TELKOM
NIKA
, Vol.13, No
.2, June 20
15
, pp. 571 ~ 5
7
7
ISSN: 1693-6
930,
accredited
A
by DIKTI, De
cree No: 58/DIK
T
I/Kep/2013
DOI
:
10.12928/TELKOMNIKA.v13i2.1124
571
Re
cei
v
ed
Jan
uary 3, 2015;
Re
vised Ma
rch 17, 2015; A
c
cepted Ap
ril 3, 2015
Integrated Syst
em Design for Broadcast Program
Infringement Detection
Sukma
w
ati Nur
Endah*
1
, Satriy
o
Adhy
2
, S
u
ti
kno
3
Informatics De
partment, F
a
cu
lt
y
of Scie
nce a
nd Mathem
atic
s, Universitas
Dipo
n
e
gor
o
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: sukma_n
e@
und
ip.ac.i
d
1
; satri
y
o
@
un
di
p.ac
.id
2
; tik@undi
p.ac.id
3
A
b
st
r
a
ct
Superv
i
sio
n
of
televisi
on
an
d radi
o bro
adc
ast pr
ogr
ams
by the Ko
misi Pe
n
y
iara
n Indo
nes
ia Jaw
a
T
enga
h is
still
bei
ng
carri
ed
out
ma
nua
lly
i.e. direct
s
u
p
e
r
v
ision
by
hu
mans. T
h
e
proc
ess certai
nly
h
a
s
some w
eakn
e
sses relate
d to hu
ma
n erro
r, such as
tiredn
ess an
d w
eary eyes. T
h
erefore, w
e
n
e
e
d
intell
ig
ent soft
w
a
re that cou
l
d
auto
m
atic
ally
detect br
oadc
a
s
t infring
e
m
e
n
t. Up to n
o
w
,
research
in th
is a
r
ea
has n
o
t b
een
c
a
rried
o
u
t. This researc
h
w
a
s
ai
me
d
at d
e
si
g
n
in
g a
n
i
n
tegra
t
ed syste
m
to
detect br
oadc
a
s
t
infrin
ge
me
nt, inclu
d
in
g archit
ecture des
ign
and
ma
in
mod
u
le i
n
terface d
e
sig
n
. T
w
o main stag
es in thi
s
system
are
Indonesian-
language speech
recogn
ition and detection
of
infrin
gem
e
nts
of the br
oadcast
progr
a
m
. W
i
th the meth
od o
f
Mel F
r
equen
cy Cepstral C
oefficie
n
ts (MFCC) and H
i
d
d
en Markov Mo
de
l
(HMM), a speech reco
gnition application
t
hat
uses
1050 s
a
m
p
le
data produ
ces
about 70% accur
a
cy rat
e
.
This rese
arch
w
ill conti
n
u
e
t
o
i
m
p
l
e
m
ent t
he
desi
gn cr
e
a
ted
usin
g s
p
eech
reco
gn
iti
on app
licati
o
n
s
that
have b
e
e
n
dev
elo
ped.
Ke
y
w
ords
:
Br
oadc
ast Infring
e
ment, Infrin
g
e
ment D
e
tecti
on,
Ko
mis
i
Pe
nyiar
an In
do
ne
sia (KPI), Spe
e
c
h
Reco
gniti
on
1. Introduc
tion
The p
ubli
c
’s
great i
n
terest
in
watchi
ng t
e
levisio
n
o
r
li
stenin
g
to th
e ra
dio
pro
m
pted the
govern
m
ent t
o
pa
ss the B
r
oad
ca
sting A
c
t No 3
2
of
2
002 in o
r
de
r t
hat t
he broad
ca
st media
could
perfo
rm the role of a healthy public info
rmation
se
rvi
c
e. One of its provisio
ns
was to set up t
he
formation
of t
he Komi
si P
e
nyiaran
Indo
n
e
sia
(KPI
)
wit
h
auth
o
rity to
overse
e the
i
m
pleme
n
tation
of the
rule
s
a
nd
cod
e
of
condu
ct a
s
we
ll as th
e
broa
dca
s
ting
sta
n
dard
s
of b
r
oa
dca
s
t p
r
o
g
ra
ms,
and also to impose pen
alties on tho
s
e
who violate
them. In perfo
rming its duti
e
s, the KPI issue
d
Reg
u
lation
o
f
the Komisi
Penyiara
n Indon
esi
a
No
. 02/P/KPI/1
2/2009
abo
u
t
the Co
de
of
Broad
ca
sting
Co
ndu
ct [1]
and
KPI Re
gulation
No
03/P/KPI/12/2009
abo
ut the Broad
ca
st
ing
Program Sta
ndards [2]. In acco
rda
n
ce
with the
le
gislation, the co
des
of broad
ca
sting
con
d
u
ct
and b
r
o
a
d
c
a
s
ting
pro
g
ra
m stan
da
rd
s, the infrin
g
e
ments can
be a
n
a
u
d
i
o/spe
e
ch-b
a
s
ed
infringe
ment,
a visual in
fringem
ent o
r
a co
ntent-based infrin
g
e
ment. Som
e
example
s
o
f
audio
-
ba
se
d
infring
e
men
t
s in a
b
r
oa
dca
s
ting
pro
g
ram
can in
