u
s
ed
f
o
r
th
e
class
if
icatio
n
o
f
im
a
g
es
th
at
d
ec
r
ea
s
e
th
e
er
r
o
r
.
E
ac
h
k
er
n
el
in
th
e
lay
er
o
f
a
c
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
lead
s
to
a
n
e
x
tr
a
ch
an
n
el
in
th
e
attr
i
b
u
te
m
ap
o
u
tp
u
t.
T
h
e
ex
citatio
n
a
n
d
s
q
u
ee
ze
s
eg
m
en
ts
in
cr
ea
s
e
th
e
co
n
ce
n
t
r
atio
n
o
f
ch
an
n
el
-
b
ased
lin
k
s
in
th
e
c
o
n
v
o
lu
ti
o
n
al
n
eu
r
al
n
etwo
r
k
.
T
h
ese
s
eg
m
en
ts
h
av
e
two
m
ain
p
r
o
ce
s
s
es.
T
h
e
co
llectio
n
o
f
attr
ib
u
te
m
ap
s
f
o
r
th
e
len
g
th
an
d
wid
th
is
p
er
f
o
r
m
e
d
b
y
a
s
q
u
ee
zin
g
p
r
o
c
ess
t
h
at
r
esu
lt
s
in
a
ch
an
n
el
d
escr
ip
to
r
th
at
h
as
o
n
e
d
im
en
s
io
n
.
T
h
e
co
m
p
letely
lin
k
ed
lay
er
s
g
iv
e
ch
an
n
el
-
b
ased
len
g
th
s
in
th
e
ex
citatio
n
p
r
o
ce
s
s
th
at
is
im
p
lem
en
ted
in
th
e
in
itial
attr
ib
u
te
m
ap
.
T
h
e
ar
c
h
itectu
r
es
o
f
R
esNet
a
s
well
a
s
ex
citatio
n
an
d
s
q
u
ee
ze
ar
e
u
tili
ze
d
f
o
r
ef
f
ec
tiv
e
lear
n
in
g
o
f
th
e
s
p
ec
tr
al
r
ep
r
esen
tatio
n
s
at
th
e
f
r
am
e
s
tag
e
at
th
e
in
p
u
t,
th
e
p
r
o
p
o
s
ed
m
o
d
el
co
n
s
is
t
s
o
f
a
R
esNet
-
ex
citatio
n
an
d
s
q
u
ee
ze
m
o
d
el.
T
h
e
m
o
d
el
h
as
eig
h
t
s
eg
m
en
ts
o
f
R
esNet
-
ex
citatio
n
an
d
s
q
u
ee
ze
.
Fo
r
ev
er
y
s
eg
m
en
t
in
th
e
StEn
s
em
b
leNe
t,
th
er
e
ex
is
ts
a
co
n
v
o
l
u
tio
n
s
lay
er
o
f
two
-
d
im
e
n
s
io
n
s
ize
in
b
atch
es
o
f
th
r
ee
,
s
u
ch
th
at
ev
er
y
l
ay
er
is
a
r
esu
lt
o
f
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
as
well
as
b
atch
n
o
r
m
aliza
tio
n
.
A
n
in
d
iv
id
u
al
r
esid
u
al
lin
k
h
as
s
im
ilar
in
p
u
t
as
th
e
s
eg
m
en
t’
s
b
r
an
ch
th
at
is
ad
d
e
d
b
ef
o
r
e
th
e
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
,
h
o
wev
e
r
,
it
is
af
ter
th
e
ex
citatio
n
an
d
s
q
u
ee
ze
.
Fo
r
ev
er
y
lin
k
,
th
er
e
ex
is
ts
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
I
n
teg
r
a
tin
g
g
a
it a
n
d
s
p
ee
ch
d
y
n
a
mics me
th
o
d
o
lo
g
ies fo
r
en
h
a
n
ce
d
… (
R
a
vikira
n
R
ed
d
a
p
p
a
R
ed
d
y
)
4873
a
lay
er
o
f
co
n
v
o
lu
tio
n
af
ter
wh
ich
b
atch
n
o
r
m
aliza
tio
n
o
cc
u
r
s
.
Glo
b
al
p
o
o
lin
g
is
u
s
ed
f
o
r
e
x
c
itatio
n
an
d
s
q
u
ee
ze
in
th
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
r
esu
lt
is
later
m
ad
e
to
en
ter
tw
o
co
m
p
letely
lin
k
ed
lay
e
r
s
,
th
e
f
ir
s
t
lay
er
h
as
th
e
ac
tiv
atio
n
R
eL
U
f
u
n
ctio
n
,
f
o
ll
o
wed
b
y
th
e
s
ig
m
o
id
g
ait
f
u
n
ctio
n
in
th
e
n
e
x
t
lay
er
.
T
h
e
o
u
tp
u
t
is
u
s
ed
f
o
r
th
e
ev
alu
atio
n
o
f
th
e
m
aj
o
r
co
n
v
o
l
u
tio
n
lay
er
u
tili
zin
g
ch
a
n
n
el
-
b
ased
m
u
ltip
licatio
n
.
T
h
e
tem
p
o
r
al
co
n
n
ec
tio
n
m
u
s
t
b
e
lear
n
ed
b
y
th
e
m
o
d
el
b
etwe
en
d
ep
ictio
n
s
lear
n
ed
f
r
o
m
s
p
ec
tr
o
g
r
am
s
,
h
en
ce
in
th
is
ca
s
e
we
u
s
e
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
s
.
E
v
er
y
lo
n
g
s
h
o
r
t
-
t
er
m
m
em
o
r
y
h
as
a
m
em
o
r
y
ce
ll
th
at
co
n
tai
n
s
d
at
a
th
at
was
p
r
esen
t
in
p
r
io
r
ce
lls
th
at
p
er
m
its
th
e
m
o
d
el
to
le
ar
n
an
d
u
n
d
er
s
tan
d
tem
p
o
r
al
lin
k
s
.
T
h
e
l
o
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
h
as
t
h
e
m
e
m
o
r
y
ce
ll
as
a
co
m
p
o
n
en
t
th
at
h
as
v
ar
io
u
s
g
ates
th
at
co
llectiv
ely
m
an
a
g
e
in
p
u
t
d
at
a
an
d
th
e
p
r
io
r
ce
ll
in
f
o
r
m
ati
o
n
th
at
is
u
s
ed
in
th
e
g
en
e
r
atio
n
o
f
th
e
p
r
esen
t
c
ell
as we
ll a
s
co
n
ce
aled
s
tates.
Sp
ec
if
ically
,
th
e
in
p
u
t g
ate
is
d
e
n
o
ted
b
y
j
u
,
th
e
f
o
r
g
et
g
ate
is
ex
p
r
ess
ed
as
g
u
th
at
is
u
s
ed
in
lear
n
in
g
th
e
d
ata
f
r
o
m
ev
er
y
ce
ll
th
at
will
b
e
s
to
r
e
d
in
t
h
e
p
r
esen
t
n
ew
ce
ll
is
d
e
n
o
ted
as
D
u
.
T
h
is
is
ex
p
r
ess
ed
in
th
e
(
1
)
to
(
3
)
.
g
u
=
ϑ
(
X
g
.
[
i
u
−
1
,
y
u
]
+
c
g
)
(
1
)
j
u
=
ϑ
(
X
j
.
[
i
u
−
1
,
y
u
]
+
c
j
)
(
2
)
D
u
=
j
u
∗
ta
n
h
ta
n
h
(
X
D
.
[
i
u
−
1
,
y
u
]
+
c
D
)
+
g
u
∗
D
u
−
1
(
3
)
p
u
=
ϑ
(
X
p
.
[
i
u
−
1
,
y
u
]
+
c
p
)
(
4
)
(
D
u
)
*
p
u
(
5
)
C
o
n
s
id
er
in
g
th
e
(
1
)
to
(
3
)
,
th
e
s
ig
m
o
id
f
u
n
ctio
n
is
r
ep
r
esen
t
ed
as
ϑ
,
an
d
a
p
o
in
t
-
b
ased
m
u
ltip
licatio
n
is
r
ep
r
esen
ted
u
s
in
g
th
e
o
p
er
at
o
r
∗
.
T
h
e
p
r
esen
t
ce
ll
with
p
u
wh
ich
is
th
e
o
u
tp
u
t
g
ate
u
tili
ze
d
f
o
r
th
e
g
en
er
atio
n
o
f
th
e
co
n
ce
aled
ce
ll
in
th
e
u
n
it
i
u
th
at
is
f
o
r
m
u
lated
in
th
e
e
q
u
atio
n
g
iv
en
(
4
)
an
d
(
5
)
.
T
h
e
m
em
o
r
y
ce
ll
as
well
as th
e
co
n
ce
aled
ce
ll is
m
o
v
ed
to
th
e
n
ex
t lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
u
n
it,
th
is
en
co
u
r
a
g
es th
e
m
o
d
el
in
lear
n
in
g
d
ep
en
d
e
n
cies
f
o
r
th
e
lo
n
g
ter
m
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
StEn
s
em
b
leNe
t
u
s
es
a
b
id
ir
ec
tio
n
al
l
o
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
th
at
h
as
two
ty
p
es
o
f
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
th
at
ar
e
co
n
tr
ar
y
in
d
ir
ec
tio
n
,
ex
p
an
d
in
g
th
e
in
f
o
r
m
atio
n
f
r
o
m
lin
k
s
o
f
th
e
f
u
tu
r
e
as
well
as
th
e
p
ast.
