Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
10
,
No.
1
,
A
pr
il
201
8
, p
p.
176
~
1
8
3
IS
S
N:
25
02
-
4752
, DO
I: 10
.11
591/
ijeecs
.
v
10
.i
1
.pp
176
-
1
8
3
176
Journ
al h
om
e
page
:
http:
//
ia
es
core.c
om/j
ourn
als/i
ndex.
ph
p/ij
eecs
Compari
son of E
ntrop
y Coding
mech
anism on IE
EE1857.
2
Lossless
Au
dio Comp
ressi
on Stand
ar
d
Fathiah
Abdu
l Muin
1
,
Te
dd
y
S
urya
Gun
aw
an
2
,
Mira
K
art
iw
i
3
, Elshei
kh M.
A.
El
she
ikh
4
1
,2,4
Depa
rtment
of
Elec
tr
ical and
Com
pute
r
Eng
i
nee
ring
,
Kul
liyyah
of
Engi
n
ee
rin
g
,
Ma
l
a
y
s
ia
3
Depa
rtment of
Inform
at
ion
S
y
s
t
ems
,
Kulliyy
ah
of
Inform
at
ion
a
nd
Com
m
unic
at
i
on
Technol
og
y
Inte
rna
ti
ona
l
Isl
a
m
ic
Univer
sit
y
Malay
s
ia,
Ja
la
n
Gom
bak,
Kuala
Lumpur,
Sel
ang
or,
Mal
a
y
s
ia
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Ja
n
7
, 201
8
Re
vised
Ma
r
9
,
201
8
Accepte
d
Ma
r
25
, 201
8
Thi
s
pape
r
has
two
objecti
v
es.
First,
we
a
im
to
re
vi
ew
and
ana
l
y
z
e
the
per
form
anc
e
of
the
IE
EE
1857
.
2
standa
rd
,
foc
u
sing
is
on
the
Golom
b
-
Ric
e
and
Arithmetic
e
ntrop
y
al
gor
it
hm
s
as
well
as
the
eff
ect
of
the
p
re
-
proc
essing
bloc
k
on
the
se
e
ntrop
y
blo
cks.
T
he
pre
-
proc
essing
bloc
k
norm
al
izes
the
err
o
r
re
sidue
of
the
L
ine
ar
Predi
ct
iv
e
enc
oder
,
which
the
n
is
passed
to
Ent
rop
y
bloc
k,
wher
e
th
e
select
or
choose
s
the
en
t
rop
y
e
ncode
r
to
use
.
The
sec
on
d
obje
c
ti
ve
is
to
pre
sent
re
sult
s
from
expe
ri
m
ent
ing
diffe
re
nt
exi
sting
al
gorit
hm
s
avai
la
bl
e
to
b
enc
h
m
ark
it
’.
The
re
sults
are
disc
uss
ed,
and
compari
sons
are
m
ade
to
id
e
nti
f
y
the
eff
e
ct
on
compress
ion
ra
ti
o
and
enc
oding
spe
ed
of
the
lossless
e
ncode
r.
As
well
as
thi
s,
compari
s
on
is
m
ade
to
ana
l
y
z
e
eff
ec
t
s
of
ena
bli
ng
an
d
disabl
ing
the
pre
-
proc
essing
o
utput
to
the
Ent
rop
y
Coding
bloc
k.
W
e
conclude
d
th
at
pre
-
p
roc
essing
bloc
k
works
well
to
fla
tten
the
o
utput
at
lower
pre
dictor
orde
r
for
al
l
the
sound
t
y
pes,
but
works
best
a
t
im
proving
th
e
r
esid
ual
ou
tput
fo
r
m
usic
sound t
y
p
e.
Ke
yw
or
d
s
:
Arithmeti
c
entro
p
y
al
gor
it
hm
s
Ent
rop
y
c
oding
Go
l
om
b
-
Ri
ce Cod
i
ng
Copyright
©
201
8
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Ted
dy S
ur
ya
G
un
a
wa
n
,
Dep
a
rtm
ent
of
Ele
ct
rical
an
d
Com
pu
te
r
E
ng
i
neer
i
ng, Kulli
yy
ah
of E
nginee
rin
g
,
In
te
r
natio
nal Is
lam
ic
U
niv
er
sit
y M
al
ay
sia
,
Jal
an Go
m
bak
,
Kuala
L
um
pu
r
, S
el
an
gor
, Ma
la
ysi
a
.
Em
a
il
:
tsgu
na
wan@ii
um
.ed
u.m
y
1.
INTROD
U
CTION
Mov
i
ng
forw
a
r
d
f
ro
m
the
firs
t
inn
ovat
io
n
of
aud
i
o
com
pr
e
ssion,
the
dem
and
for
hi
gh
e
r
qu
al
it
y
data
has rise
n
t
hro
ught
va
rio
us
f
ie
l
ds
li
ke
m
edical
inno
vations s
uc
h
as
ECG a
na
ly
sis or
heat b
e
at
an
al
ysi
s,
in t
erm
s
of
a
udio
data
s
tora
ge
an
d
a
udio
stream
ing
,
thi
s
ine
vitably
le
ads
to
a
dem
a
nd
of
lossless
aud
i
o
com
pr
es
sion
a
s
gen
e
rall
y
hig
h
-
qual
it
y
data
req
ui
res
not
on
l
y
hig
h
data
sto
rag
e
,
but
al
so
high
data
rate
for
transm
issi
on
a
nd
analy
sis
[1]
.