clud
e ha
rsh
wo
rd
s, cursing,
hara
s
sing, in
sulting o
r
de
gradi
ng mino
rities an
d
ma
rginali
s
e
d
so
ciety grou
ps
su
ch a
s
a group
with a spe
c
i
f
ic job (dom
estic worke
r
s, se
cu
rity g
uard
s
), a group
that ha
s an abe
rration
(tran
s
sexual
s), a gro
up wit
h
uno
rdin
ary
size and
phy
sical form (bu
ck te
eth, overweig
ht, midg
et,
strabi
sm
us),
a group wi
th physi
cal
di
sabilities (deaf,
blind,
mute),
a group
who have
a ment
al
retardation or disability (autism, idioti
sm),
and a group of people wi
th specific di
seases
(HIV/AIDS, lepro
s
y, epile
psy, Alzheim
e
r’s, la
tah
)
. Examples of
visual infrin
gement
s are
the
exploitation o
f
body parts
whi
c
h a
r
e co
mmonly co
ns
idere
d
to be
able to ge
nerate sexual
de
sire,
su
ch a
s
thig
hs, butto
cks,
brea
st
s an
d/or ge
ni
tals. T
he la
st one,
an exampl
e
of conte
n
t-b
a
s
ed
infringe
ment,
is for exampl
e an
ann
oun
cement that
promotes an
d e
n
co
ura
g
e
s
p
r
ostitution to
b
e
accepte
d
by religion an
d society.
The KPI con
s
ists of KPI Pusat and KPI Dae
r
ah
(KPID) at the p
r
ov
incial level. O
ne KPID
that has al
ready bee
n formed i
s
KPID in Central Java. In
carry
ing out
surv
eillance of
both
television
and radio broadcasting, KPID Cent
ral
Java
still does thi
s
manua
lly (di
r
ect
supervisi
on
by people
)
. Several televisions a
r
e turn
ed on to
be
see
n
and mo
nitored by off
i
cers to find
out
wheth
e
r th
e
delivery of
broad
ca
st
cont
ent is ag
ain
s
t the rul
e
s.
T
h
is
ha
s
certa
i
nly had
som
e
wea
k
n
e
sse
s
related to h
u
m
an erro
r, such a
s
tire
d
ness an
d we
ary eyes. Inconsi
s
ten
c
ie
s
in
asse
ssm
ent can
al
so
occur be
cau
s
e
of
the su
bj
e
c
tivity of tho
s
e
su
pe
rvisin
g the
broad
cast
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 571 – 57
7
572
prog
ram
s
. T
herefo
r
e,
we nee
d inte
lligent so
ftware th
at ca
n autom
atically detect
the
infringe
ment
s of broad
ca
st conte
n
t, which has n
e
ver b
een stu
d
ied u
p
to now.
This
re
se
arch aim
s
to
de
sign
a
syste
m
that can
a
u
tomatically i
dentify infrin
gement
s
committed
a
gain
s
t the la
w, the
cod
e
s of br
oad
ca
sting condu
ct
and th
e b
r
o
adcast
pro
g
ram
stand
ard
s
, wi
th the result that it can hel
p the
KPI’s perform
an
ce,
esp
e
ci
ally KPID Cent
ral Ja
va,
in controlling broadcast content. The
sy
stem
has t
w
o main
stages. T
hey
are
Indone
sia
n
-la
ngua
ge sp
e
e
ch recogniti
on and dete
c
tion of infri
ngeme
n
ts in
the broad
cast
prog
ram. Au
d
i
o-ba
se
d infri
ngeme
n
t ca
n
be dete
c
ted
a
u
tomatically
by the existe
nce
of a spee
ch
recognitio
n
a
pplication. As re
gards p
r
ogra
m
s b
r
oa
dca
s
t in the
Indone
sian
langu
age, th
ey
definitely re
q
u
ire
a
spee
ch re
co
gnition
system
in t
he Indo
ne
sia
n
lang
uag
e.
The
re
sea
r
ch
of
sound in Indonesi
a
is relati
vely limited, such as
the id
entification of
gamelan i
n
st
rume
nts [3] a
nd
an Indone
si
an-la
ngu
age
spee
ch re
cog
n
ition sy
stem. An Indon
esi
an-l
a
ngua
ge sp
e
e
ch
recognitio
n
system is still l
i
mited to writi
ng a simp
l
e
messag
e on
a mobile devi
c
e; it is far be
hind
E
nglish
-
spee
ch re
cog
n
it
io
n
sy
st
e
m
s,
o
f
whic
h
seve
ral h
a
ve al
re
ady bee
n a
p
p
lied to
relat
e
d
fields [4]-[6].
