I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
24
,
p
p
.
1
4
4
8
~
1
459
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
3
6
.
i
3
.
pp
1
4
4
8
-
1
4
5
9
1448
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
Deep
-
S
FER:
d
ee
p
conv
o
lutiona
l ne
ura
l net
wo
rk
and
MFCC
an
eff
ec
tive spee
ch a
nd f
a
ce
emo
tion r
ecog
nition
Ra
v
i G
um
m
ula
,
Vino
t
hk
um
a
r
Arum
ug
a
m,
Abila
s
ha
Ara
ng
a
na
t
ha
n
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
mm
u
n
i
c
a
t
i
o
n
En
g
i
n
e
e
r
i
n
g
,
D
r
.
M
.
G
.
R
.
E
d
u
c
a
t
i
o
n
a
l
a
n
d
R
e
se
a
r
c
h
I
n
st
i
t
u
t
e
,
C
h
e
n
n
a
i
,
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
Mar
13
,
2
0
2
4
R
ev
is
ed
Au
g
22
,
202
4
Acc
ep
ted
A
ug
31
,
2
0
2
4
Th
e
re
h
a
s
b
e
e
n
a
lo
t
o
f
p
r
o
g
re
ss
in
re
c
e
n
t
y
e
a
rs
in
th
e
field
s
o
f
e
x
p
e
rt
sy
ste
m
s,
a
rti
ficia
l
in
telli
g
e
n
c
e
(
AI)
a
n
d
h
u
m
a
n
m
a
c
h
in
e
in
terfa
c
e
(HMI
).
Th
e
u
se
o
f
v
o
ice
c
o
m
m
a
n
d
s
to
e
n
g
a
g
e
wit
h
m
a
c
h
in
e
ry
o
r
in
stru
c
t
it
to
d
o
a
c
e
rtain
tas
k
is
b
e
c
o
m
in
g
m
o
re
c
o
m
m
o
n
.
Nu
m
e
ro
u
s
c
o
n
su
m
e
r
e
lec
tro
n
ics
h
a
v
e
S
IRI
,
Ale
x
a
,
C
o
rtan
a
,
a
n
d
G
o
o
g
le
As
sista
n
t
b
u
il
t
in
.
I
n
t
h
e
field
o
f
h
u
m
a
n
-
d
e
v
ice
in
tera
c
ti
o
n
,
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
fr
o
m
sp
e
e
c
h
is
a
c
o
m
p
lex
re
se
a
rc
h
su
b
jec
t.
We
c
a
n
'
t
im
a
g
in
e
m
o
d
e
rn
l
ife
with
o
u
t
m
a
c
h
in
e
s,
so
n
a
tu
ra
ll
y
t
h
e
re
'
s
a
n
e
e
d
to
c
re
a
te
a
m
o
re
ro
b
u
st
fra
m
e
wo
rk
f
o
r
h
u
m
a
n
-
m
a
c
h
in
e
c
o
m
m
u
n
ica
ti
o
n
.
A
n
u
m
b
e
r
o
f
a
c
a
d
e
m
ics
a
re
n
o
w
w
o
rk
i
n
g
o
n
sp
e
e
c
h
e
m
o
ti
o
n
re
c
o
g
n
it
i
o
n
(
S
E
R)
in
a
n
e
ffo
rt
t
o
imp
ro
v
e
th
e
i
n
tera
c
ti
o
n
b
e
twe
e
n
h
u
m
a
n
s
a
n
d
m
a
c
h
in
e
s.
We
a
ime
d
to
id
e
n
ti
fy
f
o
u
r
fu
n
d
a
m
e
n
tal
e
m
o
ti
o
n
s:
a
n
g
ry
,
u
n
h
a
p
p
y
,
n
e
u
tra
l
a
n
d
j
o
y
f
u
l
fro
m
s
p
e
e
c
h
i
n
o
u
r
e
x
p
e
rime
n
t.
As
y
o
u
c
a
n
h
e
a
r
b
e
l
o
w,
we
trai
n
e
d
a
n
d
tes
ted
o
u
r
m
o
d
e
l
u
si
n
g
a
u
d
io
d
a
ta
o
f
b
rief
M
a
n
i
p
u
ri
s
p
e
e
c
h
e
s
tak
e
n
fro
m
fil
m
s.
Th
is
tas
k
m
a
k
e
s
u
se
o
f
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
(
CNN
s)
to
e
x
trac
t
fu
n
c
ti
o
n
s
fr
o
m
sp
e
e
c
h
in
o
rd
e
r
to
re
c
o
g
n
ize
d
iffere
n
t
m
o
o
d
s
u
si
n
g
t
h
e
M
e
l
-
fre
q
u
e
n
c
y
c
e
p
stra
l
c
o
e
fficie
n
t
(M
F
CC).
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
Facial
em
o
tio
n
r
ec
o
g
n
itio
n
Facial
ex
p
r
ess
io
n
r
ec
o
g
n
itio
n
I
m
ag
e
r
ec
o
g
n
itio
n
MFC
C
Sp
ee
ch
em
o
tio
n
r
ec
o
g
n
itio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
R
av
i G
u
m
m
u
la
Dep
ar
tm
e
n
t
o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
i
n
g
Dr
.
M.
G.
R
.
E
d
u
ca
tio
n
al
a
n
d
R
esear
ch
I
n
s
titu
te
C
h
en
n
ai,
I
n
d
ia
E
m
ail:
r
av
i.g
u
m
m
u
la@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Han
d
g
estu
r
es,
f
ac
ial
f
ea
t
u
r
es
an
d
v
o
ca
l
in
f
lectio
n
s
ar
e
all
p
o
ten
tial
in
p
u
ts
a
n
d
em
o
tio
n
s
t
h
at
m
ay
b
e
au
to
m
atica
lly
d
etec
ted
v
ia
th
e
th
r
ee
s
tep
s
o
f
f
ea
tu
r
e
ex
t
r
a
ctio
n
an
d
class
if
icatio
n
[
1
]
.
Desp
ite
s
ig
n
if
ican
t
ad
v
an
ce
m
e
n
ts
s
in
ce
th
e
d
awn
o
f
th
e
m
o
u
s
e
a
n
d
k
ey
b
o
ar
d
,
au
to
m
ated
s
p
ee
ch
r
ec
o
g
n
itio
n
an
d
ac
ce
s
s
ib
le
u
s
er
in
ter
f
ac
es
f
o
r
in
d
iv
id
u
als
wit
h
d
is
ab
ilit
ies,
s
tate
-
of
-
th
e
-
a
r
t
h
u
m
an
u
s
er
in
ter
f
ac
es
o
f
ten
f
ail
to
f
u
lly
le
v
er
ag
e
th
ese
v
ital
in
ter
ac
tiv
e
ca
p
ab
i
liti
es,
r
esu
ltin
g
in
less
th
an
s
atis
f
ac
to
r
y
ex
p
er
ien
ce
s
.
M
ac
h
in
es
th
at
c
o
u
l
d
in
ter
p
r
et
th
ese
em
o
tio
n
al
c
u
es
m
ay
tailo
r
th
eir
ass
is
tan
ce
to
ea
ch
in
d
iv
i
d
u
al'
s
u
n
iq
u
e
tast
es
an
d
r
eq
u
ir
e
m
en
ts
.
Acc
o
r
d
in
g
to
s
tu
d
ies
in
p
s
y
ch
o
lo
g
y
,
th
er
e
ar
e
s
ix
b
asic
h
u
m
an
em
o
tio
n
s
:
h
ap
p
in
ess
,
s
ad
n
ess
,
f
ea
r
,
wr
ath
,
d
is
g
u
est
,
an
d
s
h
o
ck
.
T
h
er
e
ar
e
ce
r
tain
em
o
tio
n
s
th
at
ca
n
o
n
ly
b
e
ex
p
r
ess
ed
v
ia
s
p
ec
if
ic
f
ac
ial
g
estu
r
es
an
d
v
o
ice
in
to
n
atio
n
[
2
]
-
[
4
]
.
E
m
o
tio
n
in
te
r
p
r
etatio
n
is
a
p
r
o
m
is
in
g
n
ew
f
ield
th
at
co
u
ld
p
r
o
v
id
e
an
s
wer
s
to
m
a
n
y
m
y
s
ter
ies
[
5
]
.
A
p
e
r
s
o
n
'
s
f
a
c
i
a
l
e
x
p
r
es
s
i
o
n
s
c
o
n
v
e
y
b
o
t
h
s
p
o
k
e
n
a
n
d
n
o
n
v
e
r
b
a
l
m
e
s
s
a
g
es
a
b
o
u
t
t
h
e
i
r
e
m
o
t
i
o
n
a
l
s
t
a
te
[
6
]
,
[
7
]
.
I
t
is
p
o
s
s
ib
le
to
in
ter
p
r
et
em
o
tio
n
al
s
tates
f
r
o
m
a
v
ar
iety
o
f
s
o
u
r
ce
s
,
s
u
ch
as
v
er
b
al,
tex
tu
al
an
d
v
is
u
al
cu
es.
A
lo
n
g
-
s
tan
d
i
n
g
talen
t
h
as
all
o
wed
u
s
t
o
co
m
p
r
e
h
en
d
m
an
y
th
in
g
s
,
i
n
clu
d
in
g
p
e
o
p
le'
s
m
in
d
s
:
th
e
ca
p
ac
ity
to
in
ter
p
r
et
th
eir
f
ee
lin
g
s
v
ia
th
e
ir
wo
r
d
s
a
n
d
ex
p
r
ess
io
n
s
.
R
e
ad
in
g
b
etwe
en
th
e
lin
es
o
f
th
ese
s
en
tim
en
ts
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Deep
-
S
F
E
R
:
d
ee
p
c
o
n
vo
l
u
tio
n
a
l n
eu
r
a
l
n
etw
o
r
k
a
n
d
MFCC
a
n
effec
tive
…
(
R
a
vi
Gu
mm
u
la
)
1449
wo
r
d
s
is
an
in
cr
ed
ib
ly
d
if
f
icu
lt
an
d
tim
e
-
co
n
s
u
m
in
g
en
d
ea
v
o
u
r
.
A
n
u
m
b
er
o
f
s
cien
tific
d
is
cip
lin
es
ar
e
n
o
w
co
llab
o
r
atin
g
o
n
a
s
o
lu
tio
n
to
th
is
p
r
o
b
lem
:
an
im
p
r
o
v
ed
m
eth
o
d
o
f
em
o
tio
n
d
etec
tio
n
f
r
o
m
a
v
a
r
iety
o
f
s
o
u
r
ce
s
,
in
clu
d
in
g
f
ac
ial
ex
p
r
e
s
s
io
n
s
an
d
v
o
ice
[
8
]
.