T
h
e
m
u
ltip
lied
o
u
tp
u
ts
o
f
th
e
n
etwo
r
k
s
r
esu
lt
i
n
a
s
in
g
le
p
a
r
ticu
lar
o
u
tp
u
t.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
two
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
th
at
ar
e
co
n
s
ec
u
tiv
e,
wh
er
e
ev
er
y
ce
ll
h
as
5
0
0
co
n
ce
aled
ce
lls
.
C
o
n
s
id
er
in
g
th
e
d
im
en
s
io
n
o
f
th
e
d
ataset,
we
av
o
id
an
y
is
s
u
es
o
f
o
v
er
f
itti
n
g
b
y
m
ask
in
g
th
e
n
eu
r
o
n
s
p
r
o
d
u
ce
d
f
o
r
tr
ain
in
g
th
e
m
o
d
el
wh
er
e
a
s
p
ar
s
e
ex
p
r
ess
io
n
o
f
th
e
in
f
o
r
m
ati
o
n
is
p
r
o
d
u
ce
d
.
T
h
e
u
s
e
o
f
atten
tio
n
tech
n
iq
u
e
s
as
well
a
s
it
s
d
iv
er
s
ity
p
er
m
its
m
o
r
e
em
p
h
asis
o
n
im
p
o
r
tan
t
p
ar
ts
o
f
em
b
ed
d
in
g
.
T
h
e
tech
n
i
q
u
es
m
en
tio
n
ed
h
a
v
e
b
ee
n
im
p
lem
e
n
ted
r
ec
e
n
tly
in
th
e
d
o
m
ain
o
f
s
p
ee
ch
d
e
tectio
n
to
f
o
cu
s
o
n
h
ig
h
er
em
o
tio
n
al
f
ea
tu
r
es
o
f
s
tam
m
er
s
an
d
ar
e
also
u
tili
ze
d
to
en
h
an
ce
StEn
s
em
b
leNe
t
to
em
p
h
asize
p
ar
ticu
la
r
r
eg
io
n
s
o
f
s
tam
m
er
s
th
at
d
o
n
o
t
h
av
e
f
lu
en
t
f
ea
tu
r
es.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
StEn
s
em
b
leNe
t
u
tili
ze
s
g
lo
b
al
atten
tio
n
m
o
d
els th
at
u
s
e
th
e
v
alu
es o
f
co
n
ce
ale
d
ce
lls
.
3
.
2
.
Ada
ptiv
e
g
ra
ph
t
o
po
lo
g
y
co
nv
o
lutio
n net
wo
r
k
T
o
an
al
y
ze
th
e
g
ait
r
ec
o
g
n
iti
o
n
,
t
h
is
r
esear
ch
wo
r
k
p
r
esen
ts
an
AGT
-
C
o
n
v
Net
to
m
o
d
el
th
e
g
ait
r
ec
o
g
n
itio
n
ac
r
o
s
s
th
e
d
ataset.
T
h
e
v
alu
e
o
f
th
e
f
in
al
r
esu
lt
in
lay
er
two
o
f
th
e
b
i
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
is
g
iv
e
n
as
i
u
an
d
D
u
is
u
s
ed
to
ex
p
r
ess
th
e
co
n
tex
tu
al
v
ec
to
r
th
at
is
ex
tr
ac
ted
b
y
th
e
a
tten
tio
n
m
o
d
el
th
at
g
en
er
ates th
e
last
class
if
ic
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
w
h
ich
is
d
en
o
ted
as
i
u
̃
.
T
h
e
f
o
r
m
u
latio
n
is
as g
iv
e
n
u
s
in
g
th
e
h
y
p
er
b
o
la
tan
f
u
n
cti
o
n
as
g
iv
en
in
(
6
)
.
i
u
̃
=
ta
n
h
ta
n
h
(
X
d
(
D
u
;
i
u
)
)
(
6
)
T
h
e
co
n
tex
tu
al
v
ec
to
r
f
o
r
th
e
a
tten
tio
n
m
eth
o
d
h
as
th
e
ad
d
e
d
weig
h
ts
o
f
all
th
e
c
o
n
ce
aled
ce
ll
o
u
tp
u
ts
o
f
th
e
e
n
co
d
er
.
A
v
ec
to
r
is
u
s
ed
as
a
lin
k
f
o
r
i
u
as
well
as
all
t
h
e
v
alu
es
o
f
th
e
co
n
ce
aled
ce
l
ls
th
at
ar
e
m
o
v
e
d
in
to
a
lay
e
r
with
s
o
f
tm
ax
t
o
d
e
p
ict
th
e
weig
h
ts
.
T
h
is
f
u
n
ctio
n
u
s
es
th
e
d
o
t
p
r
o
d
u
ct
in
th
e
atte
n
tio
n
m
et
h
o
d
.
T
h
is
is
f
o
r
m
u
lated
as g
iv
e
n
in
th
e
(
7
)
.
D
u
=
∑
u
j
=
1
i
j
(
(
f
i
u
T
.
i
j
)
(
∑
u
j
′
=
1
f
i
u
T
.
i
j
)
−
1
)
(
7
)
I
n
th
e
(
6
)
,
th
e
j
−
th
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
is
d
en
o
te
d
as
i
j
.
T
h
e
g
ait
p
r
o
f
ile
is
en
h
an
c
ed
u
s
in
g
d
u
al
-
s
p
lit
ar
ch
itectu
r
e
o
f
AGT
-
C
o
n
v
Net
f
o
r
s
k
eleta
l
m
o
tio
n
d
ata
J
is
co
n
s
id
er
ed
as
in
p
u
t
in
f
o
r
m
atio
n
.
T
h
e
f
ir
s
t
p
h
ase
o
f
th
e
ar
ch
it
ec
tu
r
e
is
b
ased
o
n
th
e
v
is
u
a
l
en
v
ir
o
n
m
e
n
ts
as
well
a
s
s
e
q
u
en
tial
attr
ib
u
tes.
Vis
u
al
-
en
h
an
ce
d
to
p
o
l
o
g
ical
lear
n
in
g
is
in
tr
o
d
u
ce
d
d
y
n
am
ically
an
d
d
en
o
ted
as
H
WB
.
T
h
e
en
h
an
ce
d
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
6
9
-
488
2
4874
to
p
o
lo
g
ical
lear
n
in
g
d
e
v
elo
p
e
d
is
u
s
ed
to
aid
a
c
o
m
b
in
e
d
a
g
g
r
eg
atio
n
f
o
r
g
r
ap
h
-
b
ased
c
o
n
v
o
lu
tio
n
s
in
th
e
n
ex
t
p
h
ase
o
f
th
e
a
r
ch
itectu
r
e.
C
r
o
s
s
-
en
tr
o
p
y
er
r
o
r
g
iv
en
as
M
DF
v
i
s
ual
is
u
s
e
d
to
e
x
am
in
e
v
is
u
al
ch
ar
a
cter
is
tic
lear
n
in
g
.
T
h
e
s
ec
o
n
d
p
h
ase
o
f
th
e
p
r
o
p
o
s
ed
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
v
ag
u
e
s
k
eleta
l
f
ea
tu
r
es
th
at
ar
e
f
ir
s
tly
o
b
tain
ed
u
s
in
g
a
lig
h
tweig
h
t
m
o
d
u
le
wh
ich
is
a
n
o
r
m
al
n
etwo
r
k
o
f
g
r
a
p
h
ical
co
n
v
o
l
u
tio
n
b
lo
ck
s
th
at
is
ex
p
r
ess
ed
in
(
1
)
.
Fu
r
t
h
er
,
th
e
b
lo
ck
s
o
f
th
e
AGT
-
C
o
n
v
Net
ar
e
g
ath
er
e
d
to
f
ilter
t
h
e
c
h
ar
ac
t
er
is
tic
f
ea
tu
r
es,
an
d
s
p
ec
if
ic
clu
es
ar
e
ex
tr
ac
ted
.
Fo
r
ev
er
y
s
eg
m
en
t,
a
jo
in
t
-
f
o
cu
s
ed
lear
n
in
g
f
ilter
is
im
p
lem
en
ted
to
p
r
o
d
u
ce
f
ilter
s
G
T
an
d
G
U
co
n
s
id
er
in
g
th
e
d
ep
t
h
.
Sp
ec
if
ically
,
G
T
alo
n
g
with
H
WB
f
r
o
m
th
e
v
is
u
al
en
h
an
ce
d
to
p
o
l
o
g
ical
lear
n
in
g
is
u
tili
ze
d
f
o
r
ex
t
r
a
ctio
n
s
p
atially
an
d
G
U
is
u
tili
ze
d
f
o
r
th
e
tem
p
o
r
al
s
y
s
tem
.