For that, Ind
o
nesi
an-l
ang
u
age
sp
e
e
ch reco
gnition
ne
eds to
be fu
rther
develop
e
d
,
so it
can
be
u
s
ed
in
vario
u
s
field
s
,
espe
cially
in
the
a
u
tomatic dete
c
tion
of infri
n
gement
s i
n
a
udio
broa
dcast p
r
o
g
ram
s
.
Re
sea
r
ch o
n
sp
ee
ch
re
co
gnition
starte
d in th
e 1
9
5
0
s [7]. Sp
eech re
co
gnition
ca
n b
e
defined
a
s
th
e p
r
o
c
e
s
s of
co
nvertin
g
v
o
ice
si
gnal
s i
n
to the
ra
nks of the
word,
by ap
plying
a
spe
c
ific
algo
ri
thm that is i
m
pleme
n
ted i
n
a comp
uter pro
g
ram. It i
s
al
so often
called Spe
e
ch
to
Text. There
are t
w
o m
a
in
pro
c
e
s
se
s in
spe
e
ch reco
gnition: featu
r
e extra
c
tion
and
re
cog
n
ition.
Variou
s meth
ods h
a
ve bee
n develop
ed to pro
duce
a
high level of accuracy. Fe
ature extra
c
ti
on
techni
que
s t
hat have b
e
en devel
ope
d incl
ude
Li
near
Predi
cti
v
e Analysis
(LPC), Cep
s
tral
Analysis [8], Mel Freq
uen
cy Cep
s
tral
Coeffici
ents
(MFCC) [9], Wavelet Ceps
tral Coeffic
i
ents
(WCC) [10] a
nd ret
r
ieval b
a
se
d
prosodi
c features [1
1]. Basically
t
here
are thre
e app
roa
c
h
e
s to
spe
e
ch re
co
g
n
ition, namel
y [12] the acousti
c-p
hon
etic app
ro
ach, pattern
recog
n
ition app
ro
a
c
h
and
artifici
al i
n
telligen
ce
a
ppro
a
ch. Th
e
sp
ee
ch
re
co
gnition te
ch
ni
que
that
is
includ
ed
in
pattern
recognitio
n
i
s
the
Hid
d
e
n
Ma
rkov M
odel
(H
MM) and S
upp
ort Vector Ma
chin
e (SVM
) [8].
Re
sea
r
ch co
ndu
cted by a
grou
p of re
se
arche
r
s fr
om
the Lab Ri
set
Sistem Ce
rd
as Jurusan Ilmu
Komputer/Inf
o
rmati
k
a h
a
s been
abl
e to
prove th
at th
e Supp
ort Ve
ctor
Ma
chine
(SVM) m
e
th
od
can
be u
s
ed t
o
cla
s
sify harsh
words
and
non-ha
rsh
word
s. Swe
a
r
words
co
nstit
u
te a violatio
n of
Law No 3
2
o
n
b
r
oa
dcastin
g
.
The
results
sh
owed
that spo
k
e
n
wo
rd
s can
b
e
cla
s
sified with
a h
i
gh
enou
gh de
gree of a
c
cura
cy (72.5%
to
92.5%) [13].
The soun
d f
eature
s
u
s
e
d
are the val
u
e of
pitch, be
cau
s
e at the begi
nning of the
study, it
was
analysed a
s
being u
s
eful f
o
r di
stingui
sh
ing
the pron
un
cia
t
ion of the wo
rd in the
form
of insults an
d non-i
n
sults
[14].
Singh et al. (2012) and O’Shaugh
nes
s
y
(2008) [12],[15] ment
ioned that the tec
hnique
for
spee
ch reco
gnition which co
nsi
s
ts
of
hund
re
ds of thou
sand
s of
words
and that i
s
still
accepte
d
up
to no
w i
s
the
Hidd
en M
a
rkov Model
(HMM), which a
ppea
red
in
1
975. Acco
rdi
ng to
Rabi
ner (198
9),
HMM i
s
a
sto
c
ha
stic p
r
oce
s
s that
o
c
curs twi
c
e,
with one
of th
e
m
bei
ng
not
a
dire
ct observ
a
tion. A hidden sto
c
ha
stic process can
be observed
only through
another
set of
stocha
stic p
r
oce
s
se
s that can p
r
o
d
u
c
e the se
que
nc
e
of obse
r
vatio
n
symbol
s. T
h
is is th
e re
a
s
on
that cau
s
e
s
HMM to p
e
rf
orm b
e
tter th
an othe
r met
hod
s [9]. In addition, the
HMM tech
niqu
e is
gene
rally a
c
cepted in
cu
rrent sp
ee
ch
reco
gnition
sy
stem
s (state-of-the-art) i
n
mode
rn tim
e
s
because of t
w
o reasons,
namely, its ability to m
odel the non-linea
r dependence of each unit of
the sou
nd on
the unit in question, an
d beca
u
se it
is a set of powe
r
ful analytical a
ppro
a
che
s
that
are
available
to estimate
the mod
e
l p
a
ram
e
ters
[1
2]. This
stud
y use
s
HMM
as
a meth
o
d
in
spe
e
ch re
cog
n
ition of broa
dca
s
ting p
r
og
ram
s
.