T
h
e
in
tr
o
d
u
ctio
n
o
f
a
r
tific
ial
in
tellig
en
t
(
AI
)
an
d
n
at
u
r
al
lan
g
u
ag
e
m
o
d
ellin
g
s
y
s
tem
s
h
a
v
e
im
p
r
o
v
ed
th
e
ac
cu
r
ac
y
o
f
th
is
r
ea
ctio
n
t
o
v
a
r
io
u
s
v
o
ca
l
-
b
ased
tech
n
i
q
u
es
an
d
u
tter
an
ce
s
[
9
]
.
W
h
en
i
n
d
o
u
b
t,
it
c
o
u
ld
b
e
wis
e
to
ex
am
in
e
f
ee
lin
g
s
.
On
e
s
u
ch
is
s
u
e
is
th
e
u
s
e
o
f
h
u
m
an
-
co
m
p
u
ter
co
o
p
er
atio
n
.
C
o
m
p
u
ter
s
ca
n
m
a
k
e
b
etter
d
ec
is
io
n
s
,
h
el
p
co
n
s
u
m
er
s
id
en
tify
em
o
tio
n
s
,
an
d
f
ac
ilit
ate
th
e
d
e
v
elo
p
m
e
n
t
o
f
m
o
r
e
r
ea
lis
tic
in
ter
ac
tio
n
s
b
etwe
en
h
u
m
an
s
an
d
r
o
b
o
ts
.
W
e
wo
u
ld
ex
a
m
in
e
th
e
c
u
r
r
en
t
s
tate
o
f
e
m
o
tio
n
r
ec
o
g
n
itio
n
m
eth
o
d
s
,
em
o
tio
n
m
o
d
els
an
d
em
o
tio
n
d
atab
ases
,
g
o
in
g
o
v
e
r
th
eir
p
r
o
s
an
d
c
o
n
s
an
d
wh
at
th
e
f
u
tu
r
e
h
o
ld
s
in
ter
m
s
o
f
th
ese
ar
ea
s
[
1
0
]
.
On
e
o
f
o
u
r
m
ain
a
r
ea
s
o
f
r
esear
c
h
is
em
o
tio
n
ass
ess
m
en
t
v
ia
th
e
u
s
e
o
f
v
o
ice
an
d
f
ac
ial
r
ec
o
g
n
itio
n
.
W
e
ex
am
i
n
ed
th
e
m
an
y
tech
n
ical
co
m
p
o
n
en
ts
th
at
co
m
p
r
is
e
co
n
tem
p
o
r
ar
y
m
eth
o
d
s
an
d
in
s
tr
u
m
en
ts
.
T
h
e
f
ield
'
s
m
o
s
t
s
ig
n
if
ican
t
ac
c
o
m
p
lis
h
m
en
ts
h
av
e
b
ee
n
c
o
m
p
leted
an
d
p
o
ten
tial
m
eth
o
d
s
f
o
r
im
p
r
o
v
e
d
o
u
tc
o
m
es h
av
e
b
ee
n
h
ig
h
lig
h
te
d
.
T
h
e
p
u
r
p
o
s
e
o
f
u
s
in
g
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs
)
in
d
ee
p
lear
n
in
g
[
1
1
]
is
to
c
o
n
s
tr
u
ct
a
tr
u
s
two
r
th
y
s
y
s
tem
ca
p
ab
le
o
f
au
to
m
atica
lly
i
d
en
tify
in
g
a
n
d
ca
teg
o
r
is
in
g
h
u
m
an
e
m
o
tio
n
s
in
v
o
ice
a
n
d
f
ac
ial
ex
p
r
ess
io
n
s
[
1
2
]
.
T
h
is
in
ter
d
i
s
cip
lin
ar
y
ap
p
r
o
ac
h
in
co
r
p
o
r
a
tes
au
d
ito
r
y
an
d
v
is
u
al
cu
es
to
p
r
o
v
i
d
e
a
m
o
r
e
co
m
p
lete
p
ictu
r
e
o
f
p
eo
p
le'
s
e
m
o
tio
n
s
.
C
NNs
ar
e
u
s
ed
b
y
th
e
m
o
d
el
to
ef
f
ec
tiv
ely
e
v
alu
at
e
f
ac
ial
ex
p
r
ess
io
n
s
[
1
3
]
.
C
NNs
ar
e
v
er
y
ef
f
ec
tiv
e
at
p
r
o
ce
s
s
in
g
s
p
atial
d
ata.
E
m
o
tio
n
al
co
m
p
o
n
e
n
ts
in
th
e
a
u
d
ito
r
y
d
o
m
ai
n
m
ay
b
e
ex
tr
ac
ted
s
im
u
ltan
eo
u
s
ly
u
s
in
g
s
p
ee
ch
d
ata.
T
h
e
p
r
im
ar
y
g
o
als
ar
e
to
m
ak
e
th
e
m
o
d
el
m
o
r
e
g
en
er
aliza
b
le
ac
r
o
s
s
d
if
f
er
e
n
t
d
atasets
to
m
ak
e
em
o
tio
n
id
en
tific
atio
n
m
o
r
e
ac
cu
r
ate
a
n
d
t
o
la
y
t
h
e
s
tag
e
f
o
r
f
u
t
u
r
e
ad
v
an
ce
m
e
n
ts
in
ar
ea
s
s
u
ch
as
s
en
tim
en
t
an
aly
s
is
,
m
en
ta
l
h
ea
lth
m
o
n
ito
r
in
g
an
d
s
m
a
r
t
m
ac
h
in
e
-
h
u
m
a
n
in
ter
ac
tio
n
s
[
1
4
]
.
T
h
e
liter
atu
r
e
s
u
r
v
ey
is
s
h
o
wn
in
T
ab
le
1
.
T
ab
le
1
.
L
iter
atu
r
e
s
u
r
v
e
y
Y
e
a
r
A
u
t
h
o
r
Ti
t
l
e
M
e
t
h
o
d
A
b
st
r
a
c
t
2
0
1
4
K
u
d
i
r
i
e
t
a
l
.
[
1
5
]
Emo
t
i
o
n
d
e
t
e
c
t
i
o
n
mac
h
i
n
e
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
.
H
u
ma
n
c
o
m
p
u
t
e
r
i
n
t
e
r
a
c
t
i
o
n
,
r
e
l
a
t
i
v
e
b
i
n
f
r
e
q
u
e
n
c
y
c
o
e
f
f
i
c
i
e
n
t
s (
R
B
F
C
)
,
r
e
l
a
t
i
v
e
su
b
-
1
i
ma
g
e
f
e
a
t
u
r
e
s (RSF
)
,
su
p
p
o
r
t
v
e
c
t
o
r
m
a
c
h
i
n
e
(
S
V
M
)
.
Est
i
m
a
t
i
o
n
o
f
h
u
ma
n
e
mo
t
i
o
n
s
u
si
n
g
EC
G
i
ma
g
e
s
a
n
d
EEG
S
i
g
n
a
l
s.
I
d
e
n
t
i
f
y
i
n
g
t
h
e
p
o
s
i
t
i
v
e
a
n
d
n
e
g
a
t
i
v
e
e
mo
t
i
o
n
s.
2
0
1
7
R
u
i
z
e
t
a
l
.
[
1
6
]
H
u
ma
n
e
mo
t
i
o
n
d
e
t
e
c
t
i
o
n
t
h
r
o
u
g
h
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
.
S
V
M
,
R
B
F
C
.
F
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
a
r
e
t
w
o
t
y
p
e
s
n
a
m
e
l
y
,
D
e
l
i
b
e
r
a
t
e
a
n
d
n
o
n
-
d
e
l
i
b
e
r
a
t
e
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
.
2
0
1
9
K
h
a
l
i
l
e
t
a
l
.
[
1
7
]
Emo
t
i
o
n
d
e
t
e
c
t
i
o
n
t
h
r
o
u
g
h
s
p
e
e
c
h
a
n
d
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
.
S
ER
,
d
e
e
p
l
e
a
r
n
i
n
g
,
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
,
d
e
e
p
B
o
l
t
z
ma
n
n
ma
c
h
i
n
e
,
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
,
C
N
N
.
Th
e
r
e
v
i
e
w
c
o
v
e
r
s
d
a
t
a
b
a
ses
u
se
d
,
e
mo
t
i
o
n
s e
x
t
r
a
c
t
e
d
a
n
d
c
o
n
t
r
i
b
u
t
i
o
n
s
mo
d
e
.
2
0
2
0
B
e
g
a
j
e
t
a
l
.
[
1
8
]
Emo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
a
s
e
d
o
n
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
s
u
si
n
g
C
N
N
.
C
N
N
,
f
a
c
i
a
l
e
x
p
r
e
ss
i
o
n
r
e
c
o
g
n
i
t
i
o
n
,
f
a
c
i
a
l
e
m
o
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
,
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
.
O
v
e
r
t
h
e
l
a
st
f
e
w
y
e
a
r
s,
t
h
e
r
e
h
a
v
e
b
e
e
n
a
n
i
n
c
r
e
a
si
n
g
n
u
m
b
e
r
o
f
s
t
u
d
i
e
s
i
n
e
mo
t
i
o
n
r
e
c
o
g
n
i
t
i
o
n
b
e
c
a
u
s
e
o
f
t
h
e
i
mp
o
r
t
a
n
c
e
a
n
d
t
h
e
i
m
p
a
c
t
t
h
a
t
i
t
h
a
s
i
n
t
h
e
i
n
t
e
r
a
c
t
i
o
n
o
f
h
u
m
a
n
s wi
t
h
c
o
m
p
u
t
e
r
s
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
2
.
1
.
T
he
2
0
1
4
e
m
o
t
io
n det
ec
t
io
n t
hro
ug
h sp
ee
ch
a
nd
f
a
cia
l e
x
press
io
ns
T
h
er
e
h
av
e
b
ee
n
n
u
m
er
o
u
s
s
tu
d
ies
d
o
n
e
r
eg
a
r
d
in
g
p
r
ed
ict
in
g
em
o
tio
n
s
th
r
o
u
g
h
e
m
o
tiv
e
s
p
ee
ch
.
E
m
o
tio
n
d
etec
tio
n
v
ia
s
p
ee
ch
is
g
r
o
win
g
in
p
o
p
u
lar
ity
s
in
ce
s
p
ee
ch
-
b
ased
in
p
u
t
s
en
s
o
r
s
ar
e
m
u
ch
ch
ea
p
er
th
an
o
th
er
tech
n
o
lo
g
ies.