T
he
idea
be
hind
lossless
au
dio
com
pr
essio
n
is
to
rem
ov
e
re
dunda
nt
bits
of
data
with
out
any
los
s
of
qual
it
y,
by
us
in
g
an
e
ncoder
c
on
ta
ini
ng
a
pr
edict
or
w
hich
can
pr
e
di
ct
the
sign
al
as
cl
os
e
as
po
ss
ible
an
d
cal
culat
ing
the
error
of
this
predict
ed
sig
nal.
Thi
s
pre
dicti
on
resid
ue
is
fi
na
ll
y
enco
de
d
with
both
it
s
pre
di
ct
or
and
va
rio
us
ot
her
neces
sary
par
am
et
ers.
Wh
en
the
us
e
r
re
cei
ves
this
enc
od
e
d
data,
it
use
s
a
dec
od
e
r
t
hat
ca
n
reconstr
uct
the
or
igi
nal
sig
na
l.
On
t
op
of
t
hi
s,
the
enc
oder
will
encode
the
resi
du
al
e
rror
us
in
g
a
n
en
tro
py
cod
i
ng,
w
hich
is
a
par
t
of
inf
or
m
at
ion
theory
to
fu
rthe
r
re
duce
the
num
ber
of
re
dunda
nt
bits
[2]
.
The
s
m
al
le
r
the
resid
ual
er
r
or
gi
ven
fro
m
the
pr
edict
or,
the
m
or
e
ef
fici
ent
the
ent
ropy
co
ding
is
and
the
sm
al
l
er
the
com
pr
essed
sig
nal as it
will
contai
n
m
or
e ze
r
oes
i
n
it
s erro
r resi
dual
.
Ov
e
rall
,
the
pr
e
dictor
an
d
e
ntr
opy
co
di
ng
are
e
ssent
ia
l
par
t
s
of
any
lossless
com
pr
essio
n
m
ech
anism
.
In
entropy
co
di
ng,
a
bitst
rea
m
m
a
y
be
co
m
pr
essed
by
r
e
m
ov
in
g
re
du
nd
a
nt
bits
th
r
ough
a
gen
e
rali
zed
al
gorithm
.
In
t
t
he
IEEE
1857.
2
lossless
au
di
o
com
pr
essio
n
sta
nd
a
r
d,
bo
th
Go
l
om
b
-
Ri
ce
and
arit
hm
etic
cod
ing
is
m
echan
ism
is
d
efined
,
t
hu
s
i
n
this
pa
pe
r,
we
wish
to
com
par
e
the
pe
rfor
m
ance
of
these
com
pr
esso
r wit
h resp
ect
t
o
t
he
I
EEE
1857.
2
predict
or al
gorithm
an
d
the
o
t
he
r
al
gorithm
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
son
of
En
tr
op
y
Co
di
ng Mec
han
ism
on I
E
EE1
857.
2
L
os
sle
ss
Aud
io
... (
Fath
i
ah A
bdul M
uin
)
177
In
m
os
t
stu
dies
an
d
descr
i
ptio
n
of
l
os
sle
ss
a
udio
c
om
pr
essio
n,
t
he
f
oc
us
is m
or
e
on
predic
ti
on
b
loc
k,
howe
ver,
entr
opy
co
din
g
is
j
ust
as
i
m
po
rtant
in
this
case
as
it
us
ed
as
a
tool
to
wr
it
e
the
inten
ded
c
om
pr
esse
d
bitst
ream
thro
ugh
it
s
al
gorith
m
.
Gen
erall
y,
the
pur
po
se
of
entr
op
y
c
odin
g
is
not
li
m
it
ed
to
au
dio
pe
r
sa
y
but
as
a
unive
rsal
co
ding
m
echan
ism
to
com
pr
ess
bitst
rea
m
da
ta
.
Nev
er
thele
ss,
th
rou
gh
this
pa
per
we
will
com
par
e
two
entr
op
y
c
oding
m
echan
is
m
sp
eci
fical
ly
desig
ne
d
f
or
the
I
EEE
1857.
2
los
sle
ss
aud
i
o
com
pr
essio
n
a
nd
t
o
fi
nd
w
hi
ch
unive
rsal
entr
op
y
c
od
i
ng
m
ay
be
m
or
e
su
it
ed
f
or
it
s
ap
plica
ti
on
.
In
t
he
pr
e
vious
pa
pe
r
,
we
ha
d
evalu
at
ed
the
perfor
m
ance
of
the
I
EEE1
857.2
Li
near
P
red
ic
to
r
Cod
i
ng
(
LPC)
blo
c
k
and
fou
nd
inte
resti
ng
relat
ionships
of
the
L
PC
blo
c
k
with
the
pr
e
processi
ng
blo
c
k
as
we
ll
,
thu
s
this
stu
dy
of
the
prep
r
ocessi
ng
blo
c
k
is
al
s
o
exte
nded
i
nto
this
pap
e
r
to
it
’s
eff
ect
on
t
he
di
ff
e
ren
t
e
nt
ropy
m
echan
ism
of
the I
E
EE1
85
7.2 s
ta
nd
a
rd
[
3]
.
In
the
rest
of
this
pap
e
r,
f
or
sect
ion
2,
we
will
discu
ss
G
olo
m
b
-
Ri
ce
Co
ding
m
e
chan
ism
a
nd
su
bse
que
ntly
Ar
it
hm
et
ic
Cod
in
g
in
sect
io
n
3.
T
hen,
in
se
ct
ion
4,
we
will
div
ide
int
o
4
su
bse
ct
io
ns,
w
ere
we
will
detai
l
com
po
nen
ts
of
t
he
IE
EE
1857.