2.
Res
earc
h
Method
A gene
ral d
e
s
cription
of the sy
stem th
at is
built
ca
n be
see
n
in
Figure 1 b
e
l
o
w. The
Integrated
S
y
stem of Inf
r
inge
ment
B
r
oad
ca
sting
Program Det
e
ction (
Siste
m
Terintegra
s
i
Dete
ksi Pel
a
ngga
ran P
r
o
g
ram
Siaran
,
abbreviate
d as SINDEPROSI) is con
s
tru
c
ted
in
accordan
ce
with the gen
e
r
al pictu
r
e in
Figure 1 belo
w
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Integrated S
y
stem
Design for Broadcast
Progr
am
Infringem
ent .... (
S
ukm
a
wati Nur Endah)
573
Figure1. System De
scripti
o
n
The followi
ng
steps of the
pro
c
e
ss in SI
NDEP
R
OSI a
r
e:
a.
Audio Input
The in
put
ca
n be
the
audi
o from
eithe
r
television
or radio
pro
g
ra
m
s
that
use th
e Indo
ne
sian
langu
age.
b. Speech
Re
co
gnition
Spoke
n
audi
o broa
dcastin
g
prog
ram
s
will be re
cog
n
ise
d
and wil
l
be written in
text form. To
desi
gn the
s
e
spee
ch
re
cognition a
ppl
ication
s
,
the
step
s of the
pro
c
e
ss
ca
n be seen i
n
Figure 2.
Figure 2. Proce
ss of Spee
ch Recognitio
n
Applicatio
n
I
nput:
Audio TV/Radio in
I
ndonesian L
a
ngua
ge
Indonesian Langua
ge Speech
Recognition Appli
cations
Transcription of
Speech in a
Br
oadcast Pr
ogr
am
Output:
To
d
a
y: .
...
...
..
Date: .
...
...
..
Ch
an
n
e
l: .....
...
Results Detection:
- Violate
Articl
e
.
.
-
No I
n
fr
ingem
e
nt
Ch
an
n
e
l:......
..
Hasil Deteksi:
- Mel
a
ng
gar Pasal
.
.
-
T
i
d
ak
ad
a
p
elan
gg
ar
an
T
h
e database of the
infr
ingem
e
nt
word/phrase/ sentence
Words Dete
ction
Result
Used during the testing
Speech Input
Pr
e-
pr
ocessing
Feature Extraction
M
odel
Generation
Pattern
Classification
Cor
pus/
speech
collectio
S
p
eechTranscri
p
ti
on
T
e
sting
The Par
a
m
e
t
e
rized
sound waves
T
r
aining
Acuistic M
odel
L
a
nguage M
odel
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 571 – 57
7
574
Here is the e
x
planation of
each pro
c
e
ss.
1) Speech
input
Speech input
can b
e
a vide
o recording o
r
ta
king a
sou
nd file with .wav extension.
2
)
Pr
e
-
pr
oc
ess
i
n
g
Pr
e
-
pr
oc
ess
i
n
g
o
n
MF
CC
inc
l
u
d
e
s
D
C
-
r
e
m
o
v
a
l
, Pr
e
-
e
m
p
h
a
s
is
e
,
F
r
a
m
e
Blo
c
k
i
n
g
and
Wind
owi
ng. DC-removal i
s
used to ob
tain
the normalise
d
valu
e of the inp
u
t data, by
cal
c
ulatin
g the averag
e of the sample
uttera
n
c
e dat
a. Pre-em
ph
asi
s
e is u
s
e
d
to reduce
the noi
se
rati
o in the
sig
n
a
l and
bala
n
ce the
spe
c
tru
m
of the voice. The F
r
am
e Blockin
g
pro
c
e
s
s is u
s
ed to
cut
a vo
ice
sig
nal
of l
ong
duration
and
sh
orten
it
, in o
r
de
r to
o
b
tain the
c
h
ar
ac
te
r
i
s
t
ic s
i
gn
a
l
pe
r
i
o
d
ic
a
lly. T
h
e W
i
n
d
o
w
i
n
g
p
r
o
c
e
s
s
a
i
ms
to re
d
u
c
e
s
p
ec
tra
l
le
ak
age
or alia
sin
g
, whi
c
h i
s
the
effect of Frame
Blo
cki
n
g
that ca
use
s
the
sign
al
to become
discontin
ued.