Sp
e
ec
h
m
ay
b
e
r
ec
o
g
n
ized
in
t
h
e
lo
u
d
est
s
itu
atio
n
s
u
s
in
g
th
e
h
u
m
an
au
d
ito
r
y
f
r
eq
u
e
n
cy
r
a
n
g
e
o
f
2
0
Hz
to
2
0
,
0
0
0
Hz.
W
h
en
m
an
ag
in
g
with
em
o
tio
n
in
n
o
is
y
s
itu
atio
n
s
th
r
o
u
g
h
s
p
ee
c
h
,
p
itch
v
ar
iatio
n
s
a
n
d
tem
p
o
o
f
s
p
ea
k
in
g
i
n
r
eg
a
r
d
t
o
tim
e
b
ec
o
m
e
s
ig
n
if
ican
t.
Nev
er
th
ele
s
s
,
th
e
p
itch
o
f
th
e
s
p
ee
ch
s
ig
n
al
ca
n
s
h
if
t
d
u
e
to
a
n
u
m
b
er
o
f
en
v
ir
o
n
m
e
n
ts
.
Ad
d
itio
n
ally
,
th
e
in
cr
ea
s
in
g
d
im
en
s
io
n
ality
o
f
th
e
v
o
ice
s
ig
n
als in
cr
ea
s
es th
e
tim
e
co
m
p
lex
ity
.
2
.
2
.
T
he
2
0
1
7
s
t
a
t
e
o
f
hu
ma
n e
m
o
t
io
n r
ec
o
g
nitio
n v
ia
v
e
rba
l a
nd
f
a
cia
l e
x
press
io
ns
I
t
h
as
p
r
o
v
en
d
if
f
icu
lt
to
u
s
e
co
m
p
u
ter
s
to
esti
m
ate
h
u
m
an
e
m
o
tio
n
s
ev
er
s
in
ce
p
eo
p
le
b
e
g
an
tak
in
g
p
ar
t
in
v
er
b
al
s
ec
ess
io
n
s
.
Usi
n
g
a
p
r
o
p
o
s
ed
h
y
b
r
id
m
eth
o
d
th
at
in
co
r
p
o
r
ates
b
o
th
s
p
o
k
en
a
n
d
f
ac
ial
em
o
tio
n
s
,
th
is
s
tu
d
y
esti
m
ates
th
e
ty
p
ic
al
em
o
tio
n
al
s
tates
o
f
a
s
u
b
je
ct
d
u
r
in
g
a
c
o
n
v
e
r
s
atio
n
al
s
ep
ar
atio
n
,
i
n
clu
d
in
g
an
g
er
,
s
ad
n
ess
,
s
atis
f
ac
tio
n
,
b
o
r
ed
o
m
,
d
is
g
u
s
t
,
an
d
asto
n
is
h
m
en
t.
Fo
r
au
d
i
o
an
d
v
id
eo
d
a
ta
in
p
ar
ticu
lar
,
we
u
s
e
r
elativ
e
b
in
f
r
eq
u
en
cy
c
o
ef
f
icien
ts
an
d
f
am
ily
m
em
b
er
s
u
b
-
im
ag
e
-
b
ased
f
ea
tu
r
es.
W
h
en
it
co
m
es
to
class
if
icatio
n
,
SVM
with
r
ad
ial
b
asis
b
it
i
s
th
e
way
t
o
g
o
.
Acc
o
r
d
in
g
to
th
is
s
tu
d
y
'
s
f
in
d
in
g
s
,
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
24
:
1
4
4
8
-
1
4
5
9
1450
r
ec
o
m
m
en
d
ed
b
len
d
ap
p
r
o
ac
h
an
d
f
ea
tu
r
e
r
em
o
v
al
with
v
o
ice
an
d
f
ac
e
ar
e
two
o
f
th
e
m
o
s
t
well
-
k
n
o
wn
f
ac
to
r
s
im
p
ac
tin
g
th
e
e
m
o
tio
n
d
etec
tio
n
s
y
s
tem
.
T
h
er
e
ar
e
f
ew
f
ac
to
r
s
th
at
m
ig
h
t
in
f
l
u
en
ce
th
e
em
o
tio
n
d
etec
tin
g
s
y
s
tem
,
alth
o
u
g
h
th
eir
im
p
ac
t
is
m
i
n
im
al.
T
h
r
o
u
g
h
th
e
u
s
e
o
f
p
u
r
p
o
s
ef
u
l
f
ac
ial
ex
p
r
ess
io
n
s
,
it
was
s
h
o
wn
th
at
th
e
b
im
o
d
al
em
o
tio
n
d
etec
tio
n
s
y
s
tem
o
u
t
p
e
r
f
o
r
m
e
d
th
e
u
n
im
o
d
al
m
eth
o
d
.
An
a
p
p
r
o
p
r
iate
d
atab
ase
is
u
s
ed
to
ad
d
r
ess
th
e
p
r
o
b
lem
.
T
h
e
r
esu
lts
in
d
icate
d
th
at
co
m
p
a
r
ed
to
t
h
e
o
th
er
s
y
s
tem
s
,
th
e
r
ec
o
m
m
en
d
ed
em
o
tio
n
r
ec
o
g
n
itio
n
m
eth
o
d
p
er
f
o
r
m
ed
m
u
c
h
b
etter
wh
en
it
ca
m
e
to
f
u
n
d
am
en
tal
em
o
tio
n
al
co
u
r
s
es.
2
.
3
.
E
mo
t
io
n r
ec
o
g
nitio
n in sp
ee
ch
wit
h t
he
u
s
e
o
f
deep
l
ea
rning
t
ec
hn
iqu
e
s
in 2
0
1
9
A
v
ital
y
et
ch
allen
g
in
g
asp
ec
t
o
f
h
u
m
an
-
co
m
p
u
ter
in
ter
a
ctio
n
(
HC
I
)
is
th
e
ab
ilit
y
to
p
er
ce
iv
e
r
ec
o
g
n
itio
n
f
r
o
m
v
o
ice
s
ig
n
als.
Sev
er
al
m
eth
o
d
s
,
in
clu
d
in
g
s
ev
er
al
well
-
estab
lis
h
ed
s
p
ee
ch
ass
ess
m
en
t
an
d
class
if
icatio
n
alg
o
r
ith
m
s
,
h
av
e
b
ee
n
u
s
ed
in
th
e
s
p
ee
c
h
em
o
tio
n
r
ec
o
g
n
itio
n
(
SER
)
liter
atu
r
e
to
d
e
-
em
o
tio
n
alize
s
ig
n
als.
A
n
ew
alter
n
ativ
e
to
tr
ad
itio
n
al
p
r
o
c
ed
u
r
es
in
SER
b
ased
o
n
d
ee
p
lear
n
in
g
tech
n
i
q
u
es
h
as
ju
s
t
b
ee
n
p
r
o
p
o
s
ed
.
T
h
is
ar
ticle
p
r
o
v
id
es
an
o
v
e
r
v
ie
w
o
f
d
ee
p
d
is
co
v
er
in
g
tech
n
iq
u
es
an
d
d
is
cu
s
s
es
r
ec
en
t
r
esear
ch
th
at
h
as
u
s
ed
th
ese
m
eth
o
d
s
to
r
ec
o
g
n
ize
em
o
tio
n
s
in
s
p
o
k
en
lan
g
u
a
g
e.
T
h
e
ass
es
s
m
en
t
d
elv
es
in
to
th
e
d
ata
s
o
u
r
ce
s
u
s
ed
,
em
o
tio
n
s
elim
in
ated
,
f
u
n
d
s
allo
ca
ted
to
SER
a
n
d
an
y
ass
o
ciate
d
r
estrictio
n
s
.
2
.
4
.
I
n 2
0
2
0
e
m
o
t
i
o
n r
ec
o
g
n
it
io
n us
ing
co
nv
o
lutio
na
l neura
l net
wo
rk
s
Facial
em
o
tio
n
r
ec
o
g
n
itio
n
h
a
s
b
ee
n
th
e
s
u
b
ject
o
f
a
n
in
c
r
e
asin
g
s
am
p
le
o
f
s
tu
d
ies
in
r
ec
en
t
y
ea
r
s
d
u
e
t
o
its
s
ig
n
if
ican
ce
an
d
i
n
f
lu
en
ce
o
n
HC
I
.
T
h
e
u
s
e
o
f
d
e
ep
lear
n
i
n
g
m
eth
o
d
s
is
b
ec
o
m
in
g
ess
en
tial
d
u
e
to
th
e
r
is
e
in
th
e
n
u
m
b
er
o
f
ch
allen
g
in
g
d
atasets
.
T
h
is
s
tu
d
y
in
v
esti
g
ates
th
e
ch
allen
g
es
o
f
em
o
tio
n
ac
k
n
o
wled
g
m
en
t
d
atasets
an
d
ex
p
er
im
en
ts
with
v
ar
io
u
s
cr
iter
ia
an
d
d
esig
n
s
o
f
C
NN
s
to
id
en
tify
s
ev
en
h
u
m
an
em
o
tio
n
s
:
an
g
s
t,
f
ea
r
,
co
n
tem
p
t,
m
o
ck
er
y
,
jo
y
,
d
e
s
p
air
,
an
d
s
h
o
c
k
in
f
ac
ial
ex
p
r
ess
io
n
s
.
T
h
e
k
ey
d
ataset
f
o
r
o
u
r
wo
r
k
is
m
u
lti
-
e
m
o
tio
n
f
ac
ial
e
x
p
r
ess
io
n
d
atas
et
(
iC
V
ME
FED)
wh
ich
is
p
r
etty
n
ew,
in
tr
ig
u
in
g
an
d
v
er
y
d
if
f
icu
lt.
2.
5
.
T
he
2
0
1
4
e
m
o
t
io
n det
ec
t
io
n t
hro
ug
h sp
ee
ch
a
nd
f
a
cia
l e
x
press
io
ns
T
h
er
e
h
a
v
e
b
ee
n
n
u
m
e
r
o
u
s
s
tu
d
ies
d
o
n
e
r
eg
a
r
d
in
g
p
r
e
d
ictin
g
em
o
tio
n
s
t
h
r
o
u
g
h
em
o
tiv
e
s
p
ee
ch
[
1
5
]
.
E
m
o
tio
n
d
etec
tio
n
v
ia
s
p
ee
ch
is
g
r
o
win
g
in
p
o
p
u
lar
ity
s
in
ce
s
p
ee
ch
-
b
ased
in
p
u
t
s
en
s
o
r
s
ar
e
m
u
ch
ch
ea
p
er
th
an
o
th
er
tech
n
o
lo
g
ies.