2,
by
re
visit
ing
the
pr
e
-
proc
essing
blo
c
k
c
ocep
t,
entr
opy
sel
ect
ion,
Go
l
om
b
-
Ri
ce
encodin
g,
a
nd
arit
hm
e
ti
c
enco
di
ng
al
gorithm
s
us
ed.
F
ollow
in
g
that,
we
will
descr
i
be
the
exp
e
rim
ental
set
up
a
nd
m
easur
em
ent
i
n
sect
ion
5,
as
well
as
the
resu
lt
s
an
d
discuss
i
on
of
this
exp
e
rim
entat
io
n
in
secti
on
6.
The
n
fi
nally
conclu
de
the
pap
er in sect
ion
7.
2.
GOLO
MB RI
CE COD
IN
G
Go
l
om
b
-
Ri
ce
Cod
i
ng
is
wide
ly
us
ed
f
or
va
rio
us
L
os
sle
s
s
A
udio
C
ompressi
on
popula
r
to
ols
a
nd
sta
nd
a
rds
,
s
uc
h
as
FLA
C
a
nd
MPEG
-
4
ALS
[4
]
,
[
5]
.
The
reason
be
ing
is
t
hat
G
ol
om
b
-
Ri
ce
codi
ng
is
a
der
i
vation
of
Huff
m
an
Co
di
ng
w
hich
is
s
uited
f
or
ti
m
e
dep
en
da
nt
ap
plic
at
ion
s,
th
us
use
fu
l
for
im
pr
ovin
g
the
encodin
g
sp
ee
d,
but
with
the
ex
pense
of
c
om
pr
ession
rati
o
[6]
.
I
n
the
la
te
r
cha
pters
,
w
e
will
ve
rify
w
hethe
r
this is ap
plica
bl
e to los
sle
ss
a
ud
i
o
c
om
pr
ession ap
plica
ti
on
s
.
Firstl
y,
the
way
this
m
et
ho
d
is
execu
te
d
is
by
giv
ing
a
uniqu
e
par
am
et
er
m
;
Then
a
po
si
ti
ve
intege
r
n,
wh
ic
h
we
w
ish
to
e
nc
od
e
i
s
div
i
ded
i
nto
a
rem
ai
nd
er
of
an
d
qu
otient,
[7]
.
If
,
t
he
cod
e
w
ord
f
or
n
w
il
l
con
sist
s o
f
k
-
le
ast
sig
nificant b
it
s
of n
,
fo
ll
ow
e
d
by
th
e
num
ber
f
or
m
ed
by
the r
em
a
inin
g
m
os
t si
gn
ific
an
t bit
s of
n
in
un
ary re
pr
ese
ntat
ion
a
nd a
sto
p bit
[2
]
. T
he
le
ngth
of this c
od
e wor
d
is,
(
1
)
Durin
g
t
he
ear
li
er
inv
e
ntio
n
of
lossless
au
di
o
com
pr
essi
on,
t
he
est
im
ati
on
f
or
the
pa
r
a
m
et
er
k
is
giv
e
n
in
a
nd
is
us
e
d
in
A
udioPaK
a
nd
it
is
base
d
on
t
he
exp
e
ct
at
ion
al
read
y
com
pute
d
in
th
e
pr
e
dictor
b
l
oc
k w
her
e
,
(
2
)
The
par
am
et
er
k
is
de
fine
d
to
be
c
on
sta
nt
over
a
n
e
nt
ire
fr
am
e
and
ta
kes
value
s
betwee
n
0
and
(b
−
1)
i
n
the
case
of
b
bi
t
aud
io
sam
ples
[2]
.
T
his
co
nc
ept
is
sti
ll
wi
dely
us
ed
i
n
c
urren
t
e
xisti
ng
cod
ecs
and
sta
ndar
ds.
Nev
e
rtheless
,
i
n
te
rm
s
of
the
lossless
a
ud
i
o
com
pr
essio
n
blo
ck
,
it
is
im
portant
to
bea
r
i
n
m
ind
that
the
G
olo
m
b
co
des
def
i
ne
d
to
be
opti
m
a
l
fo
r
e
xpone
ntial
ly
decayi
ng
pro
bab
il
it
y
distrib
ution
s
of
po
sit
ive
integers
a
nd
be
cause
the
predi
ct
ion
resid
uals
m
ay
no
t
al
l
po
sit
ive,
it
m
ay
b
e
req
ui
red
to
m
ap
the
error
re
s
idu
al
to an u
ns
i
gn
e
d value a
s
def
in
e
d
in
the e
quat
ion bel
ow
[
8]
:
(
3
)
3.
AR
IT
HMETI
C CO
DING
Anothe
r
al
te
r
na
ti
ve
f
or
univ
ersal
enc
odin
g
m
echan
is
m
i
n
L
os
sle
ss
C
om
pr
essio
n
is
arit
hm
et
i
c
cod
i
ng.
Firstl
y,
in
a
nu
ts
hell,
arit
hm
etic
cod
ing
cal
c
ulate
s
the
pro
ba
bili
ti
e
s
of
occura
nce
of
eac
h
sym
b
ol
in
a
m
essage
over
a
finite
al
phab
het.
T
his
m
et
ho
d
does
s
o
by
inco
rpor
at
in
g
t
wo
vaa
riables,
L
an
d
R,
w
he
re
L
is
the
sm
a
ll
est
bi
nar
y
val
ue
co
ns
ist
ent
with
t
he
co
de
re
pr
e
s
enting
t
he
sym
bo
ls
fo
un
d
so
far,
w
hilst
R
is
the
pro
du
ct
of
pro
bab
il
it
ie
s
of
th
e
sy
m
bo
ls
fo
und.