3) Feature
extra
c
tion
The method
us
ed for this
is
the
Mel Frequenc
y
Cepstral Coeffic
i
en
ts
(MFCC). This
s
t
age
has
several p
r
ocesse
s, whi
c
h are:
i. FFT
(
Fa
st Fo
urie
r Tra
n
sfo
r
m
)
FFT is a tra
n
sformation
method to g
e
t a si
gn
al in the frequ
e
n
cy dom
ain
of the
available di
screte sig
nal
s.
ii.
Mel-F
r
eq
uen
cy Wrappin
g
Filterba
nk i
s
carrie
d out in
orde
r to d
e
termin
e the e
nergy in th
e
sou
nd
sign
al. The
freque
ncy of a sign
al is me
asu
r
ed u
s
in
g the Mel scale
.
iii.
Cep
s
trum
Mel-F
r
eq
uen
cy Cepst
r
um
is obtai
ned
from
the
DCT
pro
c
e
s
s to g
e
t the si
gnal
back in
the time dom
ain. The re
sul
t
is called the
Mel-F
r
eq
uen
cy Cep
s
tral
Coefficient (M
F
C
C).
iv.
C
e
ps
tr
a
l
F
ilter
in
g
The MF
CC
re
sults
have se
veral wea
k
ne
sses, n
a
mely
a very sen
s
it
ive low-ord
e
r
to the
spe
c
tral
slo
p
e
and a ve
ry sen
s
itive high
-order
to
noi
se. Therefore,
the ce
pstral filtering
become
s
one
of the methods to minimi
se su
ch sen
s
itivity.
4) Model
Ge
ne
ration
This p
r
o
c
e
s
s model
s the p
a
tterns
of sp
e
e
ch into text form b
a
se
d o
n
the traini
ng
data that
are u
s
ed by e
m
ploying the
Hidd
en Ma
rkov Model (HMM).
5) Pattern
Classific
a
tion
The final
stag
e take
s fo
rm
a pattern cl
assificatio
n
that
sea
r
che
s
fo
r equivale
ncy
betwe
en
the formed g
eneration mo
dels a
nd the tested d
a
ta.
c. Speech
Tra
n
s
cription
Tran
scriptio
n
of spee
ch i
s
the re
sult
of t
he spee
ch re
cognitio
n
pro
c
e
ss for broa
dca
s
tin
g
prog
ram
s
tha
t
have been i
n
clu
ded.
d. Infringem
ent
Dete
ction
From the transcription’
s
results, the avail
able text
or
word
will be checked i
n
the i
n
fringement
regul
ation wo
rds
datab
ase
.
The re
sults of thes
e che
c
ks can b
e
p
r
inted a
s
a report for th
e
broa
dcastin
g
prog
ram ta
rg
eted.
3. Resul
t
s
and
Discus
s
ion
This sy
stem has two u
s
e
r
s, each of wh
om ac
ts a
s
Admin and Op
eration
s
Officer. Admin
is the perso
n
who is in ch
arge of ma
n
aging u
s
e
r
d
a
ta, performi
ng the trainin
g
pro
c
e
ss in
the
cre
a
tion
of I
ndon
esi
an-l
a
ngua
ge
sp
ee
ch
re
co
gnitio
n
ap
plication
s
, ma
nagi
ng
the li
st of
word
infringe
ment
s and
mana
gi
ng the
releva
nt legisl
ati
on
data related t
o
Broa
dcasti
ng La
w
No
3
2
of
2002, the
Ko
misi Penyia
ran Indo
ne
sia
Reg
u
lati
on
No 0
2
/P/KPI/12/200
9 ab
o
u
t The
Co
de
of
Broad
ca
sting
Co
ndu
ct
an
d the
Komi
si
Penyia
ran
Indon
esi
a
Re
gulation
No
03/P/KPI/12/2009
about th
e Broad
ca
sting P
r
og
ram Sta
n
d
a
rd
s. Th
e O
p
er
ation
s
Officer i
s
a
n
officer from KPI
who
run
s
thi
s
sy
stem by ente
r
in
g the b
r
oa
dca
s
ting p
r
o
g
ra
m
data a
nd a
u
d
io inp
u
t both
from televi
si
on
and ra
dio p
r
o
g
ram
s
.
The re
sulting
syste
m
d
e
si
gn
in
clu
d
e
s
architectu
re
desi
gn and
main
m
odul
e
interfa
c
e
desi
gn.
a.
A
r
chit
e
c
t
u
r
e
De
sign
Archite
c
ture desi
gn is
sho
w
n in Figu
re
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Integrated S
y
stem
Design for Broadcast
Progr
am
Infringem
ent .... (
S
ukm
a
wati Nur Endah)
575
Figure 3. Architecture de
si
gn
Here is an ex
planatio
n of each mai
n
pro
g
ram.