Sp
e
ec
h
m
ay
b
e
r
ec
o
g
n
ized
in
t
h
e
lo
u
d
est
s
itu
atio
n
s
u
s
in
g
th
e
h
u
m
an
au
d
ito
r
y
f
r
eq
u
e
n
cy
r
a
n
g
e
o
f
2
0
Hz
to
2
0
,
0
0
0
Hz.
W
h
en
m
an
ag
in
g
with
em
o
tio
n
in
n
o
is
y
s
itu
atio
n
s
th
r
o
u
g
h
s
p
ee
c
h
,
p
itch
v
ar
iatio
n
s
a
n
d
tem
p
o
o
f
s
p
ea
k
in
g
i
n
r
eg
a
r
d
t
o
tim
e
b
ec
o
m
e
s
ig
n
if
ican
t.
Nev
er
th
ele
s
s
,
th
e
p
itch
o
f
th
e
s
p
ee
ch
s
ig
n
al
ca
n
s
h
if
t
d
u
e
to
a
n
u
m
b
er
o
f
en
v
ir
o
n
m
e
n
ts
.
Ad
d
itio
n
ally
,
th
e
in
cr
ea
s
in
g
d
im
en
s
io
n
ality
o
f
th
e
v
o
ice
s
ig
n
als in
cr
ea
s
es th
e
tim
e
co
m
p
lex
ity
.
2.
6
.
T
he
2
0
1
7
s
t
a
t
e
o
f
hu
ma
n e
m
o
t
io
n r
ec
o
g
nitio
n v
ia
v
e
rba
l a
nd
f
a
cia
l e
x
press
io
ns
I
t
h
as
p
r
o
v
en
d
if
f
icu
lt
to
u
s
e
co
m
p
u
ter
s
to
esti
m
ate
h
u
m
an
e
m
o
tio
n
s
ev
er
s
in
ce
p
eo
p
le
b
e
g
an
tak
in
g
p
ar
t
in
v
er
b
al
s
ec
ess
io
n
s
.
Usi
n
g
a
p
r
o
p
o
s
ed
h
y
b
r
id
m
eth
o
d
th
at
in
co
r
p
o
r
ates
b
o
th
s
p
o
k
en
a
n
d
f
ac
ial
em
o
tio
n
s
,
th
is
s
tu
d
y
esti
m
ates
th
e
ty
p
ic
al
em
o
tio
n
al
s
tates
o
f
a
s
u
b
je
ct
d
u
r
in
g
a
c
o
n
v
e
r
s
atio
n
al
s
ep
ar
atio
n
,
i
n
clu
d
in
g
an
g
er
,
s
ad
n
ess
,
s
atis
f
ac
tio
n
,
b
o
r
ed
o
m
,
d
is
g
u
s
t
an
d
asto
n
is
h
m
en
t.
Fo
r
au
d
i
o
a
n
d
v
id
eo
d
a
ta
in
p
a
r
ticu
lar
,
w
e
u
s
e
r
elativ
e
b
in
f
r
eq
u
en
cy
c
o
ef
f
icien
ts
an
d
f
am
ily
m
em
b
er
s
u
b
-
im
ag
e
-
b
ased
f
ea
tu
r
es.
W
h
en
it
co
m
es
to
class
if
icatio
n
,
SVM
w
ith
r
ad
ial
b
asis
b
it
i
s
th
e
way
to
g
o
.
Acc
o
r
d
in
g
to
th
is
s
tu
d
y
'
s
f
in
d
in
g
s
[
1
6
]
,
th
e
r
ec
o
m
m
en
d
ed
b
len
d
ap
p
r
o
ac
h
an
d
f
ea
tu
r
e
r
em
o
v
al
with
v
o
ice
an
d
f
ac
e
ar
e
two
o
f
th
e
m
o
s
t
well
-
k
n
o
wn
f
ac
to
r
s
im
p
ac
tin
g
th
e
e
m
o
tio
n
d
etec
tio
n
s
y
s
tem
.
T
h
er
e
ar
e
f
ew
f
ac
to
r
s
th
at
m
ig
h
t
in
f
l
u
en
ce
th
e
em
o
tio
n
d
etec
tin
g
s
y
s
tem
,
alth
o
u
g
h
th
eir
im
p
ac
t
is
m
i
n
im
al.
T
h
r
o
u
g
h
th
e
u
s
e
o
f
p
u
r
p
o
s
ef
u
l
f
ac
ial
ex
p
r
ess
io
n
s
,
it
was
s
h
o
wn
th
at
th
e
b
im
o
d
al
em
o
tio
n
d
etec
tio
n
s
y
s
tem
o
u
t
p
e
r
f
o
r
m
e
d
th
e
u
n
im
o
d
al
m
eth
o
d
.
An
a
p
p
r
o
p
r
iate
d
atab
ase
is
u
s
ed
to
ad
d
r
ess
th
e
p
r
o
b
lem
.
T
h
e
r
esu
lts
in
d
icate
d
th
at
co
m
p
a
r
ed
to
t
h
e
o
th
er
s
y
s
tem
s
,
th
e
r
ec
o
m
m
en
d
ed
em
o
tio
n
r
ec
o
g
n
itio
n
m
eth
o
d
p
er
f
o
r
m
ed
m
u
c
h
b
etter
wh
en
it
ca
m
e
to
f
u
n
d
am
en
tal
em
o
tio
n
al
co
u
r
s
es.
2.
7
.
E
mo
t
io
n r
ec
o
g
nitio
n in sp
ee
ch
wit
h t
he
u
s
e
o
f
deep
l
ea
rning
t
ec
hn
iqu
e
s
in 2
0
1
9
A
v
ital
y
et
ch
allen
g
in
g
asp
e
ct
o
f
HC
I
is
th
e
ab
ilit
y
to
p
er
ce
iv
e
r
ec
o
g
n
itio
n
f
r
o
m
v
o
ice
s
ig
n
als.
Sev
er
al
m
eth
o
d
s
,
in
clu
d
in
g
s
ev
er
al
well
-
estab
lis
h
ed
s
p
ee
c
h
ass
ess
m
en
t
an
d
class
if
icatio
n
alg
o
r
ith
m
s
,
h
av
e
b
ee
n
u
s
ed
in
th
e
SER liter
atu
r
e
to
d
e
-
em
o
tio
n
alize
s
ig
n
als.
A
n
ew
alter
n
ativ
e
to
tr
ad
itio
n
a
l p
r
o
ce
d
u
r
es in
SER
b
ased
o
n
d
ee
p
lear
n
i
n
g
tech
n
iq
u
es
h
as
ju
s
t
b
ee
n
p
r
o
p
o
s
ed
.
T
h
is
ar
ticle
p
r
o
v
id
es
an
o
v
er
v
iew
o
f
d
ee
p
d
is
co
v
er
in
g
tech
n
iq
u
es
an
d
d
i
s
cu
s
s
es
r
ec
en
t
r
esear
ch
th
at
h
as
u
s
ed
th
ese
m
eth
o
d
s
to
r
ec
o
g
n
is
e
em
o
tio
n
s
in
s
p
o
k
en
lan
g
u
ag
e
[
1
7
]
.
T
h
e
ass
ess
m
en
t d
elv
es in
to
th
e
d
ata
s
o
u
r
ce
s
u
s
ed
,
em
o
tio
n
s
elim
in
ated
,
f
u
n
d
s
allo
ca
ted
to
SER
an
d
an
y
ass
o
ciate
d
r
es
tr
ictio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Deep
-
S
F
E
R
:
d
ee
p
c
o
n
vo
l
u
tio
n
a
l n
eu
r
a
l
n
etw
o
r
k
a
n
d
MFCC
a
n
effec
tive
…
(
R
a
vi
Gu
mm
u
la
)
1451
2.
8
.
I
n 2
0
2
0
e
m
o
t
i
o
n r
ec
o
g
n
it
io
n us
ing
co
nv
o
lutio
na
l neura
l net
wo
rk
s
Facial
em
o
tio
n
r
ec
o
g
n
itio
n
h
a
s
b
ee
n
th
e
s
u
b
ject
o
f
a
n
in
c
r
e
asin
g
s
am
p
le
o
f
s
tu
d
ies
in
r
ec
en
t
y
ea
r
s
d
u
e
to
its
s
ig
n
if
ican
ce
an
d
in
f
lu
en
ce
o
n
HC
I
[
1
8
]
.
T
h
e
u
s
e
o
f
d
ee
p
lear
n
in
g
m
eth
o
d
s
is
b
ec
o
m
in
g
ess
en
tial
d
u
e
to
th
e
r
is
e
in
t
h
e
n
u
m
b
e
r
o
f
ch
allen
g
in
g
d
atasets
[
1
9
]
.
T
h
is
s
tu
d
y
in
v
esti
g
ates
th
e
ch
allen
g
es
o
f
em
o
tio
n
ac
k
n
o
wled
g
m
en
t
d
atasets
an
d
ex
p
er
im
en
ts
with
v
ar
io
u
s
cr
iter
ia
an
d
d
esig
n
s
o
f
C
NN
s
to
id
en
tify
s
ev
en
h
u
m
an
em
o
tio
n
s
:
an
g
s
t,
f
ea
r
,
co
n
tem
p
t,
m
o
ck
er
y
,
jo
y
,
d
esp
air
an
d
s
h
o
ck
i
n
f
ac
ial
e
x
p
r
es
s
io
n
s
[
2
0
]
.
T
h
e
k
ey
d
ataset
f
o
r
o
u
r
wo
r
k
is
m
u
lti
-
e
m
o
tio
n
f
ac
ial
e
x
p
r
ess
io
n
d
atas
et
(
iC
V
ME
FED)
wh
ich
is
p
r
etty
n
ew,
in
tr
ig
u
in
g
an
d
v
er
y
d
if
f
icu
lt.
3.
M
E
T
H
O
D
3
.
1
.
P
r
o
blem
s
t
a
t
e
m
ent
Giv
en
th
at
h
u
m
an
s
e
n
g
ag
e
i
n
co
n
v
er
s
atio
n
al
s
ess
io
n
s
,
it
h
as
b
ee
n
d
if
f
ic
u
lt
to
e
v
alu
ate
h
u
m
an
em
o
tio
n
s
u
s
in
g
a
co
m
p
u
ter
[
2
1
]
.
T
h
is
r
esear
ch
p
r
o
ject
p
r
o
p
o
s
es
a
h
y
b
r
id
s
y
s
tem
th
at
u
s
es
h
ar
s
h
f
ac
es
an
d
v
o
ice
to
m
im
ic
th
e
b
asic
co
n
c
ep
ts
(
an
g
r
y
,
s
ad
,
j
o
y
f
u
l,
b
o
r
ed
,
d
is
g
u
s
ted
,
an
d
s
h
o
c
k
ed
)
o
f
a
n
ar
s
o
n
is
t
d
u
r
in
g
a
co
n
v
er
s
atio
n
al
s
ec
ess
io
n
.