C
o
n
v
o
lu
tio
n
s
o
f
d
im
en
s
io
n
s
1
b
y
1
ar
e
im
p
le
m
en
ted
f
o
r
in
te
r
lin
k
ed
ch
an
n
el
d
ata.
Ho
wev
e
r
,
th
e
n
etwo
r
k
o
f
th
e
g
r
ap
h
ical
co
n
v
o
l
u
tio
n
p
r
o
ce
s
s
is
g
iv
en
i
n
th
e
(
8
)
.
Y
T
=
C
on
vol
ution
1
by
1
(
X
1
,
∑
L
T
l
=
1
H
WB
l
(
Y
⊗
G
T
l
)
)
(
8
)
Y
U
=
C
on
vol
ution
1
by
1
(
X
2
,
Y
T
⊗
G
U
)
C
o
n
s
id
er
in
g
th
e
e
q
u
atio
n
s
c
o
n
v
o
lu
tio
n
b
ased
o
n
d
ep
th
is
r
ep
r
esen
ted
as
⊗
,
H
WB
l
b
e
l
on
gs
to
S
O
×
O
,
G
T
l
b
e
l
on
gs
to
S
O
×
D
,
G
U
b
e
l
on
gs
to
S
L
U
×
O
×
D
′
,
X
1
b
e
l
on
gs
to
S
D
×
D
′
,
an
d
X
2
b
e
l
on
gs
to
S
D
′
×
D
′
.
Af
ter
th
e
b
lo
ck
s
o
f
th
e
AGT
-
C
o
n
v
Net
a
r
e
p
r
o
ce
s
s
ed
,
j
o
in
t
co
n
n
ec
tio
n
h
ier
ar
ch
ical
m
a
p
p
in
g
is
im
p
le
m
en
ted
f
o
r
th
e
g
ait
ch
ar
ac
ter
is
tic
f
ea
tu
r
es
to
b
e
m
ap
p
ed
in
to
s
ix
in
d
i
v
id
u
al
s
ca
les
th
at
ar
e
b
ased
o
n
jo
in
t
-
lin
k
e
d
ar
ch
itectu
r
e.
T
h
er
e
is
a
lo
s
s
o
f
t
h
r
ee
ele
m
en
ts
d
en
o
ted
as
M
t
hr
ee
an
d
a
cir
cu
lar
lo
s
s
d
en
o
ted
as
M
cir
cl
e
wh
ich
is
im
p
lem
en
ted
in
th
e
f
ea
tu
r
es in
th
e
o
u
tp
u
t
f
o
r
tr
ai
n
in
g
m
o
n
ito
r
in
g
.
T
h
e
e
n
tire
lo
s
s
er
r
o
r
f
u
n
ctio
n
is
ca
lcu
lated
in
(
9
)
.
M
=
∂
1
M
t
hr
ee
+
∂
2
M
cir
cl
e
+
∂
3
M
DF
V
i
s
ual
(
9
)
T
h
e
(
9
)
s
h
o
w
s
t
h
a
t
l
o
s
s
e
r
r
o
r
f
u
n
c
t
i
o
n
s
a
r
e
b
a
l
a
n
c
e
d
b
y
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s
∂
1
,
∂
2
,
∂
3
.
F
u
r
t
h
e
r
m
o
r
e
,
f
e
a
t
u
r
e
s
o
f
t
h
e
j
o
i
n
t
as
w
e
l
l
a
s
t
h
e
b
o
n
e
a
r
e
c
o
m
b
i
n
e
d
f
o
r
n
e
tw
o
r
k
s
t
o
id
e
n
t
i
f
y
t
h
e
a
c
t
i
v
it
i
es
o
f
h
u
m
a
n
s
,
t
w
o
d
i
s
ti
n
c
t
s
ta
g
e
s
a
r
e
u
s
e
d
t
o
s
e
p
a
r
at
e
l
y
e
x
t
r
ac
t
f
ea
t
u
r
e
s
o
f
t
h
e
b
o
n
e
a
n
d
t
h
e
j
o
i
n
t
a
n
d
t
h
e
r
e
s
u
l
ts
o
f
b
o
t
h
a
r
e
c
o
m
b
i
n
e
d
.
T
h
e
f
o
c
u
s
o
f
f
e
a
t
u
r
e
s
o
f
b
o
n
e
i
s
s
i
m
i
l
a
r
t
o
t
h
a
t
o
f
j
o
i
n
t
s
e
x
c
e
p
t
t
h
e
i
n
p
u
t
o
m
it
s
t
h
e
c
o
o
r
d
i
n
a
te
s
v
a
l
u
e
o
f
a
d
jac
e
n
t
j
o
i
n
ts
.
T
h
e
f
i
r
s
t
p
h
a
s
e
d
a
ta
is
t
h
e
f
e
at
u
r
e
s
f
r
o
m
jo
i
n
t
s
a
n
d
t
h
e
s
e
c
o
n
d
p
h
as
e
d
at
a
i
s
t
h
e
f
e
at
u
r
e
s
f
r
o
m
b
o
n
e
s
.
T
o
s
im
p
l
i
f
y
,
t
h
e
f
e
a
t
u
r
es
o
f
b
o
t
h
p
h
a
s
e
s
a
r
e
c
o
m
b
i
n
e
d
wi
t
h
t
h
e
d
i
m
e
n
s
i
o
n
o
f
1
2
b
y
D
o
ut
p
ut
a
n
d
i
s
i
m
p
le
m
e
n
t
e
d
as
t
h
e
o
u
t
p
u
t
.
3
.
3
.
Ada
ptiv
e
g
ra
ph
t
o
po
lo
g
y
co
nv
o
lutio
n net
wo
r
k
lea
rn
ing
I
t
is
o
b
s
er
v
ed
th
at
v
a
r
io
u
s
p
a
r
ts
o
f
th
e
b
o
d
y
s
h
o
w
v
ar
io
u
s
ex
ten
t
o
f
d
i
f
f
er
en
ce
as
well
a
s
lev
el
o
f
f
r
ee
d
o
m
co
n
s
id
er
in
g
th
e
s
k
ele
tal
s
tr
u
ctu
r
e,
th
e
jo
in
t
f
o
cu
s
ed
lear
n
in
g
f
ilter
is
u
tili
ze
d
to
e
x
p
lain
th
e
s
ep
ar
ate
ch
ar
ac
ter
is
tics
th
at
ar
e
tem
p
o
r
al
an
d
s
p
at
ial
f
o
r
v
a
r
io
u
s
g
ai
t
s
er
ies
b
y
d
ev
elo
p
in
g
c
u
s
to
m
f
ilter
s
.
Her
e,
two
d
is
tin
ct
p
h
ases
f
o
r
m
th
e
tem
p
o
r
al
an
d
s
p
atial
f
ilter
s
th
at
ar
e
u
s
ed
f
o
r
th
e
e
x
tr
ac
tio
n
o
f
s
p
a
tial
f
ea
tu
r
es
as
well
as th
e
ac
q
u
is
itio
n
o
f
tem
p
o
r
al
f
ea
tu
r
es.
T
h
e
ef
f
icien
cy
is
ex
p
an
d
ed
d
u
e
t
o
t
h
e
u
s
e
o
f
d
ep
th
-
b
ased
f
ilter
s
.
T
h
e
g
ait
c
h
ar
ac
ter
is
tics
Y
b
elo
n
g
s
to
S
U
×
O
×
D
is
u
tili
ze
d
f
o
r
in
p
u
t
an
d
th
e
en
h
a
n
ce
d
tem
p
o
r
al
p
o
o
lin
g
is
u
s
ed
to
g
ain
th
e
o
u
tp
u
t
Y
q
b
e
l
on
gs
to
S
U
q
×
O
×
D
,
in
th
is
ca
s
e
U
q
ex
p
r
ess
es
th
e
p
o
o
lin
g
d
im
en
s
io
n
.
C
o
n
s
id
er
in
g
tem
p
o
r
al
as
well
as
s
p
atial
p
h
ases
,
at
f
ir
s
t,
we
u
s
e
co
n
v
o
lu
t
io
n
s
th
at
ar
e
tem
p
o
r
al
f
o
r
lear
n
in
g
co
n
tex
t
-
b
ased
d
ata
at
ev
er
y
jo
in
t.
Fu
r
th
er
m
o
r
e,
p
o
o
lin
g
th
at
is
tem
p
o
r
al
is
ap
p
lied
in
th
e
s
p
atial
p
h
ase
to
g
ath
e
r
g
lo
b
al
tem
p
o
r
al
d
ata.
L
ater
,
two
f
u
lly
lin
k
ed
la
y
er
s
alo
n
g
with
a
lay
er
o
f
b
atc
h
n
o
r
m
aliza
tio
n
a
n
d
an
ac
tiv
at
io
n
f
u
n
ctio
n
R
eL
U
ar
e
u
tili
ze
d
f
o
r
th
e
c
o
n
s
tr
u
ct
io
n
o
f
in
ter
-
lin
k
ed
ch
a
n
n
el
c
o
m
m
u
n
icatio
n
an
d
r
esu
lt
in
a
f
ilter
g
t
th
at
h
as
d
im
en
s
io
n
s
1
by
O
by
(
L
t
by
D
)
.