1)
Authenticatio
n and auth
o
ri
sation
This me
nu is
use
d
as an id
entifier for users of
the sy
stem, by displaying the log-in menu
for the usern
a
me and p
a
ssword e
n
tere
d by the user
of the system
.
2)
User Data M
anag
ement
This me
nu is
use
d
to mana
ge user data.
3) Speech
Re
co
gnition
Speech
re
co
gnition i
s
a
major p
r
og
ra
m in th
e fo
rm of Ind
one
sian
-lan
gua
g
e
spee
ch
recognitio
n
a
pplication
s
.
4) Infringem
ent
Dete
ction
This me
nu is
also the m
a
in
prog
ram, whi
c
h is
u
s
e
d
to detect violati
ons of the b
r
oad
ca
st
prog
ram.
5)
Provisio
n Dat
a
Manag
eme
n
t
Provisio
n Dat
a
Manag
eme
n
t is use
d
to manag
e the d
a
ta legisl
ation
.
b.
Main Mod
u
le
Interface De
sign
Main mod
u
le
interface de
si
gn is divide
d into three p
r
o
c
ed
ure
s
a
s
follows:
Procedu
re Data Traini
ng (spe
ech_trai
n
ning_
data: voice, vartrai
nni
ng_recognitio
n_text:
cha
r
)
Initial State
: System waiting for
the dat
a input of voice traini
ng
Final State
: System issu
ed data mod
e
lling with HM
M
Algorithm
:
1)
The sy
stem a
c
cepts d
a
ta input of voice
training
2)
The sy
stem p
r
ep
ro
ce
ssi
ng
voice traini
ng
data
3)
The sy
stem e
x
tracting feat
ure extra
c
tion
4)
The sy
stem produces
data
modelling wit
h
HMM
Procedu
re Data Testing
(spee
ch_te
s
t_d
a
ta:
voice, broad
ca
sting_
p
r
og
ram_
data:
voice,
varre
co
gnitio
n_re
s
ult_text: char)
Initial State
: System waiting for data in
put
of voice test or b
r
oa
dcast pro
g
ram data
Final State
: System issu
es text recog
n
ition re
sults
Algorithm
:
1)
The sy
stem a
c
cepts d
a
ta input of
voice
test or broad
cast pro
g
ram data
2)
The sy
stem p
r
ep
ro
ce
ssi
ng
voice te
st dat
a or broad
ca
st program dat
a
3)
The sy
stem e
x
tracting the f
eature
4)
The sy
stem reco
gni
sing
sp
eech pattern
with user dat
aba
se a
s
the input
5)
System issue
s
text reco
gni
tion results
Main
Program
Autenticati
o
n
&Autorization
User Data
Mana
geme
n
t
Speec
h
Reco
gniti
on
Infrigement
Detectio
n
Provisio
n
Data
Mana
geme
n
t
Data
T
r
ainning
Data
T
e
sting
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 16
93-6
930
TELKOM
NIKA
Vol. 13, No. 2, June 20
15 : 571 – 57
7
576
Procedu
re Infringe
ment De
tection (re
c
og
niti
on_result_
text: char, pro
v
ision_
data: char,
varinfrigm
ent
_rep
ort: ch
ar)
Initial State
: System awa
i
ts input text reco
gnition re
sults
Final State
: System issu
es infrin
gem
e
n
t repo
rt
Algorithm
:
1)
The sy
stem a
c
cepts in
put text recog
n
itio
n results
2)
The sy
stem checks re
cog
n
i
tion result
s text with database of infrin
gement
s data
as the inp
u
t
3)
The sy
stem d
e
termin
es
wh
ether the recognition resul
t
s text is inclu
ded a
s
an
offence o
r
not
4)
The sy
stem stores th
e re
co
gnition re
sult
s text as an infringe
ment if the
recognitio
n
re
sults text is in
clud
ed in the
infringe
ment
data
5)
System issue
s
an infrin
ge
ment rep
o
rt
In creating
spee
ch
re
cog
n
ition, a
spe
e
ch
datab
ase is
used, in
volving 50
p
eople
wit
h
variation
s
of the followi
ng chara
c
te
risti
c
s:
a.
Young G
r
ou
p
:
15-25 yea
r
s
old
Numb
er of m
a
le data sam
p
les
: 15 people
Numb
er of fe
male data sa
mples
: 15 people
b.
Adult Grou
p: 25-4
5
years o
l
d
Numb
er of m
a
le data sam
p
les
: 10 people
Numb
er of fe
male data sa
mples
: 10 people
Selection
of the ab
ove ag
e gro
u
p
s
is
consi
der
ed b
a
s
ed
on the fa
ct that the b
r
oad
ca
st pro
g
ram
contai
ns infri
ngeme
n
ts mo
stly spo
k
en b
y
the age gro
up.