Fam
ily
m
em
b
e
r
in
r
elativ
e
s
u
b
-
im
ag
e
an
d
f
r
eq
u
e
n
cy
co
ef
f
i
cien
ts
ty
r
es
th
at
ar
e
b
ased
o
n
s
p
ec
ialis
ed
m
ater
i
als
ar
e
u
s
ed
f
o
r
v
is
u
al
an
d
au
d
ito
r
y
d
ata
[
2
2
]
.
Su
p
p
o
r
t
c
lass
if
icatio
n
is
ac
co
m
p
lis
h
ed
u
s
in
g
v
ec
to
r
eq
u
ip
m
en
t
with
a
r
ad
ial
b
asis
k
er
n
el.
B
ased
o
n
th
e
r
esu
lts
o
f
th
is
r
esear
ch
,
th
e
r
ec
o
m
m
en
d
ed
f
u
n
ctio
n
e
x
tr
ac
tio
n
u
s
in
g
v
o
ice
an
d
f
ac
e
is
a
m
ajo
r
f
ac
to
r
in
f
lu
e
n
cin
g
th
e
em
o
tio
n
d
is
co
v
er
y
s
y
s
tem
,
esp
ec
ially
wh
en
co
m
b
in
ed
with
th
e
p
r
o
p
o
s
ed
a
p
p
r
o
ac
h
.
T
h
er
e
ar
e
a
f
ew
f
ac
t
o
r
s
th
at
m
ig
h
t
in
f
l
u
en
ce
th
e
em
o
tio
n
d
etec
tin
g
s
y
s
tem
,
b
u
t
th
ey
wo
n
'
t
h
av
e
m
u
c
h
o
f
an
ef
f
ec
t.
C
o
m
p
ar
e
d
to
a
u
n
i
m
o
d
al
s
y
s
tem
th
at
u
s
es
in
ten
tio
n
al
f
ac
es
t
o
d
eter
m
in
e
em
o
tio
n
s
,
th
e
b
im
o
d
al
ap
p
r
o
ac
h
is
less
ef
f
ec
tiv
e
[
2
3
]
.
A
r
elev
an
t
d
ata
s
o
u
r
ce
is
u
s
ed
to
tack
le
th
e
p
r
o
b
lem
.
Acc
o
r
d
i
n
g
to
t
h
e
r
esu
lts
,
th
e
s
u
g
g
ested
em
o
tio
n
d
is
co
v
er
y
m
eth
o
d
o
u
tp
er
f
o
r
m
ed
th
e
co
m
p
etitio
n
wh
en
it c
am
e
to
b
asic p
s
y
ch
o
l
o
g
ical
ca
teg
o
r
ies.
3
.
2
.
P
r
o
po
s
ed
s
y
s
t
em
T
h
e
im
p
o
r
ta
n
ce
an
d
im
p
ac
t
o
f
f
ac
ial
em
o
tio
n
r
ec
o
g
n
itio
n
i
n
HC
I
h
as
led
to
an
u
p
s
u
r
g
e
i
n
r
esear
ch
in
to
th
e
to
p
ic
in
r
ec
en
t
y
ea
r
s
.
I
t
is
n
o
w
n
ec
ess
ar
y
to
u
s
e
d
ee
p
lear
n
in
g
al
g
o
r
ith
m
s
d
u
e
to
th
e
in
cr
ea
s
in
g
d
iv
er
s
ity
o
f
d
e
m
an
d
in
g
d
at
asets
[
2
4
]
.
I
n
th
is
p
a
p
er
,
we
tak
e
a
lo
o
k
at
th
e
p
r
o
b
lem
s
with
f
ee
lin
g
ac
k
n
o
wled
g
m
en
t
d
atasets
an
d
tr
y
v
ar
io
u
s
cr
iter
ia
an
d
C
NN
ar
ch
itectu
r
es
to
d
etec
t
s
ev
en
h
u
m
an
e
m
o
tio
n
s
:
an
g
s
t,
co
n
ce
r
n
,
co
n
tem
p
t,
jo
y
,
s
ad
n
ess
,
an
d
s
u
r
p
r
is
e
in
f
ac
ial
ex
p
r
ess
io
n
s
[
2
5
]
.
B
ec
au
s
e
it
is
n
o
v
el,
in
ter
esti
n
g
an
d
ch
allen
g
in
g
,
we
h
av
e
c
h
o
s
en
iC
V
ME
FED
as th
e
p
r
im
ar
y
d
ataset
f
o
r
o
u
r
r
esear
ch
.
3
.
3
.
C
o
nv
o
lutio
na
l neura
l net
wo
rk
T
h
e
C
NN
is
co
m
m
o
n
ly
u
s
ed
in
d
ee
p
lear
n
in
g
f
o
r
im
a
g
e
r
e
co
g
n
itio
n
an
d
co
m
p
u
ter
v
is
io
n
task
s
.
I
t's
g
r
ea
t a
t c
ap
tu
r
i
n
g
s
p
atial
r
elatio
n
s
h
ip
s
in
d
ata.
Fig
u
r
e
1
e
x
p
la
in
s
th
e
C
NN
p
r
o
ce
s
s
.
Fig
u
r
e
1
.
C
NN
p
r
o
ce
s
s
3
.
3
.
1
.
H
er
e
a
re
t
he
k
e
y
a
s
pect
s
o
f
CNNs
C
NNs
ar
e
d
esig
n
ed
with
a
n
u
m
b
er
o
f
lay
er
s
s
u
ch
as
f
u
ll
y
co
n
n
ec
te
d
,
p
o
o
lin
g
an
d
co
n
v
o
lu
tio
n
al
o
n
es.
C
o
n
v
o
lu
tio
n
al
lay
er
s
ca
p
tu
r
e
f
ea
tu
r
e
h
ier
ar
c
h
ies
b
y
a
p
p
ly
in
g
f
ilter
s
(
k
er
n
els)
t
o
in
c
o
m
in
g
d
ata.
C
NNs
ar
e
co
m
m
o
n
ly
u
s
ed
f
o
r
im
a
g
e
an
d
v
id
e
o
r
ec
o
g
n
itio
n
,
m
ed
ical
im
ag
e
an
aly
s
is
,
o
b
j
ec
t
d
etec
tio
n
,
f
ac
ial
r
ec
o
g
n
itio
n
,
an
d
o
th
er
p
u
r
p
o
s
es.
T
h
ey
ar
e
also
u
s
ed
in
n
o
n
-
v
is
u
al
ac
tiv
ities
wh
ich
in
clu
d
es
n
atu
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
an
d
tim
e
-
s
er
ies
an
aly
s
is
.
T
h
ese
attr
ib
u
tes
co
n
tr
i
b
u
te
to
C
NNs
ex
ce
p
tio
n
al
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
r
ep
r
esen
tatio
n
s
k
ills
,
m
ak
in
g
th
em
a
co
r
n
er
s
to
n
e
o
f
m
o
d
er
n
c
o
m
p
u
ter
v
is
io
n
a
n
d
r
elate
d
f
ield
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
24
:
1
4
4
8
-
1
4
5
9
1452
B
ac
k
p
r
o
p
ag
atio
n
is
a
m
ec
h
an
is
m
f
o
r
u
p
d
atin
g
th
e
weig
h
ts
in
a
n
etwo
r
k
b
ased
o
n
an
er
r
o
r
.
I
t
en
tails
co
m
p
u
tin
g
th
e
g
r
ad
ien
t o
f
th
e
lo
s
s
f
u
n
ctio
n
in
r
elatio
n
to
ea
c
h
weig
h
t a
n
d
ch
an
g
in
g
t
h
e
weig
h
ts
ac
co
r
d
in
g
ly
.
o
p
tim
izatio
n
alg
o
r
ith
m
s
:
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
S
GD)
,
Ad
am
,
an
d
R
MSp
r
o
p
ar
e
th
r
ee
p
o
p
u
lar
o
p
tim
izer
s
,
ea
ch
with
its
o
wn
s
tr
ateg
y
f
o
r
m
o
d
if
y
in
g
lear
n
in
g
r
ate
an
d
o
th
er
p
ar
am
eter
s
.
3
.
3
.
2
.
L
a
y
er
s
t
ha
t
us
e
co
nv
o
l
utio
n
I
n
c
o
n
v
o
l
u
tio
n
,
f
ea
tu
r
es
ar
e
ex
tr
ac
ted
b
y
s
lid
in
g
a
f
ilter
ac
r
o
s
s
th
e
in
p
u
t
d
ata,
wh
ile
s
p
atial
co
r
r
elatio
n
s
ar
e
p
r
eser
v
ed
.
B
y
lear
n
in
g
h
ier
ar
c
h
ical
r
ep
r
esen
tatio
n
s
,
m
an
y
f
ilter
s
ca
n
id
en
tify
a
wid
e
r
an
g
e
o
f
ch
ar
ac
ter
is
tics
.
B
y
lo
wer
in
g
th
e
n
u
m
b
er
o
f
s
p
atial
d
im
e
n
s
io
n
s
an
d
th
e
c
o
m
p
u
tatio
n
al
ef
f
o
r
t,
p
o
o
lin
g
lay
e
r
s
lo
wer
s
am
p
le
f
ea
tu
r
e
m
ap
s
.
Ma
x
p
o
o
lin
g
(
c
h
o
o
s
in
g
th
e
m
ax
im
u
m
v
alu
e
)
an
d
av
er
a
g
e
p
o
o
lin
g
ar
e
two
co
m
m
o
n
way
s
to
p
o
o
l d
ata.
3
.
3
.
3
.
L
a
y
er
s
t
ha
t
a
re
f
ull
y
c
o
nn
ec
t
ed
I
n
o
r
d
er
to
ag
g
r
eg
ate
lear
n
t
f
ea
tu
r
es
f
o
r
u
s
e
in
r
eg
r
ess
io
n
o
r
class
if
icatio
n
task
s
,
f
u
lly
co
n
n
ec
ted
lay
er
s
ar
e
u
s
ed
.
E
ac
h
n
eu
r
o
n
in
o
n
e
lay
er
is
co
n
n
ec
ted
to
e
v
er
y
n
e
u
r
o
n
in
th
e
lay
e
r
b
elo
w
an
d
ab
o
v
e
it
v
ia
th
ese
lay
er
s
.
Fu
lly
co
n
n
ec
ted
lay
er
s
,
also
k
n
o
wn
as
d
en
s
e
lay
er
s
,
p
e
r
f
o
r
m
a
s
ig
n
if
ica
n
t
f
u
n
ctio
n
in
n
eu
r
al
n
etwo
r
k
s
,
in
p
a
r
ticu
lar
f
o
r
r
e
g
r
ess
io
n
an
d
class
if
icatio
n
.