The
sim
ple
m
echan
ism
of
arit
hm
e
ti
c
cod
ing
can
be
d
es
crib
e
d
as the
fo
ll
owin
g
se
quence
, wit
h
ste
ps 2
to 4
occurin
g recu
r
sively
as n
e
w s
ym
bo
ls i
s pr
oc
essed
[
9]
.
1)
In
it
ia
li
zat
ion
of
an
d
2)
Enc
od
e
n
e
xt sy
m
bo
l (
th
of
t
he
alph
a
bet)
b
y
3)
New
4)
Ou
t
pu
t
seq
ue
nc
e b
it
c, f
or
eac
h bit
seq
ue
nce
betwee
n
an
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
10
, N
o.
1
,
A
pr
il
201
8
:
176
–
1
8
3
178
Wh
e
re
a
nd
is
the
pro
bab
i
li
ty
of
the
it
h
(c
urren
t
)
a
nd
j
t
h
(
ne
xt)
sym
bo
l
in
the
a
lph
a
bet
resp
ect
ively
.
4.
IEE
E 185
7.2 ST
AND
ARD
4.1.
Preproces
sing
Bl
ock
As
m
entioned
in
the
pr
e
vi
ou
s
pap
e
r,
one
of
t
he
ad
va
ntages
of
the
IEEE
1857.2
lo
ssless
c
om
pr
essio
n
is
that
the
bit
error
int
rod
uced
durin
g
tran
sm
issi
on
ca
n
be
m
ini
m
iz
ed
by
the
pre
-
processi
ng
blo
c
k
[
3]
.
T
his
is
because
the
a
ud
i
o
f
ram
es
c
an
be
dec
od
e
d
at
one
f
ram
e
inter
val
sin
c
e
the
inf
or
m
at
ion
of
a
fr
am
e
are
ind
e
pende
nt
of
each
oth
er
.
H
ow
e
ve
r,
the
pr
edict
ion
resid
ue
s
will
be
bi
gger
com
par
e
d
to
the
rest
of
th
e
fr
am
e
con
te
nt,
res
ulti
ng
in
t
he
inc
rease
of
the
predict
io
n
resi
dues
dy
nam
ic
r
ang
e
.
T
o
com
pensat
e
for
thi
s,
the
e
ntr
op
y e
nc
od
e
r
the
n
i
ncr
ease
s both
the calc
ulati
on
c
om
plexity
an
d t
he
alph
a
bet size
.
The
pr
e
-
proce
sso
r
bl
ock
inte
nd
s
to
overc
om
e
this
sit
uation
by
ta
king
t
he
pr
e
dicti
on
r
esi
du
es
an
d
dow
ns
hi
ft
them
to
a
certai
n
degree.
T
his
operati
on
al
lows
the
a
m
plit
ud
e
to
decr
ease
and
the
en
vel
ope
of
th
e
pr
e
dicti
on
re
s
idu
es
is
flat
te
ned
by
adapt
ive
norm
aliz
at
ion
,
w
hich
was
al
so
pro
ved
to
im
pr
ove
the
com
pr
essio
n
r
at
e
on
MPE
G
4
-
AL
S
[
8]
.
The
resid
ual
s
a
m
ple
are
qu
a
ntize
d
int
o
po
wer
of
2
us
in
g
th
e
PA
RC
OR
c
oeffici
ent
k
to
m
ake
sure
that
it
is
integer
s
f
or
l
os
sle
ss
c
od
i
ng
as
well
norm
alized
by
do
wn
-
s
hifted
with
as foll
ow
ed
[10
]
,
[
1
1
]
:
(
4
)
(
5
)
The
te
rm
within
t
he
s
umm
a
tio
n
are
pr
e
-
co
m
pu
te
d
by
R
A
_s
hi
ft
a
nd
RA
_s
hi
ft1
2
fixe
d
ta
ble,
wh
ic
h
wer
e
provide
d by the
sta
ndar
d f
or
fast c
om
pu
ta
ti
on
and
de
vi
ce p
or
ta
bili
ty
.
4.2.
Entropy
Codi
ng
Sele
ctio
n
As
m
entioned
pr
e
viously
,
the
IEEE
1857.
2
s
ta
nd
a
rd
has
a
s
el
ect
ion
of
tw
o
ty
pes
of
e
nthr
op
y
c
odin
g,
wh
ic
h
are
Go
l
o
mb
-
Ri
ce
Codi
ng
a
nd
Ar
it
hm
et
ic
Cod
ing
and
it
util
iz
es
on
e
bit
in
th
e
ou
t
pu
t
bitst
rea
m
to
switc
h
betwee
n
the
tw
o
m
eth
ods
f
or
t
he
de
cod
e
r
t
o
rec
ognize
w
hich
m
et
ho
d
was
use
d
in
the
e
nc
od
i
ng
process
[1
2
]
.
This
is
il
lustra
te
d
in
the
F
ig
ure
1
bel
ow,
s
o
the
co
de
r
will
sel
ect
the
entr
op
y
ty
pe
t
o
use
an
d
encode
t
his
se
le
ct
ion
as
well
.
In
the
sta
ndar
d
it
m
entions
that
t
his
s
el
ect
ion
occ
ur
s
de
pendin
g
on
the
env
i
ronm
ent
of
the
us
er
s,
but
fo
r
sim
plici
t
y
in
this
stud
y,
we
ex
plici
tl
y
sel
ect
the
entropy
ty
pe
by
parsi
ng
an
a
rgum
ent to the
too
l.
Figure
1
.