Of the 50
d
a
ta sa
mple
s,
som
e
come
from diffe
re
nt tribe
s
an
d
are
a
s in In
done
sia,
becau
se e
a
ch tribe
or
reg
i
on ha
s it
s o
w
n a
c
cent
i
n
uttering
a
syllable, word
or p
h
ra
se
in
the
Indone
sia
n
la
ngua
ge. M
e
a
n
whil
e, the
et
hnicity a
n
d
re
gion
s of
data
sampl
e
s in
th
e data
retriev
a
l
words a
r
e Ja
vanese, Sundane
se, Beta
wi, Sumatra, Medan, Sum
a
tra Padan
g,
Sumatra Batak,
Jambi, Ri
au, Bali and Sula
we
si.
Each
pa
rticip
ant in
coll
ect
i
ng d
a
ta p
r
o
noun
ce
d 15
7
2
sylla
ble
s
a
nd 1
658
wo
rds; the
words uttered
re
pre
s
e
n
t all
existing
sylla
bles,
wh
ere
syllabific
a
tion refers the Great Dic
t
ionary
of
the Indonesi
a
n Language (
KBBI). Spoken syllabl
e has 11 spec
ies consi
s
ting
of Singing
(V) and
con
s
o
nant
s (K). The type
s an
d their
1
1
numb
e
r
of syllable
s
a
r
e
V (5), VK (6
1
)
, KV (97
)
, KVK
(919), KKV (56), KKVK (250),
VKK (10), KKVKK (44), KKKV (5),
KKKVK (11), and KVKK (114).
Testing
for
sp
eech recognit
i
on is
pe
rform
ed
on
21
existing wo
rd
s; th
e sa
mple
dat
a used
are 10
50 dat
a, and the da
ta are divided
into 60%
training data an
d 40% testin
g data. This test
use
s
th
e M
F
CC coefficie
n
t
8, an
d 1
4
in
the H
MM stat
e.
Experim
en
ts
a
r
e ca
rri
ed out
by cha
ngi
ng
the voice dat
a use
d
in the training p
r
o
c
e
ss.
Ta
ble 1 shows the exp
e
rime
ntal re
sults.
Table 1. Experime
n
ts Result
Training Data
Experiments
Degree o
f
Accuracy (%
)
Man Speech
Woman Speech
Man Speech
1
80.95
23.8
2
85.71
19.05
Average
83.33
21.42
Woman Speech
1
23.8
80.95
2
19.05
71.42
Average
21.45
76.18
Man and Woman
Speech
1
76.19
61.90
2
76.19
76.19
Average
76.19
69.04
Table
1 sho
w
s th
at the
gend
er u
s
e
d
as trai
nin
g
data have
a
great i
n
fluen
ce o
n
the
results
of the
accu
ra
cy of
the
sy
stem.
The te
sts
usi
ng trai
ning
d
a
ta containi
n
g
both
male
and
female uttera
nce
s
can p
r
o
duce a level
of accura
cy t
hat is rel
a
tively compa
r
abl
e betwe
en te
sting
with mal
e
vo
ice
data
and
female voi
c
e data.
Thu
s
in the
proce
s
s of d
e
tecti
on of
bro
a
d
c
ast
prog
ram inf
r
i
ngeme
n
ts, training data t
hat contai
n the voice
s
of both men an
d wome
n will
be
use
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
1693-6
930
Integrated S
y
stem
Design for Broadcast
Progr
am
Infringem
ent .... (
S
ukm
a
wati Nur Endah)
577
4. Conclu
sion
The desi
gn result
s
of
th
e broa
dcast progra
m
infri
n
g
e
ment d
e
tecti
on
system
ha
ve bee
n
built in the fo
rm of a
r
chite
c
tural
de
sign
and t
he main
modul
e
interface
d
e
si
gn. This de
sign h
a
s
been te
sted
for the spee
ch recognitio
n
testing p
h
a
s
e. Implem
e
n
tation of th
e system
will
be
develop
ed u
s
ing trainin
g
d
a
ta that conta
i
n men and
women from yo
ung an
d adult
group
s.
Referen
ces
[1]
Send
jaj
a
SD (
C
hief of Komis
i
Pen
y
iar
an Ind
ones
ia Pus
a
t). Reg
u
lati
on of
Komisi Pe
n
y
iar
an Ind
ones
ia
No 02/P/KPI/12/200
9 ab
out
T
he Co
de of Bro
adcasti
ng C
o
n
duct. Jakarta. 200
9.
[2]
Send
jaj
a
SD (
C
hief of Komis
i
Pen
y
iar
an Ind
ones
ia Pus
a
t). Reg
u
lati
on of
Komisi Pe
n
y
iar
an Ind
ones
ia
No 03/P/KPI/12/200
9 ab
out T
he Broa
dcasti
n
g
Program Sta
ndar
d. Jakarta.
2009.