T
h
ese
lay
er
s
ar
e
r
ef
er
r
e
d
to
b
e
"f
u
lly
co
n
n
ec
ted
"
b
e
c
a
u
s
e
e
a
c
h
n
e
u
r
o
n
i
n
o
n
e
l
a
y
e
r
c
o
m
m
u
n
i
c
a
t
e
s
w
it
h
e
v
e
r
y
n
e
u
r
o
n
i
n
t
h
e
l
a
y
e
r
s
e
x
cl
u
s
i
v
el
y
b
e
l
o
w
a
n
d
a
b
o
v
e
i
t.
T
h
is
d
esig
n
en
ab
les
th
e
n
etwo
r
k
to
ca
p
tu
r
e
an
d
ag
g
r
eg
ate
co
m
p
licated
in
ter
ac
tio
n
s
b
etwe
en
lear
n
ed
f
ea
tu
r
es.
T
h
ey
id
en
tify
p
atter
n
s
o
r
s
eq
u
en
ce
s
in
th
e
d
ata,
s
u
ch
as
ed
g
es
in
an
im
ag
e
o
r
tem
p
o
r
al
i
n
ter
ac
tio
n
s
in
tim
e
s
er
ies.
Fo
llo
win
g
th
e
f
ea
tu
r
e
ex
tr
ac
tio
n
p
r
o
ce
d
u
r
e,
f
u
lly
co
n
n
ec
ted
lay
er
s
ar
e
u
s
ed
to
ag
g
r
eg
ate
all
o
f
th
e
k
n
o
wled
g
e
g
ai
n
ed
an
d
g
e
n
er
at
e
th
e
f
in
al
f
o
r
ec
ast.
T
h
is
in
f
o
r
m
atio
n
a
g
g
r
eg
atio
n
m
ak
es
f
u
ll
y
co
n
n
ec
ted
lay
er
s
n
ec
ess
ar
y
f
o
r
p
r
o
g
r
am
s
s
u
ch
as
r
eg
r
ess
io
n
an
d
class
if
icat
io
n
.
I
n
ea
ch
f
u
lly
co
n
n
ec
ted
lay
er
,
ea
ch
n
eu
r
o
n
co
m
p
u
tes
a
weig
h
ted
s
u
m
o
f
t
h
e
p
r
ev
io
u
s
lay
er
'
s
o
u
tp
u
ts
,
ad
d
s
a
b
ias
ter
m
,
an
d
s
en
d
s
th
e
r
esu
lt
em
p
lo
y
in
g
a
n
ac
tiv
atio
n
f
u
n
ctio
n
.
Fu
lly
co
n
n
ec
ted
lay
er
s
ar
e
ad
d
itio
n
al
ly
h
elp
f
u
l
in
d
im
en
s
io
n
ality
r
ed
u
ctio
n
.
As
d
ata
p
ass
es th
r
o
u
g
h
th
ese
lay
er
s
,
th
e
h
ig
h
-
d
im
e
n
s
io
n
al
f
ea
tu
r
e
s
p
ac
e
u
s
u
ally
b
ec
o
m
es f
latten
ed
,
m
ak
in
g
it e
asier
to
id
en
tify
o
r
d
eter
m
in
e
o
u
tco
m
es.
Fo
r
class
if
icatio
n
task
s
,
th
e
f
in
al
f
u
lly
co
n
n
ec
ted
lay
er
is
f
r
eq
u
e
n
tly
co
n
s
tr
u
cted
u
p
o
f
as
m
a
n
y
n
e
u
r
o
n
s
as
class
es,
with
a
s
o
f
tm
ax
ac
tiv
atio
n
f
u
n
ctio
n
p
r
o
d
u
ci
n
g
p
r
o
b
ab
ilit
ies
f
o
r
ea
ch
.
3
.
4
.
M
el
f
re
qu
ency
ce
ps
t
ra
l c
o
ef
f
icient
Mel
-
f
r
eq
u
e
n
cy
ce
p
s
tr
al
co
ef
f
i
cien
t
(
MFC
C
)
i
s
a
wid
ely
u
s
e
d
tech
n
iq
u
e
f
o
r
ex
tr
ac
tin
g
au
d
io
f
ea
tu
r
es
an
d
th
e
f
l
o
w
ch
ar
t is sh
o
wn
in
th
e
Fig
u
r
e
2
.
T
h
e
p
r
im
ar
y
g
o
a
ls
ar
e:
−
E
lim
in
ate
p
itch
in
f
o
r
m
atio
n
w
h
ich
is
v
o
ca
l f
o
ld
ex
citatio
n
.
−
Sep
ar
ate
th
e
r
etr
iev
e
d
ch
ar
ac
t
er
is
tics
f
r
o
m
ea
ch
o
th
e
r
.
−
Mo
d
if
y
to
a
cc
o
m
m
o
d
ate
th
e
w
ay
p
eo
p
le
h
ea
r
d
if
f
er
e
n
t v
o
lu
m
es a
n
d
f
r
e
q
u
en
cies o
f
s
o
u
n
d
.
−
R
ec
o
r
d
h
o
w
co
n
tex
t is ch
an
g
in
g
.
Featu
r
e
is
th
e
ac
o
u
s
tic
p
r
o
p
er
ty
o
f
th
e
v
o
ice
s
ig
n
al.
T
h
e
ac
o
u
s
tic
f
ea
tu
r
es
o
f
t
h
e
s
p
ee
ch
s
ig
n
al
ar
e
p
r
eser
v
ed
th
r
o
u
g
h
o
u
t a
n
aly
s
is
b
y
ex
tr
ac
t
in
g
ju
s
t a
lim
ited
q
u
a
n
tity
o
f
d
ata
f
r
o
m
th
e
s
ig
n
al
[
2
6
]
.
Fig
u
r
e
2
.
MFC
C
f
lo
w
ch
ar
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Deep
-
S
F
E
R
:
d
ee
p
c
o
n
vo
l
u
tio
n
a
l n
eu
r
a
l
n
etw
o
r
k
a
n
d
MFCC
a
n
effec
tive
…
(
R
a
vi
Gu
mm
u
la
)
1453
3
.
5
.
F
a
ce
e
m
o
t
io
n us
ing
CNN
T
h
e
f
ac
e
em
o
tio
n
u
s
in
g
C
NN
p
r
o
ce
s
s
s
tep
s
ar
e
ex
p
lain
ed
i
n
Fig
u
r
e
3
.
(
a)
First
th
in
g
to
d
o
(
Step
1
)
g
ath
er
in
f
o
r
m
atio
n
:
o
b
tain
a
c
o
llectio
n
o
f
im
ag
es
o
f
f
ac
es
lab
elled
with
m
atch
in
g
em
o
tio
n
s
(
s
u
ch
as h
ap
p
in
ess
,
s
o
r
r
o
w
an
d
an
g
e
r
)
.
Get
th
e
im
ag
es r
ea
d
y
f
o
r
p
r
o
c
ess
in
g
b
y
d
o
in
g
th
in
g
s
lik
e
r
esizin
g
,
n
o
r
m
alis
in
g
p
ix
el
v
alu
es a
n
d
a
d
d
in
g
m
e
tad
ata
if
n
ec
ess
ar
y
.
(
b
)
Sty
le
o
f
th
e
m
o
d
el
(
Step
2
)
: CNN sty
le
is
ex
p
lain
ed
h
er
e.
-
L
ay
er
f
o
r
in
p
u
t:
t
ak
es
p
ictu
r
es
with
a
ce
r
tain
d
im
en
s
io
n
ality
.
A
lay
er
o
f
co
n
v
o
l
u
tio
n
s
:
d
ea
c
tiv
ate
th
e
p
ictu
r
e
f
u
n
ctio
n
ality
.
Af
ter
e
v
er
y
c
o
n
v
o
lu
tio
n
al
lay
er
,
th
e
r
e
s
h
o
u
l
d
b
e
ac
tiv
atio
n
f
u
n
ct
io
n
s
(
lik
e
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
)
.
L
ay
er
m
er
g
i
n
g
(
e.
g
.
,
m
ax
p
o
o
lin
g
)
to
r
e
d
u
ce
s
p
atial
d
im
e
n
s
io
n
s
an
d
im
p
o
r
tan
t le
ad
in
g
f
u
n
ctio
n
s
.
-
Flatten
in
g
lay
er
:
b
ef
o
r
e
f
ee
d
in
g
th
e
o
u
tp
u
t
o
f
th
e
c
o
n
v
o
lu
t
io
n
al
lay
er
s
in
to
f
u
lly
co
n
n
ec
ted
lay
er
s
,
f
latten
th
em
.
-
I
n
ter
co
n
n
ec
ted
lay
e
r
s
:
u
s
e
ch
ar
ac
ter
is
tics
ex
tr
ac
ted
b
y
co
n
v
o
lu
tio
n
al
lay
e
r
s
to
d
o
ca
teg
o
r
izatio
n
.
I
n
th
e
r
esu
lt lay
er
,
th
e
r
e
ar
e
as m
an
y
n
er
v
e
ce
lls
as th
er
e
ar
e
cla
s
s
es (
e.
g
.
,
em
o
tio
n
al
s
tates).
(
c)
Fo
r
war
d
p
r
o
p
ag
atio
n
(
Step
3
)
:
a
p
p
ly
th
e
C
NN's
p
r
ed
ef
in
ed
lay
er
s
to
th
e
in
p
u
t
p
ictu
r
es.
Dete
r
m
in
e
th
e
n
etwo
r
k
'
s
o
u
tp
u
t b
y
u
s
in
g
th
e
lear
n
ed
b
iases
an
d
weig
h
ts
.
(
d
)
Dete
r
m
in
e
l
o
s
s
(
Step
4
)
:
t
o
f
in
d
th
e
d
if
f
er
en
ce
b
etwe
en
th
e
a
ctu
al
an
d
p
r
o
jecte
d
lab
els,
u
s
e
a
lo
s
s
f
ea
tu
r
e
th
at
is
ex
ce
llen
t
f
o
r
ca
teg
o
r
y
task
s
,
s
u
ch
as
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
.
Fifth
ly
,
wh
en
wo
r
k
in
g
with
f
ac
to
r
s
lik
e
weig
h
ts
an
d
p
r
e
d
is
p
o
s
itio
n
s
,
u
s
e
b
ac
k
p
r
o
p
a
g
at
io
n
to
d
eter
m
in
e
t
h
e
lo
s
s
f
u
n
ctio
n
'
s
s
lo
p
es.