Enc
oder
Co
der Sele
ct
ion
(abo
ve)
a
nd D
ec
oder
Co
der Sele
ct
ion (
belo
w)
=
1
1
−
2
=
+
1
=
log
2
1
1
−
2
=
+
1
+
0
.
5
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
son
of
En
tr
op
y
Co
di
ng Mec
han
ism
on I
E
EE1
857.
2
L
os
sle
ss
Aud
io
... (
Fath
i
ah A
bdul M
uin
)
179
4.3.
Adapt
i
ve Gol
omb
E
ncodi
n
g
Fo
r
t
he
G
olo
m
b
-
Ri
ce
E
ntryp
y
sel
ect
ion
,
the
sta
nd
a
rd
util
iz
es
an
ada
ptive
Go
l
om
b
-
Ri
ce
m
echan
ism
,
an
ad
van
ce
d
m
et
ho
d,
wh
e
re
by
un
li
ke
it
predecess
or
s
,
it
do
e
sn’t
on
ly
update
the
Ri
ce
par
am
et
er
(le
adze
r
o,
div
isi
on
an
d
re
m
ai
nd
er
bits)
,
bu
t
it
al
so
upda
te
s
the
m
val
ue
after
eac
h
bl
ock
is
proces
s
ed,
w
hic
h
is
fixed
i
n
the origi
nal Go
lom
b
-
Ric
e
m
eth
od.
Firstl
y,
in
the
encode
r,
t
he
L
PC
resid
ue
is
div
ide
d
i
nto
s
ubbl
ocks
of
,
w
her
e
can
be
se
le
ct
ed
t
o
be
bet
ween
1
and
up
t
o
5.
T
hen
i
niti
al
m
i
s
cal
culat
e
d
by
the
m
ean
(
)
of
the
L
PC
re
sidu
al
er
ror,
by
the
fo
ll
owin
g
e
qua
ti
on
,
(
6
)
w
he
n
eac
h
s
ubblo
c
k
is
enc
od
ed,
the
m
is
updated
base
d
on
the
pa
ram
et
er
RICE_
NU
M
_M
UL
=
32
a
fter
the
Go
l
om
b
-
Ri
ce
Enc
od
e
r.
Fig
ure
1
ab
ove,
il
lu
strat
es
this
up
dating
process
of
m
and
the
f
ollow
i
ng
e
quat
i
ons
descr
i
bes h
ow
m
is updated
e
ach tim
e b
y ea
ch
s
ubbl
ock,
−
−
−
(
7
)
(
8
)
(
9
)
The
s
um
in
th
is
case
i
s
t
he
Go
l
om
b
-
Ri
ce
cum
ulati
ve
sum
wh
ic
h
is
ca
lc
ulate
d
based
on
t
he
ea
c
h
su
bbl
ock resi
dual
cu
m
ulati
ve
su
m
that is p
r
ocesse
d
at
that
tim
e.
Figure
2
.
G
olom
b
-
Ri
ce Encoder Pr
ocess
4.4.
Arithme
tic En
codin
g
Nex
t,
f
or
the
Ar
it
hm
etic
Enc
od
i
ng
sel
ect
ion
of
IEE
E1
857.2
blo
c
k
it
is
al
so
s
i
m
i
la
r
in
it
’
s
su
bbl
ock
i
ng
of
the
LPC
residu
e
,
but
x
on
l
y
go
es
up
to
3.
Firstl
y,
in
this
blo
ck
,
the
m
ean
of
the
flat
te
ned
resid
ue
is c
ompu
te
d,
w
hich
t
hen is lo
gar
it
hm
ic
l
y qu
a
ntize
d
by the
f
ollowi
ng
e
quat
io
n,
(
10
)
A
fter
this
,
this
qu
at
iz
ed
m
ean
ind
e
x
is
local
ly
deq
ua
ntize
d
an
d
then
us
e
d
to
gen
e
rate
the
Pr
oba
ba
bili
ty
ta
ble
from
a
set
of
pr
ob
a
bi
li
ty
tem
plate
.
The
f
ollo
wing
equ
at
io
n
is
us
e
d
in
this
s
cal
ing
proce
dure,
(
11
)
The
pr
ob
a
bili
ty
tem
plate
is
e
ssentia
ll
y
a
set
of
prob
a
bili
ty
den
sit
y
value
s
from
a
trai
ned
aud
i
o
data,
wh
ic
h
is
a
ppr
ox
im
at
el
y
a
Gau
ssio
n
functi
on
(m
ean
-
0.1
an
d
sta
nda
rd
de
viati
on
0.6
),
bu
t
uniq
ue
to
the
IEEE
1857.
2
de
fine
d
sta
nda
r
d.
Pr
e
vious
w
ork
has
s
how
n
that
t
his
tra
ined
ta
ble
can
achieve
a
lo
ssless
com
pr
essio
n p
erfor
m
ace bett
er th
a
n
t
hat of t
he Gau
ssio
n functi
on, thus
thi
s table
us
ed
[
10]
.
Figure
3
desc
ribes
t
he
ov
e
rall
proces
s
of
t
he
Ar
it
hm
etic
enco
de
r.
Als
o
de
fine
d
in
this
process
is
th
e
MSB
/LSB
sp
li
t
blo
ck,
w
hic
h
is
an
adv
anta
ge
to
the
Go
l
om
b
-
rice
m
e
thod.
This
is
bec
ause
the
Ar
it
hm
eti
c
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
10
, N
o.
1
,
A
pr
il
201
8
:
176
–
1
8
3
180
Enc
od
e
r
can
f
urt
her
s
plit
the
LPC
resid
ue
to
MSB
and
LSB
par
ts
so
that
in
can
furthe
r
increase
the
ef
fi
ci
ency
of the
Ar
it
hm
et
ic
En
co
ding
A
lgorit
hm
, in
th
e case
of
t
rail
ing
bits.