[3]
T
j
ah
y
a
nto A,
Supr
apto
Y
K
, W
u
lan
dari
DP. Sp
ectra
l
-bas
ed F
e
atu
r
es R
anki
n
g
for Gamel
a
n
Instruments Id
entificati
on
us
i
ng F
ilter T
e
chniq
ues.
T
E
LK
OMNIKA T
e
lecommunic
a
tio
n
Co
mputi
n
g
Electron
ics an
d Contro
l
. 201
3; 11(1): 95
–10
6.
[4]
Jadhav A, Patil A. A Smart T
e
x
t
i
ng S
y
stem for Andro
i
d Mobi
le Us
er
s.
Internation
a
l
Journ
a
l
of
Engi
neer
in
g R
e
searc
h
an
d Applic
atio
ns
. 20
12; 2(2): 11
26-
112
8.
[5]
Jadh
av A, Pa
til A. Andro
i
d
Speec
h to
T
e
xt C
onv
erter
for SMS Ap
plicati
on.
IOSR Jour
nal
of
Engi
neer
in
g
. 2012; 2(3): 4
20-
423.
[6]
Sharma F
R
,
W
a
sson SG.
Speec
h R
e
co
gniti
on
an
d S
y
nt
hetis T
ool:
Assistive T
e
chno
log
y
f
o
r
Ph
y
s
ica
l
l
y
Dis
abl
ed Perso
n
s
.
Internation
a
l
Journ
a
l of Co
mp
ut
er Scie
nc
e and T
e
l
e
co
mmu
n
icati
ons
.
201
2; 3(4): 86-
91.
[7]
Patel I, R
ao Y
S
. Speec
h R
e
c
ogn
ition
us
ing
HMM
w
i
th
MF
CC-AN A
n
a
l
ysi
s
usi
ng F
r
e
q
u
e
nc
y Sp
ectral
Decom
posi
on T
e
chnique
. Si
g
nal & Image Pr
ocessi
ng
:
An Internati
o
n
a
l Jo
urna
l.
201
0; 1(2):
101-1
10.
[8]
Anusu
y
a
MA, Katti SK. Speec
h Rec
o
g
n
itio
n b
y
Mac
h
in
e: A Revi
e
w
.
International
Jo
urna
l of
Co
mp
uter Scie
nce an
d Infor
m
ation Sec
u
rity (IJCSIS)
. 2009; 6(3): 181
–2
05.
[9]
Kumar K, Aggar
w
a
l RK. Hi
n
d
i Spe
e
ch Re
cogn
ition S
y
st
em Using HT
K.
Internationa
l Journa
l of
Co
mp
uting a
n
d
Business R
e
s
earch
. 20
11; 2(
2): 2229-
61
66.
[10]
Adam T
B
, Salam MS, G
una
w
a
n T
S
, W
a
velet
C
eps
tral Co
efficien
ts for Isolate
d
Spe
e
ch.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol
. 20
13; 11(5): 2
731
–27
38.
[11]
W
ang Y, Ya
ng
X, Z
ou J. R
e
se
ar
ch of Emoti
o
n Rec
ogn
itio
n
Based
on S
p
e
e
ch a
nd F
a
c
i
al
Expr
essi
on.
T
E
LKOMNIKA T
e
leco
mmunic
a
tion C
o
mputi
n
g Electron
ics a
nd Co
ntrol
. 20
13; 11(1): 8
3–9
0.
[12]
Sing
h B, Kap
u
r
N, Kaur
P. S
peec
h R
e
cog
n
i
t
ion
w
i
t
h
Hi
dd
e
n
Markov M
o
d
e
l: A Rev
i
e
w
.
In
te
rn
a
t
io
na
l
Journ
a
l of Adv
ance
d
Res
earc
h
in Co
mputer
Scienc
e an
d Softw
are Engin
e
e
rin
g
.
201
2; 2(3): 400-4
03.
[13]
Enda
h SN,
N
ugra
hen
i DMK
.
Klasifik
asi
U
c
apa
n Kat
a
d
eng
an S
u
p
por
t Vector Mac
h
in
e.
Jurnal
Masyarak
a
t
Informatika
. 20
12
; 3(6): 7-14.
[14]
Enda
h SN, N
u
grah
eni
DMK.
Anal
isis Pitch
dan F
o
r
m
ant
Sinya
l
Uca
p
a
n
Kata
. Prosi
d
i
ng Sem
i
n
a
r
Nasio
n
a
l
Ilmu Komputer U
n
iv
ersitas Di
pon
e
goro 2
0
1
2
. Semaran
g
. 201
2; 101
–1
07.
[15]
O’Shau
gh
ness
y
D. Invite
d p
a
per: Autom
a
tic
Spe
e
ch
Rec
o
gniti
on: H
i
stor
y, Methods
a
n
d
Ch
all
eng
es.
Pattern Reco
g
n
itio
n Scienc
e
Direct.
200
8; 41: 2965-
29
79.
Evaluation Warning : The document was created with Spire.PDF for Python.