Similar
ly
,
th
e
f
ilter
is
r
esh
ap
ed
in
to
d
im
en
s
io
n
s
L
t
by
O
by
D
th
at
u
s
es
b
atch
n
o
r
m
aliza
tio
n
th
at
o
m
its
s
m
all
o
r
lar
g
e
f
ilter
attr
ib
u
tes
to
b
e
u
s
ed
.
T
h
e
p
r
o
ce
d
u
r
e
th
at
o
cc
u
r
s
at
th
e
s
p
atial
p
h
ase
is
ev
alu
ated
u
s
in
g
th
e
(
1
0
)
.
g
T
=
G
(
G
(
UQ
(
UD
(
Y
q
)
)
,
X
3
)
,
X
4
)
(
1
0
)
G
T
=
B
a
tc
h
N
or
ma
l
iza
tion
(
R
e
s
ha
pin
g
(
g
t
)
)
I
n
(
3
)
,
UD
is
u
s
ed
to
d
e
n
o
te
co
n
v
o
lu
tio
n
th
at
is
tem
p
o
r
al,
a
c
o
m
p
letely
lin
k
ed
lay
er
is
ex
p
r
ess
ed
as
G
,
an
d
UQ
is
u
s
ed
to
ex
p
r
ess
tem
p
o
r
al
p
o
o
lin
g
.
Her
e,
X
3
b
e
l
on
gs
to
S
D
×
D
s
th
at
is
u
s
ed
t
o
d
ec
r
ea
s
e
th
e
s
ize
o
f
th
e
ch
an
n
el
b
y
s
r
atio
an
d
X
4
b
e
l
on
gs
to
S
D
s
×
D
g
ain
s
b
a
ck
th
e
s
ize
o
f
th
e
ch
a
n
n
el.
Fo
r
v
ar
io
u
s
s
p
atial
p
h
ases
,
th
e
tem
p
o
r
al
p
ar
t
is
u
tili
ze
d
to
ex
p
lain
th
e
m
o
v
in
g
f
ea
tu
r
es
th
at
ex
clu
d
e
th
e
tem
p
o
r
al
p
o
o
lin
g
t
o
r
etain
a
tem
p
o
r
al
ar
ch
itectu
r
e
as
well
as
im
p
lem
en
t
two
co
m
p
letely
lin
k
ed
lay
e
r
s
th
at
h
av
e
ac
tiv
atio
n
f
u
n
ctio
n
R
eL
U
f
o
r
tem
p
o
r
al
s
ize.
T
h
e
m
ain
aim
is
to
ac
h
ie
v
e
ef
f
ec
tiv
e
tem
p
o
r
al
lin
k
s
f
o
r
v
ar
io
u
s
m
o
tio
n
s
to
in
v
esti
g
at
e
m
o
v
i
n
g
f
ea
tu
r
es.
Fu
r
th
er
m
o
r
e
,
a
p
r
o
ce
s
s
o
f
n
o
r
m
aliza
tio
n
is
im
p
lem
en
te
d
to
g
u
ar
a
n
tee
th
at
th
e
f
ea
tu
r
e
d
i
s
tr
ib
u
tio
n
is
s
tab
le.
F
in
ally
,
th
e
p
r
o
ce
s
s
f
o
r
th
e
te
m
p
o
r
al
p
h
ase
is
g
iv
en
as
(
1
1
)
.
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
I
n
teg
r
a
tin
g
g
a
it a
n
d
s
p
ee
ch
d
y
n
a
mics me
th
o
d
o
lo
g
ies fo
r
en
h
a
n
ce
d
… (
R
a
vikira
n
R
ed
d
a
p
p
a
R
ed
d
y
)
4875
g
U
=
G
(
G
(
UD
(
Y
q
)
,
X
5
)
,
X
6
)
(
1
1
)
G
U
=
B
a
tc
h
N
or
ma
l
iza
tion
(
g
U
)
C
o
n
s
id
er
in
g
th
e
(
1
1
)
,
X
5
b
e
l
on
gs
to
S
U
Q
×
(
β
×
U
Q
)
ex
p
an
d
s
th
e
tem
p
o
r
al
s
ize
b
y
β
r
atio
an
d
X
6
b
e
l
on
gs
to
S
(
β
×
U
Q
)
×
L
U
d
ec
r
ea
s
es
it
to
a
p
ar
ticu
lar
d
i
m
en
s
io
n
L
U
.
T
h
er
e
ex
is
t
a
f
ew
m
eth
o
d
o
lo
g
ies
o
f
g
ait
o
n
a
v
is
u
al
b
asis
th
at
u
tili
ze
v
ar
io
u
s
s
tag
e
ap
p
r
o
ac
h
es
to
d
ev
elo
p
th
e
lo
ca
l
attr
ib
u
tes
o
f
th
e
m
o
d
el.
T
h
ese
tech
n
iq
u
es
s
h
o
w
s
o
m
e
s
im
ilar
ities
to
jo
in
t
f
o
cu
s
ed
lear
n
in
g
f
i
lter
.
T
h
e
jo
in
t
-
f
o
cu
s
ed
lear
n
in
g
f
ilter
is
co
m
p
a
r
ed
to
th
e
v
ar
io
u
s
s
tag
e
tech
n
iq
u
es,
wh
ich
y
ield
s
th
e
f
o
llo
win
g
r
esu
lts
:
i
)
th
e
d
if
f
er
en
t
s
tag
e
tech
n
iq
u
e
ex
tr
ac
ts
attr
ib
u
tes
o
f
v
ar
io
u
s
s
eq
u
en
ce
s
th
at
u
s
e
co
n
s
tan
t
co
n
v
o
lu
tio
n
s
,
in
th
is
ca
s
e,
jo
in
t
jo
i
n
t
-
f
o
c
u
s
ed
lear
n
in
g
f
ilte
r
th
e
attr
ib
u
tes
ex
tr
ac
ted
ar
e
f
o
r
ev
er
y
s
eq
u
e
n
ce
;
ii
)
th
e
v
ar
io
u
s
s
tag
e
tech
n
iq
u
e
u
tili
ze
s
co
n
v
o
lu
tio
n
s
th
at
ca
n
n
o
t
b
e
s
h
ar
ed
,
wh
er
e
th
e
attr
ib
u
tes
u
s
ed
ar
e
b
ig
g
er
th
an
th
e
co
n
v
o
lu
tio
n
s
.
W
h
er
ea
s
,
f
o
r
jo
in
t
f
o
cu
s
ed
lear
n
in
g
f
ilter
s
av
es
m
o
s
tly
h
alf
o
f
th
e
attr
ib
u
tes
an
d
m
ak
es
it
m
o
r
e
e
f
f
icie
n
t
;
an
d
iii
)
th
e
v
a
r
io
u
s
s
tag
e
tech
n
iq
u
e
attain
s
th
e
f
ea
tu
r
es
b
y
p
ar
titi
o
n
i
n
g
m
an
u
a
lly
,
wh
er
ein
th
e
s
em
an
tics
alig
n
m
en
t
is
n
o
t
ef
f
icien
t.
Ho
wev
e
r
,
th
e
jo
in
t
-
f
o
cu
s
ed
lear
n
in
g
f
ilter
h
as f
ea
tu
r
es th
at
ar
e
alig
n
ed
well
u
s
in
g
in
p
u
ts
f
r
o
m
th
e
s
k
eleto
n
.
3
.
4
.
Vis
ua
l
f
ea
t
ure
e
x
t
ra
ct
i
o
n
T
h
e
v
is
u
al
en
h
a
n
ce
d
to
p
o
lo
g
i
ca
l
lear
n
in
g
m
eth
o
d
u
s
es
in
h
er
en
t
v
is
u
al
d
ata
f
o
r
ev
e
r
y
s
eq
u
en
ce
f
o
r
lear
n
in
g
to
ad
a
p
t
to
v
is
u
al
en
v
ir
o
n
m
e
n
ts
.
C
o
n
s
id
er
J
as
th
e
in
itial
g
ait
s
eq
u
en
ce
th
at
is
at
tain
ed
th
r
o
u
g
h
th
e
in
p
u
t.
I
n
itially
,
th
e
m
et
h
o
d
is
ap
p
lied
f
o
r
t
h
e
ex
tr
ac
tio
n
o
f
v
is
u
al
f
ea
tu
r
es,
af
ter
wh
ich
g
lo
b
al
m
ea
n
p
o
o
lin
g
is
u
s
ed
to
co
m
b
i
n
e
th
e
g
lo
b
al
v
i
s
u
al
d
ata.
L
ater
,
a
co
m
p
letely
lin
k
ed
lay
er
is
u
s
ed
to
g
ain
th
e
v
is
u
al
class
if
y
in
g
v
ec
to
r
g
w
b
e
l
on
gs
to
S
L
W
,
th
e
v
is
u
al
co
u
n
t
is
d
e
n
o
ted
u
s
in
g
L
W
.
I
n
th
is
ca
s
e,
th
e
c
r
o
s
s
-
en
tr
o
p
y
er
r
o
r
f
o
r
g
w
r
esu
lts
in
a
lo
s
s
M
DF
v
i
s
ua
l
th
at
is
u
tili
ze
d
to
m
o
n
ito
r
f
ea
tu
r
e
attr
ib
u
te
lear
n
in
g
.