Figure
1
.
A
rith
m
et
ic
Cod
ing
Enc
od
e
r
Bl
oc
k Diag
ram
5.
E
X
PERI
MEN
TAL SET
UP
AND ME
A
SUREME
NT
The
a
naly
sis
was
c
onduct
ed
in
a
c
om
bin
a
ti
on
of
MATL
AB
a
pp
li
cat
io
n
a
nd
(
.ex
e
)
file
com
piled
from
C
on
a
DELL
la
pt
op,
wh
ic
h
was
r
unning
with
I
ntel
cor
e
i7
proce
sso
r
to
m
easure
the
per
f
or
m
ance
of
each
t
oo
l.
Th
e
datab
ase
c
ons
ist
ed
of
A
ud
i
o
Bo
ok
s
w
hich
we
re
at
l
east
1
hour
lo
ng
w
it
h
a
c
om
bin
at
ion
of
En
glish
a
nd
A
rab
ic
book
rec
ordin
gs
.
Table
1
s
hows
a
sam
ple
of
the
A
udio
books
datab
ase
w
hich
were
use
d
in the i
nv
e
sti
ga
ti
on
.
Table
1
.
Sam
pl
e Aud
i
o
B
ook Data
base
Ty
p
e
Filen
a
m
e
Play
b
ack L
en
g
t
h
Eng
lish
Aud
io
Bo
o
k
ArtOfWa
r8k
.wav
1
:1
2
:1
8
Arabic Aud
io
Bo
o
k
Bo
o
k
(
A
m
in
),
20
1
6
1
2
2
0
.wav
1
:0
0
:4
3
Qu
r’
an
ic Reci
tatio
n
Al
-
Kah
f
(A
m
in
),
2
0
1
6
1
2
1
9
.wav
1
:0
2
:0
7
Fo
r
the
m
easurem
ents,
o
ne
of
the
basic
m
ea
su
rem
ent
for
a
lossless
au
dio
cod
ec
is
t
he
c
om
pr
essio
n
rati
o,
w
hich
is
m
easur
ed
by
the
com
par
iso
n
of
the
outp
ut
file
to
the
ori
gin
al
raw
file
with
the
f
ollo
wing
form
ula:
(
12
)
This
c
om
pr
essibil
it
y
determ
in
ed
th
e p
r
oport
ion
al
it
y
of b
it
s
in w
hic
h
wer
e
c
om
pr
essed
,
the
la
rg
e
r
t
hi
s
rati
o,
the
m
ore
bits
wer
e
c
om
pr
essed
[
5].
As
well
as
thi
s,
it
is
al
so
i
m
po
rtant
to
ensure
that
the
rate
of
com
pr
essin
g
is
eff
ic
ie
nt
,
th
us
enc
od
i
ng
s
pee
d
will
al
so
nee
d
to
be
m
easur
ed.
The
enc
oding
/
decodin
g
s
peeds
are
m
easur
ed
in
te
rm
s
of
how
fast
the
enc
od
e
r
proce
sses
relat
ive
to
the
total
le
ng
th
of
the
Audio
file
in
MB
.
This is
def
i
ned b
y t
he
foll
owing eq
uatio
ns
:
(
13
)
6.
RESU
LT
S
AND A
N
ALY
S
IS
In
t
his
ex
pe
rim
ent,
the
inte
r
nal
set
ti
ng
s
of
the
IEE
E1
857.
2
e
ntropy
to
ol
was
c
om
par
ed
by
ex
plici
tl
y
add
i
ng
a
par
se
d
op
ti
on
t
o
t
he
too
l
sel
ect
ing
ei
ther
Go
l
om
b
Ri
ce
(G
R)
or
Ar
it
hm
et
ic
Cod
in
g
(A
C
),
as
well
as
enab
li
ng
a
nd
disablin
g
of
t
he
Pr
e
processi
ng
f
or
each
Ent
ropy
Co
der.
A
dd
it
io
nally
,
it
is
al
so
be
nc
hm
ark
ed
with
the
la
te
st
FLA
C
a
nd
MP
EG
-
ALS
with
var
yi
ng
pr
e
dictor
order
t
o
fin
d
ho
w
well
ea
ch
al
gorithm
wor
ks
=
=
(
)
/
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
son
of
En
tr
op
y
Co
di
ng Mec
han
ism
on I
E
EE1
857.
2
L
os
sle
ss
Aud
io
... (
Fath
i
ah A
bdul M
uin
)
181
with
hi
gh
e
r
pr
eci
sion
.
F
or
a
fair
com
par
iso
n,
the
to
ol
wa
s
recreated
i
n
C
cod
e
on
t
op
of
A
VS
Ch
i
na
Cod
ec
and v
e
rified
b
y
co
m
par
in
g
the
outp
ut f
il
e
of the
decode
r
to
the
or
i
gin
al
f
il
e.