R
ev
is
e
th
e
r
eq
u
ir
em
en
ts
with
t
h
e
h
elp
o
f
an
o
p
tim
izatio
n
p
r
o
g
r
am
m
e
lik
e
R
MSp
r
o
p
,
Ad
am
o
r
SGD.
(
e)
Data
t
r
ain
in
g
(
Step
5
)
:
p
u
t
t
h
e
C
NN
th
r
o
u
g
h
its
p
ac
es
with
th
e
ca
teg
o
r
is
ed
d
ataset.
R
e
p
ea
t
iter
ativ
ely
ac
r
o
s
s
p
ictu
r
e
b
atch
es,
d
o
in
g
f
o
r
war
d
p
r
o
p
a
g
atio
n
,
lo
s
s
co
m
p
u
tatio
n
an
d
b
ac
k
p
r
o
p
a
g
at
io
n
.
C
o
n
tin
u
e
u
n
til th
e
m
er
g
i
n
g
r
e
q
u
ir
em
e
n
t
s
ar
e
m
et
o
r
u
n
til a
s
p
ec
if
ied
n
u
m
b
er
o
f
d
ates h
a
v
e
p
ass
ed
.
(f)
E
v
alu
atio
n
(
Step
6
)
:
t
r
y
o
u
t
th
e
s
ea
s
o
n
ed
m
o
d
el
o
n
a
f
r
esh
b
atch
o
f
v
alid
atio
n
d
ata
to
s
ee
h
o
w
well
it
p
er
f
o
r
m
s
.
E
v
alu
ate
p
a
r
am
eter
s
lik
e
F1
-
s
co
r
e,
r
ec
all,
ac
c
u
r
ac
y
an
d
p
r
ec
is
io
n
.
(
g
)
Ver
if
y
in
g
(
Step
7
)
:
t
ak
e
a
lo
o
k
at
th
e
ex
p
e
r
ien
ce
d
m
o
d
el
o
n
its
g
en
er
alis
atio
n
e
f
f
icien
cy
o
n
a
d
if
f
er
en
t
test
.
T
o
v
er
if
y
t
h
e
d
esig
n
'
s
ef
f
icac
y
o
n
h
id
d
e
n
d
ata,
c
o
m
p
u
te
co
m
p
ar
ab
le
e
v
alu
atio
n
m
ea
s
u
r
es.
(
h
)
Op
tim
izatio
n
an
d
f
in
e
-
tu
n
in
g
(
Step
8
)
:
a
d
j
u
s
t
d
esig
n
s
ty
le
a
n
d
h
y
p
er
p
ar
am
eter
s
(
s
u
c
h
as
lear
n
in
g
r
ate
an
d
b
atch
d
im
en
s
io
n
)
in
lig
h
t
o
f
test
r
esu
lts
.
T
o
av
o
id
o
v
er
f
itti
n
g
,
u
s
e
ap
p
r
o
ac
h
es
s
u
ch
as
f
ailu
r
e
o
r
L
2
r
eg
u
lar
is
atio
n
.
(
i)
E
x
ec
u
tio
n
(
Step
9
)
:
u
s
e
f
r
esh
f
ac
e
im
ag
es
to
tr
ain
t
h
e
s
k
illed
v
er
s
io
n
to
m
ak
e
in
f
er
en
ce
s
.
I
n
co
r
p
o
r
ate
th
e
m
o
d
el
in
t
o
s
y
s
tem
s
o
r
ap
p
s
f
o
r
p
r
ac
tical
u
s
e,
s
u
c
h
d
is
co
v
e
r
in
g
em
o
tio
n
s
in
v
id
e
o
s
o
r
u
s
in
g
it
in
r
ea
l
-
tim
e.
Fig
u
r
e
3
.
Face
em
o
tio
n
u
s
in
g
C
NN
3
.
6
.
S
peec
h
em
o
t
io
n
us
ing
CNN
a
nd
M
F
CC
All th
e
s
tag
es o
f
th
is
p
r
o
ce
s
s
ar
e
ex
p
lain
ed
in
Fig
u
r
e
4
.
(
a)
Stag
e
1
d
ata
p
r
ep
a
r
atio
n
:
N
o
r
th
Am
er
ican
ac
ce
n
t
p
h
r
ases
m
ay
b
e
f
o
u
n
d
i
n
th
e
r
y
er
s
o
n
au
d
io
-
v
is
u
al
d
atab
ase
o
f
em
o
tio
n
al
s
p
ee
ch
an
d
s
o
n
g
(
R
AVDE
SS
)
.
E
x
p
r
e
s
s
io
n
s
o
f
p
leasu
r
e,
s
ad
n
ess
,
a
n
g
er
,
s
u
r
p
r
is
e
an
d
d
is
g
u
s
t
ar
e
all
p
ar
t
o
f
th
e
lan
g
u
ag
e
we
u
s
e.
T
h
er
e
ar
e
two
tier
s
o
f
em
o
tio
n
al
in
ten
s
it
y
(
r
eg
u
lar
an
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
24
:
1
4
4
8
-
1
4
5
9
1454
s
o
lid
)
an
d
a
th
ir
d
,
m
o
r
e
s
u
b
tle,
lev
el
o
f
em
o
tio
n
.
C
o
llect
a
c
o
llectio
n
o
f
au
d
i
o
clip
s
lab
elle
d
with
v
a
r
io
u
s
em
o
tio
n
al
s
tates.
-
Get
th
e
au
d
i
o
f
iles
r
ea
d
y
:
t
h
e
au
d
io
f
iles
m
u
s
t
b
e
p
r
ep
ar
ed
b
ef
o
r
e
an
y
th
i
n
g
else.
A
n
in
d
iv
id
u
al'
s
d
is
tin
ct
ch
ar
ac
ter
in
t
h
e
f
ile'
s
th
ir
d
d
ig
it,
w
h
ich
r
e
p
r
esen
ts
th
e
em
o
tio
n
k
ey
,
allo
ws
th
e
s
o
u
n
d
f
r
am
ewo
r
k
to
r
ec
o
g
n
is
e
d
if
f
er
en
t
em
o
tio
n
s
.
Fiv
e
d
is
tin
ct
em
o
tio
n
s
:
ca
lm
,
an
g
r
y
,
w
o
r
r
ied
,
d
is
g
u
s
ted
,
an
d
s
u
r
p
r
is
e
wh
ich
ar
e
p
ar
t
o
f
th
e
d
ata
s
et.
T
r
an
s
f
o
r
m
th
e
u
n
p
r
o
ce
s
s
ed
au
d
io
s
am
p
les
in
t
o
a
s
u
itab
le
f
o
r
m
at
s
u
ch
as
wav
ef
o
r
m
au
d
io
f
ile
(
W
AV)
f
iles
.
E
d
it
th
e
au
d
io
s
am
p
les
s
u
ch
th
at
MFC
C
s
ar
e
n
o
t
p
r
esen
t.
(
b
)
Stag
e
2
cr
ea
tin
g
th
e
m
o
d
el
:
w
h
en
d
ec
id
in
g
o
n
s
p
ee
c
h
tag
s
,
class
if
icat
io
n
ac
co
r
d
in
g
to
th
e
n
u
m
b
e
r
o
f
class
es
is
r
eq
u
ir
ed
.
T
h
e
f
o
ll
o
win
g
ca
teg
o
r
ies
ar
e:
class
if
icatio
n
:
g
o
o
d
an
d
p
o
o
r
ex
ce
llen
t
:
u
n
well,
co
n
ten
t.
Po
o
r
:
a
n
x
iety
,
d
ep
r
ess
io
n
,
an
d
ir
r
itab
ilit
y
.
An
g
e
r
,
s
ad
n
ess
,
co
n
ten
tm
e
n
t,
f
ea
r
,
an
d
tr
an
q
u
illy
ar
e
th
e
class
es.
E
m
o
tio
n
al
s
tate:
f
u
r
y
,
d
ep
r
ess
io
n
,
jo
y
,
wo
r
r
y
,
ir
r
itab
ilit
y
,
tem
p
er
,
an
d
d
r
ea
d
.
-
Def
in
e
th
e
C
NN
a
lg
o
r
ith
m
h
e
r
e:
t
h
e
in
p
u
t la
y
e
r
is
ca
p
ab
le
o
f
r
ec
eiv
in
g
MFC
C
s
.
-
L
ay
er
s
o
f
co
n
v
o
l
u
tio
n
:
u
s
e
o
n
e
-
d
im
e
n
s
io
n
al
c
o
n
v
o
lu
tio
n
al
f
ilter
s
to
p
ick
u
p
o
n
p
atter
n
s
in
th
e
s
u
r
r
o
u
n
d
in
g
a
r
ea
.
-
L
ay
er
s
f
o
r
p
o
o
lin
g
:
r
ed
u
ce
s
p
atial
d
im
en
s
io
n
s
b
y
d
o
wn
s
a
m
p
lin
g
attr
ib
u
te
m
a
p
s
.
Div
id
e
th
e
in
p
u
t
r
esu
lt
in
to
co
m
p
letely
co
n
n
ec
ted
lay
er
s
b
y
f
latten
in
g
th
e
lay
er
.
E
x
ec
u
te
ca
teg
o
r
y
b
ased
o
n
d
r
awn
o
u
t
attr
ib
u
tes:
f
u
lly
co
n
n
ec
ted
lay
er
s
.
T
h
e
o
u
tp
u
t
lay
er
co
n
s
i
s
ts
o
f
n
er
v
e
ce
lls
th
at
ar
e
ac
tiv
ated
b
y
s
o
f
tm
ax
an
d
co
r
r
esp
o
n
d
t
o
d
if
f
er
en
t e
m
o
tio
n
ty
p
es.
(
c)
Stag
e
3
d
ata
tr
ain
in
g
:
h
e
r
e,
y
o
u
'
ll
b
u
ild
th
e
C
NN
m
o
d
el's
l
ay
er
s
,
ch
o
o
s
e
th
e
ac
tiv
atio
n
f
u
n
ctio
n
,
m
a
k
e
th
e
ap
p
r
o
p
r
iate
d
ec
is
io
n
an
d
t
ell
So
f
tm
ax
to
s
p
lit
th
e
d
is
cu
s
s
io
n
in
to
m
an
y
team
s
.
Ma
k
e
u
s
e
o
f
tr
ain
in
g
d
ata
to
h
o
n
e
th
e
m
o
d
el
an
d
t
h
en
test
it
with
n
ew
d
ata.
E
v
alu
ate
th
e
esti
m
ated
co
s
t
in
r
elatio
n
to
th
e
ac
tu
al
co
s
ts
.