T
h
is
ass
u
r
es
ef
f
ec
tiv
e
v
is
u
al
p
r
ed
ictio
n
.
Similar
ly
,
t
h
e
f
u
n
c
tio
n
So
f
tMa
x
is
u
s
ed
to
r
esu
lt
in
a
n
o
r
m
alize
d
v
alu
e
v
ec
t
o
r
g
w
̃
b
e
l
on
gs
to
S
L
W
.
A
s
et
o
f
to
p
o
lo
g
ies
th
at
r
eq
u
i
r
e
to
b
e
lear
n
ed
ar
e
d
en
o
ted
as
H
s
et
=
{
H
W
1
,
H
W
2
,
…
.
,
H
W
L
W
}
th
at
ex
p
r
ess
es
th
e
m
o
d
if
icatio
n
o
f
v
is
u
al
en
h
an
c
ed
ab
ilit
y
,
wh
er
e
H
W
j
b
e
l
on
gs
to
S
L
T
×
O
×
O
r
ep
r
esen
ts
th
e
r
elate
d
to
p
o
l
o
g
y
g
ain
ed
at
th
e
j
−
th
v
is
u
al.
Fo
r
th
e
g
e
n
er
atio
n
o
f
v
is
u
ally
en
h
a
n
ce
d
to
p
o
lo
g
y
H
WB
,
th
e
h
ig
h
est
v
alu
e
i
n
d
ex
to
b
e
g
ain
ed
in
g
w
,
wh
ich
is
g
iv
en
as
(
1
2
)
.
I
n
th
e
(
5
)
,
ie
w
is
u
s
ed
to
d
e
n
o
te
t
h
e
s
eq
u
en
ce
th
at
h
as
th
e
h
ig
h
est
en
co
u
n
ter
s
.
Fu
r
th
er
,
th
e
r
elate
d
to
p
o
lo
g
y
is
ch
o
s
en
f
r
o
m
th
at
H
s
et
by
ie
w
as g
iv
en
in
(
1
3
)
.
ie
w
=
high
e
s
t
a
r
g
a
r
g
g
w
̃
(
1
2
)
H
1
=
H
s
et
[
ie
w
]
(
1
3
)
C
o
n
s
i
d
e
r
i
n
g
(
6
)
,
s
p
ec
i
f
i
c
f
e
a
t
u
r
e
s
a
r
e
d
e
n
o
t
e
d
u
s
i
n
g
H
1
t
o
c
o
r
r
e
s
p
o
n
d
i
n
g
v
i
s
u
a
ls
.
F
o
r
t
h
e
t
es
t
i
n
g
d
a
t
a
s
et
s
,
l
e
v
e
l
-
1
v
i
s
u
a
l
c
la
s
s
i
f
i
ca
t
i
o
n
w
it
h
h
i
g
h
a
c
c
u
r
a
c
i
e
s
o
f
9
8
%
i
s
s
h
o
w
n
t
o
b
e
r
e
l
i
a
b
l
e
.
H
e
n
c
e
,
f
o
r
e
v
e
r
y
s
e
q
u
e
n
c
e
,
k
n
o
w
n
a
s
t
h
e
r
e
s
u
lt
o
f
v
i
s
u
a
l
cla
s
s
i
f
i
c
at
i
o
n
p
r
e
d
i
c
t
i
o
n
,
a
n
a
c
c
u
r
a
t
e
c
h
o
i
c
e
c
a
n
b
e
m
a
d
e
f
o
r
e
n
h
a
n
c
e
d
t
o
p
o
l
o
g
y
f
o
r
t
h
e
r
e
l
a
t
e
d
v
is
u
a
l
.
W
e
a
s
s
u
m
e
t
h
a
t
t
h
e
i
n
i
ti
a
l
iz
a
t
i
o
n
o
f
l
e
a
r
n
i
n
g
a
t
t
r
i
b
u
t
e
s
i
s
d
o
n
e
t
h
r
o
u
g
h
t
o
p
o
l
o
g
y
,
a
n
d
u
p
d
a
t
e
s
c
a
n
b
e
m
a
d
e
t
o
m
o
d
i
f
y
t
h
e
v
is
u
a
l
c
h
a
r
a
c
te
r
i
s
t
i
cs
u
s
i
n
g
t
h
e
m
e
th
o
d
o
f
b
a
c
k
w
a
r
d
p
r
o
p
a
g
a
t
i
o
n
.
A
l
t
h
o
u
g
h
,
H
1
is
n
o
t
e
n
o
u
g
h
t
o
e
x
p
r
e
s
s
a
l
l
k
i
n
d
s
o
f
i
n
t
r
a
-
d
i
s
ti
n
c
t
e
x
i
s
t
e
n
c
e
f
o
r
t
h
i
s
v
i
s
u
a
l
,
t
h
e
r
e
f
o
r
e
a
n
a
u
g
m
e
n
t
a
t
i
o
n
i
s
i
n
t
r
o
d
u
c
e
d
.
W
e
a
s
s
u
m
e
t
h
e
d
i
s
t
r
i
b
u
t
i
o
n
o
f
i
n
f
o
r
m
a
t
i
o
n
i
n
g
w
̃
d
e
p
i
c
t
s
t
h
e
s
e
q
u
en
t
i
a
l
f
e
a
t
u
r
es
,
c
o
n
s
i
d
e
r
i
n
g
t
h
e
i
n
f
o
r
m
a
t
i
o
n
i
s
i
n
t
e
r
m
s
o
f
l
i
n
ea
r
w
e
i
g
h
ts
f
o
r
c
o
m
b
i
n
a
t
i
o
n
w
i
t
h
t
o
p
o
l
o
g
y
f
o
r
H
s
et
.
T
h
i
s
i
s
e
q
u
a
t
e
d
as
g
i
v
e
n
i
n
(
1
4
)
.
H
2
=
∑
L
W
j
=
1
g
̃
w
j
H
W
j
(
1
4
)
H
WB
=
h
1
H
1
+
h
2
H
2
+
h
3
H
3
(
1
5
)
Fu
r
th
er
,
H
3
is
th
e
co
n
s
tan
t
to
p
o
lo
g
y
th
at
is
u
s
ed
f
o
r
th
e
e
x
tr
ac
tio
n
o
f
n
o
r
m
al
f
ea
tu
r
es.
Fin
ally
,
H
1
,
H
2
,
H
3
is
co
m
b
in
ed
with
h
1
,
h
2
,
h
3
as
th
e
co
ef
f
icien
ts
th
at
r
esu
lt
in
H
WB
g
iv
en
i
n
(
1
5
)
.
T
h
e
jo
in
ts
o
f
t
h
e
b
o
d
y
ar
e
lin
k
ed
d
y
n
a
m
ically
u
s
in
g
co
n
v
o
lu
tio
n
s
in
th
e
g
r
ap
h
th
at
in
co
r
p
o
r
ate
d
ata
a
b
o
u
t
t
h
e
jo
in
ts
f
o
r
a
d
is
tan
t
r
an
g
e
as
well
as
h
av
e
a
jo
in
t
co
r
r
elatio
n
with
th
e
lo
ca
tio
n
s
.
T
h
e
lin
k
s
ar
e
d
y
n
a
m
ic
as
well
as
co
m
p
licated
wh
ich
ar
e
m
o
d
if
ie
d
ef
f
ec
tiv
ely
to
en
h
an
ce
th
e
ca
p
ac
ity
o
f
cr
o
s
s
-
v
is
u
al
en
v
ir
o
n
m
en
ts
.
4.
P
E
RF
O
RM
A
NCE
E
VA
L
U
AT
I
O
N
T
h
e
p
er
f
o
r
m
a
n
ce
ev
al
u
atio
n
e
n
co
m
p
ass
es
th
e
C
h
in
ese
Aca
d
em
y
o
f
Scien
ce
s
I
n
s
titu
te
o
f
A
u
to
m
atio
n
(
C
ASI
A
-
B
)
an
d
Osak
a
Un
iv
er
s
ity
m
u
lti
-
v
iew
lar
g
e
p
o
p
u
latio
n
(
OU
-
MV
L
P
)
d
atasets
f
o
r
g
ait
r
ec
o
g
n
itio
n
.
Var
io
u
s
m
etr
ics
ar
e
u
tili
ze
d
t
o
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s
ac
r
o
s
s
d
if
f
er
en
t
v
iewin
g
an
g
les.
I
n
th
e
s
p
ee
c
h
em
o
tio
n
r
ec
o
g
n
itio
n
d
o
m
ain
,
a
wid
e
ar
r
a
y
o
f
m
et
h
o
d
s
,
in
clu
d
in
g
class
im
b
alan
ce
h
a
n
d
lin
g
ap
p
r
o
ac
h
es,
ar
e
ev
alu
ate
d
o
n
t
h
e
SEP
-
2
8
k
(
Stu
tter
in
g
ev
e
n
ts
in
p
o
d
ca
s
ts
)
d
ataset,
co
n
s
id
er
in
g
m
etr
i
cs
s
u
ch
as
P
,
R
,
in
,
F1
-
Sco
r
e,
B
,
an
d
to
ta
l
ac
cu
r
ac
y
.