The follo
wing
are the
opti
ons
with it
s
descr
i
pt
ion
wh
ic
h
a
re
par
se
d
t
o
eac
h of t
he
t
oo
l:
FLA
C:
o
f: F
or
ce
over
w
rite
if output
file
n
am
e exists
o
l <
m
ax_
pre
dictor
_or
der
>:
Set
m
ax
predict
or
order o
f fixe
d m
od
el
MPEG
-
4 AL
S:
o
n<fram
e_size>
: Set
f
ram
e size
o
o<pre
dictor
_or
der
>:
Set p
red
i
ct
or
orde
r
IEEE 1
857.2:
o
f<1
>:
A
ud
i
o
St
or
a
ge Fi
le
outp
ut for
m
at
o
p<pre
dictor
_or
der
>:
set
predi
ct
or
orde
r
Fr
om
the
G
raph
in
F
i
gure
5,
we
can
see
tha
t
the
pr
e
proces
sing
blo
c
k
do
e
s
i
m
pr
ove
the
perform
anc
e
of
Entr
opy
co
ding
f
or
the
G
olo
m
b
-
Ri
ce
Bl
ock
with
hi
gh
e
r
pre
dictor
ord
er
by
at
le
ast
1.
29%
f
ro
m
pr
e
dicto
r
order
14.
This
i
m
pr
ovem
ent
i
ncr
ease
s
with
pr
e
dictor
ord
er
value
.
H
owev
er,
the
s
am
e
ca
n’
t
be
c
on
cl
uded
f
or
Ar
it
hm
et
ic
Cod
in
g,
a
s
the
re
i
s
no
cl
ea
r
patte
rn
on
w
hethe
r
t
he
pr
e
processi
ng
m
et
ho
d
im
pr
oves
t
he
Ar
it
hm
et
ic
Cod
i
ng or
no
t.
In
te
rm
s
of
Com
pr
essio
n
rati
o,
the
prep
rocessi
ng
blo
c
k
do
e
s
i
m
pr
oves
the
com
pr
ession
rati
o
by
at
le
at
0.
37%
f
or
th
e
G
olo
m
b
-
Ri
ce
Cod
i
ng
i
n
F
ig
ur
e
4,
ho
wev
e
r
ve
ry
ins
ign
ific
a
nt
i
m
prov
em
ent
is
seen
with
the Arit
hm
et
ic
blo
c
k
as
w
el
l a
s this the
p
e
rfo
rm
ance so
m
eho
w
worse
n at
predict
or
order
16 and
20.
Com
par
ing
t
he
value
s
of
t
he
enc
od
i
ng
s
pe
ed
a
nd
com
pr
essio
n
rati
o
f
or
A
rithm
etic
Cod
i
ng
an
d
Go
l
om
b
-
Ri
ce
Cod
i
ng
it
sel
f,
the
G
olo
m
b
-
R
ic
e
m
et
ho
d
ha
s
a
m
uch
faste
r
enc
odin
g
s
pe
ed
c
om
par
ed
t
o
th
e
Ar
it
hm
et
ic
Co
ding
m
echan
ism
,
wh
ere
as
th
e
Ar
it
hm
et
ic
C
od
i
ng
has
bett
er
com
pr
essio
n
rati
o
ove
rall
to
the
Go
l
om
b
-
Ri
ce
Cod
i
ng
f
or
los
s
le
ss
aud
io
c
om
pr
ession
ap
pl
ic
at
ion
.
This
is
con
sist
ent
to
the
pr
evi
ous
wor
k
wh
ic
h
c
om
par
es the t
wo m
et
h
od
s
for i
m
age co
m
pr
essio
n
a
pp
li
cat
io
n.
In
a
ddit
ion
t
o
t
his
com
par
iso
n,
a
nother
inte
resti
ng
fin
ding
was
fou
nd
w
he
n
c
om
par
ing
t
he
database
to
the
ot
her
Al
g
ori
thm
s.
As
s
een
in
the
F
ig
ure
6,
des
pite
pr
evio
us
w
ork
[1
0]
,
the
MPE
G
-
ALS
has
the
t
he
best
com
pr
essio
n
ra
ti
o,
w
he
re
as
t
he
F
LAC
faste
st
in
te
rm
s
of
e
ncodin
g
sp
ee
d.
O
ver
al
l
FL
AC
an
d
MPE
G
-
4
ALS
al
gorithm
s
su
rp
asses
the
I
E
EE1
857.2
co
m
pr
ession
rati
o,
as
well
as
the
encodin
g
sp
eed
as
we
ll
.
One
po
s
sibil
it
y
fo
r
this
occurre
nc
e
cou
ld
be
relat
ed
to
the
fact
that
the
data
base
us
e
d
are
al
l
Au
dio
books
file
s,
wh
e
re
as
the
previ
ou
s
wor
k
wer
e
al
l
m
us
ic
file
s,
with
sho
rter
le
ng
t
h.
T
his
exp
la
ins
wh
y
MPEG
-
4
AL
S
is
the
best
as
the
sta
nd
a
r
d
us
es
lo
ng
te
rm
LPC
residu
e
pre
dicti
on
w
hich
is
m
or
e
su
it
ed
f
or
s
peech
file
s,
w
her
e
as
none
of
the
re
oth
e
r
al
gorith
m
s
con
ta
in
thi
s
[1
3
]
.
The
F
L
AC
com
es
in
a
cl
os
e
seco
nd,
bu
t
it
al
so
co
ntains
a
n
aud
i
o
detect
io
n
w
hich
s
witc
hes
be
t
ween
diff
e
ren
t
ty
pes
of
sp
eec
h
m
od
el
li
ng
oth
e
r
tha
n
LPC,
th
us
al
lowi
ng
bette
r
er
ror
re
sidu
e
c
om
par
e
d
to
IEEE
1857.
2.
Additi
on
a
ll
y,
fr
om
the
FLA
C
c
od
e
,
it
us
es
it
s
ow
n
sysc
al
l
PO
S
IX
inter
fa
ce
f
or
rea
d/wr
i
te
protocal,
w
hi
ch
al
lows
fa
ste
r
rea
d
a
nd
wr
i
te
sp
ee
d
to
the
stdli
b
inter
face
.