T
h
e
ac
c
u
r
ac
y
o
f
th
e
s
ettin
g
is
s
h
o
wn
b
y
t
h
is
co
m
p
a
r
is
o
n
.
Div
i
d
e
th
e
d
ata
s
et
in
to
th
r
e
e
p
ar
ts
: tr
ain
in
g
,
r
ec
o
g
n
itio
n
an
d
test
in
g
.
(
d
)
Stag
e
4
f
ir
s
t,
we
n
ee
d
to
tr
ai
n
th
e
C
NN
m
o
d
el:
p
e
r
f
o
r
m
iter
atio
n
s
u
s
in
g
s
ets
o
f
MFC
C
s
.
I
n
itiate
b
ac
k
p
r
o
p
a
g
atio
n
,
f
o
r
war
d
b
r
ee
d
in
g
an
d
lo
s
s
co
m
p
u
tatio
n
.
Use
an
o
p
tim
is
atio
n
f
o
r
m
u
la
(
s
u
c
h
as
Ad
am
o
r
SGD)
to
u
p
d
ate
th
e
p
ar
am
eter
s
.
Fo
u
r
th
p
iece
o
f
ad
v
ice:
test
th
e
v
eter
an
m
o
d
el
o
n
th
e
v
alid
atio
n
d
ata
s
et:
d
eter
m
in
e
m
ea
s
u
r
es lik
e
F1
-
s
c
o
r
e,
r
ec
all,
ac
c
u
r
ac
y
a
n
d
p
r
ec
i
s
io
n
.
(
e)
Stag
e
5
ass
ess
in
g
:
t
ak
e
a
lo
o
k
at
th
e
test
ed
m
o
d
el
u
s
in
g
th
e
tr
ain
in
g
s
et:
f
in
d
an
aly
tical
m
etr
ics
to
s
ee
h
o
w
well
g
en
er
alis
atio
n
wo
r
k
s
.
(f)
Stag
e
6
test
in
g
:
m
ak
e
av
ailab
le
th
e
tr
ain
ed
m
o
d
el
f
o
r
u
s
e
in
in
f
er
r
in
g
f
r
esh
au
d
io
s
am
p
les
in
r
ea
l
tim
e:
t
ak
e
au
d
io
s
am
p
les
as
in
p
u
t.
Get
MF
C
C
s
o
u
t
o
f
th
e
d
ata.
I
n
p
u
t
th
e
MFC
C
s
in
to
th
e
ex
p
er
t
d
esig
n
to
f
o
r
etell
th
e
em
o
tio
n
.
Fig
u
r
e
4
.
Sp
ee
c
h
em
o
tio
n
u
s
in
g
C
NN
4.
RE
SU
L
T
S
I
n
o
r
d
er
to
ac
c
o
m
p
lis
h
th
is
task
,
we
h
av
e
t
r
ain
ed
a
C
NN
alg
o
r
ith
m
u
s
in
g
th
e
R
AVDE
SS
S
o
u
n
d
Data
s
et
to
id
en
tify
em
o
tio
n
s
i
n
s
p
o
k
en
lan
g
u
a
g
e
an
d
we
h
av
e
u
tili
ze
d
th
e
em
o
tio
n
f
ac
ial
e
x
p
r
ess
io
n
p
ictu
r
es
d
ataset
to
id
en
tify
f
ac
ial
e
x
p
r
e
s
s
io
n
s
.
Usi
n
g
th
e
s
am
e
alg
o
r
ith
m
,
we
h
av
e
tr
ain
ed
b
o
th
t
h
e
s
p
ee
ch
d
ataset
an
d
th
e
f
ac
e
p
h
o
to
d
ataset
u
s
in
g
C
NN
v
er
s
io
n
s
.
Yo
u
ca
n
f
in
d
t
h
e
f
ac
e
d
ataset
wh
ich
co
n
tain
s
th
e
p
h
o
to
g
r
ap
h
s
o
f
em
o
tio
n
f
ac
es sh
o
wn
in
Fig
u
r
e
5
in
th
e
'
Data
s
et
'
f
o
ld
er
.
W
e
ar
e
s
h
o
win
g
d
ataset
o
f
au
d
i
o
f
iles
in
Fig
u
r
e
6
an
d
th
is
d
ataset
s
av
ed
in
s
id
e
s
p
ee
ch
em
o
tio
n
d
ataset
f
o
ld
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Deep
-
S
F
E
R
:
d
ee
p
c
o
n
vo
l
u
tio
n
a
l n
eu
r
a
l
n
etw
o
r
k
a
n
d
MFCC
a
n
effec
tive
…
(
R
a
vi
Gu
mm
u
la
)
1455
Fig
u
r
e
5
.
E
m
o
tio
n
f
aces
Fig
u
r
e
6
.
E
m
o
tio
n
a
u
d
io
d
atasets
Ob
s
er
v
ed
h
er
e
ar
e
2
4
s
ep
ar
ate
s
tar
r
ec
o
r
d
in
g
s
o
f
h
u
m
an
s
p
ee
ch
,
s
p
an
n
in
g
8
d
is
tin
ct
em
o
tio
n
al
s
tates
:
n
eu
tr
al,
tr
an
q
u
il,
d
elig
h
ted
,
d
ep
r
ess
in
g
,
u
p
s
et,
af
r
ai
d
,
d
is
g
u
s
ted
,
an
d
asto
n
is
h
ed
.
T
o
s
ee
au
d
io
in
f
o
r
m
atio
n
,
ju
s
t
g
o
to
an
y
f
o
ld
er
in
th
e
ab
o
v
e
d
is
p
lay
.
E
v
er
y
wav
e
f
ile
is
as
s
o
ciate
d
with
a
n
u
m
b
er
th
at
is
s
p
lit
b
y
th
e
'
-
'
s
y
m
b
o
l;
th
r
ee
d
ig
its
r
ep
r
esen
t
th
e
id
,
o
n
e
r
ep
r
esen
ts
th
e
g
en
d
er
,
an
d
th
e
th
ir
d
p
lace
s
a
v
alu
e
b
etwe
en
o
n
e
an
d
eig
h
t th
at
r
ep
r
esen
ts
th
e
em
o
ti
o
n
.
T
h
e
d
ata
s
et
lo
ad
ig
a
n
d
p
r
e
-
p
r
o
ce
esin
g
d
ata
s
h
o
wn
i
n
Fig
u
r
e
s
7
an
d
8
.
Fig
u
r
e
7
.
Data
s
et
lo
ad
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
In
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
3
6
,
No
.
3
,
Dec
em
b
er
20
24
:
1
4
4
8
-
1
4
5
9
1456
Fig
u
r
e
8
.
Pre
-
p
r
o
ce
s
s
in
g
d
ata
Fig
u
r
e
9
ex
p
lain
s
ab
o
u
t
th
e
alg
o
r
ith
m
r
esu
lts
o
f
C
NN
p
r
o
ce
s
s
.
T
h
e
g
r
ap
h
s
h
o
wn
in
Fig
u
r
e
1
0
ex
p
lain
s
th
at
b
o
th
alg
o
r
ith
m
s
s
h
o
wn
in
Fig
u
r
e
9
ac
h
iev
ed
a
p
r
ec
is
io
n
o
f
1
an
d
b
o
th
f
o
r
m
u
lae
ac
h
iev
ed
a
lo
s
s
v
alu
e
o
f
0
,
with
th
e
x
-
ax
is
r
ep
r
esen
tin
g
e
p
o
c
h
an
d
th
e
y
-
ax
is
r
ep
r
esen
tin
g
p
r
ec
is
io
n
an
d
lo
s
s
v
alu
es
r
esp
ec
tiv
ely
.
T
h
e
ac
cu
r
ac
y
o
f
f
ac
ial
ex
p
r
ess
io
n
s
(
g
r
ee
n
lin
e)
an
d
v
o
ca
lizatio
n
s
(
b
lu
e
lin
e)
ar
e
s
h
o
wn
in
t
h
e
g
r
ap
h
ab
o
v
e.
T
o
p
o
s
t
a
f
ac
e
p
h
o
to
g
r
ap
h
an
d
s
ee
th
e
r
esu
lts
s
h
o
wn
in
Fig
u
r
e
s
11
to
1
3
,
click
th
e
"Pr
e
d
ict
Facial
E
m
o
tio
n
"
b
u
tto
n
r
ig
h
t n
o
w.
Fig
u
r
e
9
.
Alg
o
r
ith
m
r
esu
lts
(
C
NN)
Fig
u
r
e
1
0
.
E
m
o
tio
n
lo
s
s
an
d
a
cc
u
r
ac
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2
5
0
2
-
4
7
52
Deep
-
S
F
E
R
:
d
ee
p
c
o
n
vo
l
u
tio
n
a
l n
eu
r
a
l
n
etw
o
r
k
a
n
d
MFCC
a
n
effec
tive
…
(
R
a
vi
Gu
mm
u
la
)
1457
Fig
u
r
e
1
1
.
T
est
im
ag
e
r
esu
lts
Fig
u
r
e
1
2
.
Au
d
io
test
f
iles
Fig
u
r
e
1
3
.
Sp
ee
ch
em
o
tio
n
tes
t
5.
CO
NCLU
SI
O
N
On
e
p
r
o
m
is
in
g
av
en
u
e
f
o
r
ad
v
an
ce
m
en
t
in
em
o
tio
n
al
co
m
p
u
tin
g
is
th
e
co
m
b
in
atio
n
o
f
f
ac
ial
an
d
v
o
ca
l
an
aly
s
is
with
d
ee
p
lear
n
in
g
'
s
C
NN
s
f
o
r
em
o
ti
o
n
id
e
n
tific
atio
n
.
B
y
c
o
m
b
in
in
g
h
el
p
f
u
l
em
o
tio
n
al
c
u
es
f
r
o
m
s
p
ee
ch
s
ig
n
als
with
th
e
r
o
b
u
s
tn
ess
o
f
C
NNs
f
o
r
s
p
a
tial
f
u
n
ctio
n
ex
tr
ac
tio
n
,
th
is
m
eth
o
d
en
a
b
les
th
e
d
etailed
an
aly
s
is
o
f
f
ac
e
em
o
tio
n
s
.
T
h
e
co
m
b
i
n
atio
n
o
f
b
o
th
v
is
u
al
an
d
au
d
ito
r
y
m
o
d
al
ities
im
p
r
o
v
es
th
e
d
esig
n
'
s
ca
p
ac
ity
to
ca
p
tu
r
e
t
h
e
n
u
an
ce
an
d
co
m
p
lex
ity
o
f
h
u
m
an
em
o
tio
n
s
.
T
h
is
m
ak
es
th
e
in
d
u
s
tr
ialized
Evaluation Warning : The document was created with Spire.PDF for Python.