T
h
ese
ev
alu
atio
n
s
ar
e
co
m
p
ar
ed
with
th
e
s
tate
-
of
-
th
e
-
ar
t
tech
n
iq
u
es
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
6
9
-
488
2
4876
o
f
th
e
PS
in
a
d
d
r
ess
in
g
t
h
e
u
n
iq
u
e
ch
allen
g
es
p
o
s
ed
b
y
g
ait
an
d
s
p
ee
ch
r
ec
o
g
n
itio
n
task
s
an
d
th
e
r
esu
lts
ar
e
s
h
o
wn
in
th
e
f
o
r
m
o
f
g
r
ap
h
s
a
n
d
tab
les.
4
.
1
.
Da
t
a
s
et
det
a
ils
SEP
-
28k
[
6
]
:
T
h
e
s
elec
tio
n
p
r
o
ce
s
s
in
v
o
lv
ed
id
e
n
tify
in
g
a
n
d
in
clu
d
in
g
a
t
o
tal
o
f
3
8
5
p
o
d
ca
s
ts
to
f
o
r
m
th
e
SEP
-
2
8
k
s
tu
tter
in
g
d
ataset.
T
h
e
d
u
r
atio
n
o
f
th
e
in
itial
p
o
d
ca
s
t
r
ec
o
r
d
in
g
s
ex
h
ib
its
v
ar
ia
b
ilit
y
.
T
h
e
p
o
d
ca
s
t
ad
h
er
es
to
a
s
eq
u
en
tial
f
r
am
e
wo
r
k
,
co
m
p
r
is
in
g
a
d
iv
e
r
s
e
r
a
n
g
e
o
f
s
eg
m
en
ts
th
at
ca
n
s
p
a
n
f
r
o
m
4
0
to
2
5
0
in
n
u
m
b
er
.
E
ac
h
s
eg
m
e
n
t
is
ch
a
r
ac
ter
ized
b
y
a
p
r
ed
eter
m
in
e
d
d
u
r
atio
n
o
f
th
r
ee
s
ec
o
n
d
s
.
T
h
e
to
tal
n
u
m
b
er
o
f
s
eg
m
en
ts
o
b
tain
ed
is
2
8
,
1
7
7
.
T
h
e
SEP
-
2
8
k
d
ataset
co
n
s
is
t
s
o
f
two
d
is
tin
ct
lab
el
ca
teg
o
r
ies:
s
tu
tte
r
in
g
an
d
non
-
s
tu
tter
in
g
.
T
h
e
s
p
ee
ch
ca
teg
o
r
ies
th
at
d
o
n
o
t
d
em
o
n
s
tr
ate
s
tu
tter
in
g
an
d
a
r
e
n
o
t
p
er
tin
en
t
to
o
u
r
s
tu
d
y
en
co
m
p
ass
in
co
h
er
en
t
s
p
ee
ch
,
u
n
ce
r
tain
s
p
ee
ch
,
ab
s
en
ce
o
f
s
p
ee
ch
,
lo
w
au
d
io
q
u
ality
,
a
n
d
th
e
p
r
esen
ce
o
f
m
u
s
ic.
Stu
tter
in
g
is
a
s
p
ee
ch
d
is
o
r
d
er
th
at
en
co
m
p
ass
es
v
ar
io
u
s
s
p
ee
ch
b
eh
av
io
r
s
s
u
ch
as
b
l
o
ck
s
,
p
r
o
lo
n
g
atio
n
s
,
r
ep
etitio
n
s
,
in
ter
jectio
n
s
,
an
d
f
lu
en
t
s
eg
m
en
ts
.
T
h
e
d
atase
t
co
m
p
r
is
es
v
ar
io
u
s
s
p
ee
ch
t
y
p
es,
wh
ich
ar
e
as
f
o
llo
ws:
1
0
.
3
5
h
o
u
r
s
o
f
f
l
u
en
t
s
p
ee
ch
,
3
.
3
4
h
o
u
r
s
o
f
in
ter
j
ec
tio
n
s
,
2
.
7
4
h
o
u
r
s
o
f
r
e
p
etitio
n
,
1
.
4
8
h
o
u
r
s
o
f
p
r
o
lo
n
g
atio
n
,
1
.
7
5
h
o
u
r
s
o
f
b
l
o
ck
,
an
d
an
ad
d
itio
n
al
1
.
4
8
h
o
u
r
s
o
f
p
r
o
lo
n
g
atio
n
.
Af
ter
th
e
co
m
p
letio
n
o
f
th
e
tag
g
in
g
p
r
o
ce
s
s
,
ea
ch
v
o
ice
f
r
ag
m
en
t,
wh
ich
h
as
a
d
u
r
atio
n
o
f
3
s
ec
o
n
d
s
,
is
s
u
b
jecte
d
to
d
o
wn
s
am
p
lin
g
.
T
h
e
dow
n
s
am
p
lin
g
o
p
er
atio
n
r
ed
u
ce
s
th
e
f
r
eq
u
e
n
cy
o
f
th
e
v
o
ice
f
r
ag
m
en
t t
o
1
6
k
Hz.
C
ASI
A
-
B
[
2
0
]
:
T
h
e
C
ASI
A
-
B
d
ataset
is
em
p
lo
y
ed
f
o
r
b
o
t
h
th
e
tr
ain
i
n
g
a
n
d
test
in
g
p
h
a
s
es
o
f
th
e
ex
p
er
im
en
t
i
n
th
is
s
tu
d
y
.
T
h
e
C
ASI
A
-
B
d
atab
ase,
wh
ich
is
a
co
m
p
r
eh
en
s
iv
e
m
u
lti
-
v
iew
g
ait
d
atab
ase,
was
co
m
p
iled
an
d
i
n
itially
r
elea
s
ed
to
th
e
p
u
b
lic
in
2
0
0
5
.
T
h
e
s
tu
d
y
in
v
o
l
v
ed
a
co
h
o
r
t
o
f
1
2
4
p
a
r
ticip
an
ts
,
with
g
ait
d
ata
b
ein
g
c
o
llected
f
r
o
m
1
1
d
if
f
er
en
t
p
er
s
p
ec
tiv
es.
T
h
e
p
e
r
s
p
e
c
t
i
v
es
w
e
r
e
e
s
ta
b
l
is
h
e
d
b
y
u
t
i
l
i
zi
n
g
a
n
g
l
e
s
t
h
at
s
p
a
n
n
e
d
f
r
o
m
0
t
o
1
8
0
d
e
g
r
e
e
s
,
w
i
t
h
a
c
o
n
s
i
s
t
e
n
t
i
n
te
r
v
a
l
o
f
1
8
d
e
g
r
e
e
s
b
e
t
w
ee
n
e
a
c
h
p
e
r
s
p
e
cti
v
e
.
T
h
e
C
AS
I
A
-
B
s
y
s
tem
ca
teg
o
r
izes
walk
in
g
in
to
th
r
ee
d
is
tin
ct
s
ce
n
ar
io
s
:
n
o
r
m
al
walk
in
g
(
NM
)
,
walk
in
g
w
h
ile
ca
r
r
y
in
g
a
b
ag
(
B
G)
,
an
d
walk
in
g
wh
ile
wea
r
in
g
a
co
at
o
r
jack
et
(
C
L
)
.
T
h
e
f
r
am
e
r
ate
o
f
th
e
C
ASI
A
-
B
d
ataset
is
2
5
f
r
am
es
p
er
s
ec
o
n
d
(
FP
S).
T
h
e
d
ataset
co
m
p
r
is
es
a
cu
m
u
lativ
e
co
u
n
t
o
f
8
5
4
,
0
0
0
tr
ain
in
g
f
r
am
es
a
n
d
7
9
8
,
0
0
0
test
in
g
f
r
am
es.
T
h
e
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
o
l
o
g
y
u
tili
ze
d
in
th
is
s
tu
d
y
h
as
u
n
d
er
g
o
n
e
tr
ain
in
g
an
d
ev
alu
atio
n
u
s
in
g
th
e
C
ASI
A
-
B
d
ataset.
OU
-
MV
L
P
[
2
1
]
:
T
h
e
OU
-
MV
L
P
d
ataset
is
a
well
-
e
s
tab
lis
h
ed
b
en
ch
m
a
r
k
u
tili
ze
d
f
o
r
g
ait
id
en
tific
atio
n
,
wid
el
y
ac
k
n
o
wl
ed
g
ed
with
in
th
e
f
ield
.
T
h
e
d
a
taset
is
th
e
lar
g
est
in
ter
m
s
o
f
p
u
b
lic
a
v
ailab
ilit
y
,
en
co
m
p
ass
in
g
a
t
o
tal
o
f
1
0
,
3
0
7
d
is
tin
ct
p
ar
ticip
an
ts
.
T
h
e
d
at
aset
co
m
p
r
is
es
a
to
tal
o
f
1
4
p
e
r
s
p
ec
tiv
es,
s
p
an
n
in
g
f
r
o
m
0
to
9
0
d
e
g
r
ee
s
an
d
f
r
o
m
1
8
0
to
2
7
0
d
eg
r
ee
s
.
E
ac
h
v
iew
is
lin
k
ed
to
two
s
eq
u
e
n
ce
s
,
s
p
ec
if
ically
id
en
tifie
d
as
0
0
an
d
0
1
.
T
h
e
r
e
ar
e
a
to
tal
o
f
5
,
1
5
3
in
d
i
v
id
u
als
em
p
l
o
y
ed
s
p
ec
if
ically
f
o
r
test
in
g
p
u
r
p
o
s
es,
with
th
e
r
em
ain
in
g
5
,
1
5
4
s
u
b
jects a
llo
ca
ted
f
o
r
tr
ain
in
g
.
T
h
e
s
eq
u
e
n
c
es in
d
ex
ed
as
#
0
0
a
n
d
#
0
1
ar
e
ex
p
licitly
ass
ig
n
ed
as g
aller
ies an
d
p
r
o
b
es,
r
esp
ec
tiv
ely
.
4
.
2
.
Resul
t
s
f
o
r
g
a
it
o
n CA
SI
A
-
B
a
nd
O
U
-
M
VL
P
da
t
a
s
e
t
T
h
e
d
ata
i
n
Fig
u
r
e
1
s
h
o
wca
s
es
a
d
etailed
an
aly
s
is
o
f
d
if
f
e
r
en
t
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s
at
v
ar
io
u
s
o
r
ien
tatio
n
an
g
les,
ea
ch
co
r
r
e
s
p
o
n
d
in
g
to
a
d
eg
r
ee
f
r
o
m
0
°
to
1
8
0
°,
th
e
p
er
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
o
f
v
ar
io
u
s
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s
b
ased
o
n
d
if
f
er
en
t
m
etr
ics
is
ca
r
r
i
ed
o
u
t.
T
h
e
PS
ac
h
iev
es
r
em
ar
k
ab
ly
h
i
g
h
s
co
r
es,
co
n
s
is
ten
tly
ex
ce
ed
in
g
9
9
%
in
m
o
s
t
m
etr
ics,
w
h
ich
in
d
icate
s
its
ex
ce
p
tio
n
al
p
e
r
f
o
r
m
an
c
e
in
g
ait
r
ec
o
g
n
itio
n
.
Oth
er
m
eth
o
d
s
lik
e
B
iFu
s
io
n
[
2
2
]
also
p
er
f
o
r
m
well
b
u
t
f
a
ll
s
lig
h
tly
s
h
o
r
t
in
co
m
p
ar
is
o
n
with
th
e
PS
.
T
h
e
r
esu
lts
s
u
g
g
est
th
at
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
d
em
o
n
s
tr
ates
s
u
p
er
io
r
ef
f
ec
tiv
en
ess
an
d
ac
cu
r
ac
y
in
g
ait
r
ec
o
g
n
itio
n
co
m
p
ar
ed
to
t
h
e
o
th
e
r
m
eth
o
d
s
.
I
n
co
m
p
ar
is
o
n
with
th
e
e
x
is
tin
g
s
y
s
tem
(
ES
)
th
e
ac
cu
r
ac
y
v
alu
e
r
an
g
es
f
r
o
m
9
6
to
9
9
in
c
o
m
p
ar
is
o
n
with
th
e
PS
w
h
ic
h
g
en
er
ates a
h
ig
h
er
v
alu
e
f
o
r
v
ar
io
u
s
s
ettin
g
v
alu
e
s
.
Fig
u
r
e
1
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
s
tate
-
of
-
ar
t te
ch
n
i
q
u
es with
PS
f
o
r
NM
#
5
-
6
75
80
85
90
95
100
105
0°
18°
36°
54°
72°
90°
108°
126°
144°
162°
180°
v
alu
e
Angle
G
ait
s
e
t
MT
3D
G
ait
G
L
G
ait
P
art
BiF
u
s
i
o
n
G
ait
T
A
KE
(
E
S
)
Pro
p
o
s
e
d
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
I
n
teg
r
a
tin
g
g
a
it a
n
d
s
p
ee
ch
d
y
n
a
mics me
th
o
d
o
lo
g
ies fo
r
en
h
a
n
ce
d
… (
R
a
vikira
n
R
ed
d
a
p
p
a
R
ed
d
y
)
4877
T
h
e
p
r
o
v
id
e
d
d
ata
p
r
esen
ts
th
e
m
ea
n
p
er
f
o
r
m
a
n
ce
o
f
d
if
f
er
en
t
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s
.
I
n
Fig
u
r
e
2
ea
ch
m
eth
o
d
is
co
m
p
a
r
ed
alo
n
g
with
th
e
PS
f
o
r
th
e
r
esp
ec
tiv
e
m
ea
n
r
ec
o
g
n
itio
n
s
co
r
e.
T
h
e
PS
g
iv
es
b
etter
p
er
f
o
r
m
an
ce
d
ep
ictin
g
a
m
ea
n
s
co
r
e
o
f
9
9
.
6
.
B
iFu
s
io
n
[
2
2
]
also
d
em
o
n
s
tr
ates
s
tr
o
n
g
p
er
f
o
r
m
an
ce
with
a
m
ea
n
s
co
r
e
o
f
9
8
.
7
.
Oth
er
m
eth
o
d
s
li
k
e
GaitGL
[
2
3
]
,
GaitPar
t
[
2
4
]
,
GaitT
AKE
(
E
S)
[
2
5
]
,
m
o
d
u
lar
th
r
ee
-
d
im
en
s
io
n
al
m
u
ltis
p
ec
ies
(
MT
3
D)
[
2
6
]
,
an
d
Gaitset
[
2
7
]
ex
h
ib
it
m
ea
n
s
co
r
es
r
an
g
in
g
f
r
o
m
9
5
.
7
to
9
7
.
5
,
r
ef
lectin
g
th
eir
r
esp
ec
tab
le
r
ec
o
g
n
itio
n
ca
p
a
b
i
liti
es.
Fig
u
r
e
2
.
Me
an
v
alu
e
c
o
m
p
ar
i
s
o
n
f
o
r
NM
#
5
-
6
In
Fig
u
r
e
3
th
e
r
esu
lts
ar
e
d
i
s
p
lay
ed
f
o
r
B
G#
1
-
2
,
th
e
PS
co
n
s
is
ten
tly
en
s
u
r
es
b
etter
p
e
r
f
o
r
m
a
n
ce
,
ac
h
iev
in
g
t
h
e
h
ig
h
est
r
ec
o
g
n
it
io
n
s
co
r
es
ac
r
o
s
s
m
o
s
t
an
g
les.
T
h
is
h
ig
h
lig
h
ts
its
r
em
ar
k
a
b
le
co
n
s
is
ten
cy
an
d
ef
f
ec
tiv
en
ess
in
r
ec
o
g
n
izin
g
g
ait
p
atter
n
s
,
with
s
co
r
es
r
e
ac
h
in
g
as
h
ig
h
as
9
9
.
8
4
%.
B
iFu
s
io
n
[
2
2
]
also
d
em
o
n
s
tr
ates
s
tr
o
n
g
p
er
f
o
r
m
a
n
ce
,
p
a
r
ticu
lar
ly
at
0
°,
1
8
°,
a
n
d
5
4
°
an
g
les,
with
r
ec
o
g
n
itio
n
s
co
r
es
ab
o
v
e
9
7
%.
Me
an
wh
ile,
GaitT
AKE
(
E
S)
[
2
5
]
d
eliv
e
r
s
b
etter
r
esu
lts
,
es
p
ec
ially
at
3
6
°,
5
4
°,
an
d
7
2
°
an
g
les,
with
s
co
r
es
p
ea
k
in
g
at
9
9
%.
Oth
er
m
eth
o
d
s
lik
e
MT
3
D
[
2
6
]
,
GaitGL
[
2
3
]
,
Gaitset
[
2
7
]
,
an
d
GaitPar
t
[
2
4
]
o
f
f
er
v
ar
y
in
g
p
er
f
o
r
m
an
ce
,
r
an
g
i
n
g
f
r
o
m
8
4
.
9
%
to
9
8
.
8
%.
I
n
c
o
n
clu
s
io
n
,
th
e
PS
g
iv
es
b
etter
p
er
f
o
r
m
an
ce
in
c
o
m
p
ar
is
o
n
with
th
e
ES
.
Fig
u
r
e
4
en
s
u
r
es
th
e
s
u
m
m
ar
y
o
f
th
e
p
er
f
o
r
m
an
ce
m
ea
n
ac
r
o
s
s
v
ar
io
u
s
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s,
PS
en
s
u
r
es
b
est
p
er
f
o
r
m
an
ce
wit
h
a
m
ea
n
s
co
r
e
o
f
9
8
.
8
6
.
B
iFu
s
io
n
[
2
2
]
also
ex
h
i
b
its
s
tr
o
n
g
p
er
f
o
r
m
an
ce
with
a
m
ea
n
s
co
r
e
o
f
9
6
,
s
ig
n
if
y
in
g
i
ts
ef
f
ec
tiv
en
ess
in
g
ait
r
e
co
g
n
i
tio
n
.
GaitT
AKE
(
E
S)
[
2
5
]
f
o
ll
o
ws
clo
s
ely
with
a
m
ea
n
s
co
r
e
o
f
9
7
.
5
,
em
p
h
asizin
g
its
r
eliab
ilit
y
in
r
ec
o
g
n
izin
g
g
ait
p
atter
n
s
.
Me
th
o
d
s
s
u
ch
as
MT
3
D
[
2
6
]
an
d
GaitGL
[
2
7
]
d
em
o
n
s
tr
ate
co
n
s
is
ten
t
p
er
f
o
r
m
an
ce
with
m
ea
n
s
co
r
es
o
f
9
4
.
5
,
u
n
d
er
s
co
r
in
g
th
eir
s
u
itab
ilit
y
f
o
r
g
ait
r
ec
o
g
n
itio
n
task
s
.
Gaitset
[
2
7
]
an
d
GaitPar
t
[
2
4
]
d
eliv
e
r
m
ea
n
s
co
r
es
o
f
9
3
a
n
d
9
1
.
5
,
r
esp
ec
tiv
ely
.
T
h
is
d
ata
in
d
icate
s
th
at
th
e
PS
s
tan
d
s
o
u
t
as
th
e
m
o
s
t
e
f
f
ec
tiv
e
g
ait
r
ec
o
g
n
itio
n
ap
p
r
o
ac
h
,
wit
h
th
e
h
ig
h
est
m
ea
n
s
co
r
e,
s
u
g
g
esti
n
g
its
s
tr
o
n
g
p
o
ten
tial f
o
r
ap
p
licatio
n
s
in
b
io
m
etr
ics,
s
ec
u
r
ity
,
o
r
s
u
r
v
eill
an
ce
.
Fig
u
r
e
3
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
s
tate
-
of
-
ar
t te
ch
n
i
q
u
es with
PS
f
o
r
B
G#
1
-
2
93
94
95
96
97
98
99
100
G
ait
s
e
t
MT
3D
G
ait
G
L
G
ait
P
art
BiF
u
s
i
o
n
G
ait
T
A
KE
(
E
S
)
Pro
p
o
s
e
d
Me
an
v
alu
e
Me
th
o
d
o
logy
85
90
95
100
0°
18°
36°
54°
72°
90°
108°
126°
144°
162°
180°
v
alu
e
An
gle
G
ait
s
e
t
MT
3D
G
ait
G
L
G
ait
P
art
BiF
u
s
i
o
n
G
ait
T
A
KE
(
E
S
)
Pro
p
o
s
e
d
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.
13
,
No
.
4
,
Dec
em
b
er
2
0
2
4
:
4
8
6
9
-
488
2
4878
Fig
u
r
e
4
.
Me
an
v
alu
e
c
o
m
p
ar
i
s
o
n
f
o
r
B
G#
1
-
2
I
n
F
ig
u
r
e
5
th
e
an
aly
s
is
is
ca
r
r
ied
o
u
t
f
o
r
C
L
#
1
-
2
wh
ich
s
h
o
ws
th
at
th
e
P
S
co
n
s
is
ten
tly
ac
h
iev
es
th
e
h
ig
h
est
r
ec
o
g
n
itio
n
r
ates
ac
r
o
s
s
d
if
f
er
en
t
an
g
les,
with
v
alu
es
well
ab
o
v
e
9
0
%.
I
n
co
m
p
ar
is
o
n
,
s
o
m
e
o
th
er
m
eth
o
d
s
,
s
u
ch
as
Gaitset
[
2
7
]
an
d
MT
3
D
[
2
6
]
,
ex
h
i
b
it
s
im
ilar
p
er
f
o
r
m
an
ce
,
m
ain
tain
i
n
g
r
ec
o
g
n
itio
n
r
ates
ar
o
u
n
d
8
0
%
at
m
o
s
t
an
g
les.
GaitGL
[
2
3
]
d
em
o
n
s
tr
ates
s
lig
h
tly
b
etter
ac
cu
r
ac
y
,
wit
h
r
ec
o
g
n
itio
n
r
ates
ex
ce
ed
in
g
9
0
%
at
s
p
ec
if
ic
an
g
les.
GaitPar
t
[
2
4
]
la
g
s
b
eh
in
d
t
h
e
r
est,
with
r
ec
o
g
n
itio
n
r
ates
r
an
g
in
g
f
r
o
m
6
6
.
5
%
to
8
6
.
9
%.
B
iFu
s
io
n
[
2
2
]
p
er
f
o
r
m
s
in
an
av
e
r
ag
e
m
an
n
e
r
,
s
u
r
p
ass
in
g
9
5
%
ac
cu
r
ac
y
at
s
ev
er
al
an
g
les.
GaitT
AKE
(
E
S)
[
2
5
]
also
ac
h
iev
es st
r
o
n
g
r
ec
o
g
n
itio
n
r
ates,
with
a
d
ip
to
8
7
% a
t 1
8
0
°.
I
n
co
n
clu
s
io
n
,
th
e
PS
o
u
tp
er
f
o
r
m
s
o
th
er
m
eth
o
d
s
,
s
h
o
wca
s
in
g
h
i
g
h
er
ac
cu
r
ac
y
a
n
d
r
o
b
u
s
tn
ess
in
g
ait
r
ec
o
g
n
itio
n
ac
r
o
s
s
v
ar
io
u
s
v
iewin
g
an
g
les.
Fig
u
r
e
6
p
r
esen
ts
a
co
m
p
a
r
i
s
o
n
o
f
v
ar
i
o
u
s
g
ait
r
ec
o
g
n
itio
n
m
eth
o
d
s
,
f
o
cu
s
in
g
o
n
th
e
ir
o
v
er
all
p
er
f
o
r
m
an
ce
f
o
r
th
e
m
ea
n
s
co
r
e.
T
h
e
PS
s
h
o
wca
s
es
a
m
ea
n
r
ec
o
g
n
itio
n
r
ate
o
f
9
5
.
3
7
%.
I
t
d
em
o
n
s
tr
ates
th
e
h
ig
h
est ac
cu
r
ac
y
am
o
n
g
all
m
eth
o
d
s
,
in
d
icatin
g
its
ef
f
ec
tiv
en
ess
in
g
ait
r
ec
o
g
n
itio
n
.
GaitT
AKE
(
E
S)
[
2
5
]
an
d
B
iFu
s
io
n
[
2
2
]
also
p
er
f
o
r
m
ex
ce
p
tio
n
ally
well,
ac
h
iev
in
g
r
e
co
g
n
itio
n
r
ates o
f
9
2
.
2
% a
n
d
9
2
.
1
%,
r
esp
ec
tiv
ely
.
GaitGL
[
2
3
]
f
o
llo
ws
clo
s
ely
with
a
r
ec
o
g
n
itio
n
r
ate
o
f
8
3
.
6
%,
s
u
r
p
ass
in
g
Gaitset
[
2
7
]
an
d
MT
3
D
[
2
6
]
,
wh
ich
b
o
th
ex
h
ib
it a
m
ea
n
r
ec
o
g
n
itio
n
r
ate
o
f
8
1
.
5
%.
GaitPar
t [
2
6
]
lag
s
,
with
a
m
ea
n
r
ec
o
g
n
itio
n
r
ate
o
f
7
8
.
7
%.
T
h
e
r
esu
lts
s
u
g
g
est
th
at
th
e
PS
o
f
f
er
s
a
s
ig
n
if
ican
t
ad
v
an
ce
m
e
n
t
in
g
ait
r
ec
o
g
n
itio
n
,
o
u
t
p
er
f
o
r
m
in
g
its
co
m
p
etito
r
s
in
ter
m
s
o
f
ac
cu
r
ac
y
a
n
d
r
elia
b
ilit
y
.
Fig
u
r
e
5
.
Acc
u
r
ac
y
c
o
m
p
ar
is
o
n
o
f
s
tate
-
of
-
ar
t te
ch
n
i
q
u
es with
PS
f
o
r
C
L
#
1
-
2
Fig
u
r
e
6
.
Me
an
v
alu
e
c
o
m
p
ar
i
s
o
n
f
o
r
C
L
#
1
-
2
86
88
90
92
94
96
98
100
G
ai
t
se
t
M
T
3
D
G
ai
t
G
L
Gai
t
P
ar
t
B
i
F
us
i
o
n
G
ai
t
T
A
K
E
(
E
S
)
P
r
o
po
se
d
Me
an
valu
e
Me
th
o
d
0
20
40
60
80
100
120
0°
18°
36°
54°
72°
90°
108°
126°
144°
162°
180°
v
alu
e
A
n
g
l
e
G
ait
s
e
t
MT
3D
G
ait
G
L
G
ait
P
art
BiF
u
s
i
o
n
Gait
T
A
KE (E
S
)
P
ro
p
o
s
ed
0
20
40
60
80
100
120
G
ait
s
e
t
MT
3D
G
ait
G
L
G
ait
P
art
BiF
u
s
i
o
n
G
ait
T
A
KE
(
E
S
)
Pro
p
o
s
e
d
CL
#
1-2
Me
an
v
alu
e
Me
th
o
d
Evaluation Warning : The document was created with Spire.PDF for Python.