T
his
cou
l
d be the
r
e
aso
n beh
i
nd it
’s
sig
nificant
di
ff
e
ren
ce
in
e
nc
od
i
ng s
peed to
the o
t
her
s
[
4]
.
Figure
2
.
Com
par
is
on of
Ar
it
hm
etic Cod
in
g (left)
and
G
olo
m
b
Ri
ce Cod
i
ng (rig
ht)
C
ompressi
on Rat
io
by
Ena
bling an
d Disabli
ng
Pr
e
proces
sin
g
Bl
oc
k
1.5
1.52
1.54
1.56
1.58
1.6
1.62
2
4
6
8
10
12
14
16
18
20
C_I
EAC_PR
E
C_I
EAC
1.34
1.36
1.38
1.4
1.42
1.44
1.46
2
4
6
8
10
12
14
16
18
20
C_I
EGC_
P
RE
C_I
EGC
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2502
-
4752
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci,
Vol
.
10
, N
o.
1
,
A
pr
il
201
8
:
176
–
1
8
3
182
Figure
3
.
Com
par
is
on of
Ar
it
hm
etic Cod
in
g (left)
and
G
olo
m
b
Ri
ce Cod
i
ng (rig
ht)
E
nc
odin
g
S
pee
d by
Ena
bling an
d Disabli
ng
Pr
e
proces
sin
g
Bl
oc
k
Figure
4
.
Com
par
is
on of C
om
pr
ession
Ra
ti
o (left) a
nd E
nc
od
i
ng S
peed
(
rig
ht) for
Var
i
ou
s
A
l
gorithm
s
7.
CONCL
US
I
O
N
Ov
e
rall
,
the
e
xp
e
rim
ent
sh
ows
that
f
or
lo
ssless
au
dio
c
om
pr
essio
n,
th
e
arit
hem
etic
cod
i
ng
has
a
higher
c
om
pr
ession
rati
o
to
golom
b
rice
codi
ng
in
the
IEE
E1
857.2,
wit
h
the
exp
e
ns
e
of
encodin
g
sp
ee
d
an
d
the
pr
e
-
pr
oces
sing
bl
oc
k
does
i
m
pr
ov
e
th
e
com
pr
ession
rati
o
of
entr
opy
cod
i
ng
es
pe
ci
al
ly
Go
lom
b
-
Ri
c
e
cod
i
ng,
bu
t
m
ay
worse
it
’s
encodin
g
s
pee
d
f
or
hi
gh
e
r
predict
or
order.
Nev
e
rtheless
,
wh
e
n
c
om
pari
ng
it
to
oth
e
r
m
et
ho
ds,
it
sti
l
l
fall
s
beh
in
d
to
MPEG
-
4
an
d
FL
AC
in
te
rm
s
of
co
m
pr
ession
rati
o
an
d
encodin
g
sp
ee
d
especial
ly
in
th
e
case
of
au
dio
books
.
Ultim
a
te
ly
,
IEEE
1857.
2
m
ay
be
m
or
e
s
uited
for
m
us
ic
file
s,
but
f
ur
t
her
i
m
pr
ovem
ent
c
ou
l
d
be
done
by
detect
ion
of
sp
eec
h
file
s
and
lo
ng
-
te
rm
pr
e
dicti
on
m
ec
han
ism
.
In
te
r
m
s
of
sp
ee
d,
the
I
EE
E1
857.2
an
d
MPEG
-
4
ALS
cou
l
d
ha
ve
a
bette
r
com
par
ison
with
the
F
LAC
co
dec
by
us
in
g
FLA
C sy
scal
l
PO
S
IX inter
fa
ce rea
d/wr
it
e i
nterf
ace
.
ACKN
OWLE
DGE
MENT
The
aut
hors
ha
ve
gr
at
e
fu
ll
y
ackno
wled
ged
that
this
research
ha
s
bee
n
su
pport
ed
by
Mi
nistry
of
Higher
Educat
i
on Mal
ay
sia
Resea
rch F
und,
FRGS
15
-
194
-
0435.
REFERE
NCE
S
[1]
Mori
y
a
Rese
arch L
ab
,
“
Lossless Com
pre
ss
ion
of
Audio
Dat
a
(M
PEG
-
4
ALS a
d
i
ts
Applicati
on)
,”
200
3.
[2]
Hans
M
.
and
Sc
haf
er
R
.
W.
,
“
L
oss
le
ss
compress
ion
of
digita
l
a
udio
,”
IEEE
S
ig
nal
Proc
ess
Mag
,
vol
.
18
,
pp
.
21
–
32
,
2001
.
6
6.2
6.4
6.6
6.8
7
7.2
7.4
2
4
6
8
10
12
14
16
18
20
T_I
EAC_PR
E
T_I
EAC
7.5
8
8.5
9
9.5
10
2
4
6
8
10
12
14
16
18
20
T_I
EGC_
P
RE
T_I
EGC
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
4
6
8
10
12
14
16
18
20
C_F
LAC
C_I
EAC_PR
E
C_I
EGC_
P
RE
C_MALS
0
10
20
30
40
50
60
70
80
2
4
6
8
10
12
14
16
18
20
T_FLA
C
T_I
EAC_PR
E
T_I
EGC_
P
RE
T_MALS
Evaluation Warning : The document was created with Spire.PDF for Python.
Ind
on
esi
a
n
J
E
le
c Eng &
Co
m
p
Sci
IS
S
N:
25
02
-
4752
Compari
son
of
En
tr
op
y
Co
di
ng Mec
han
ism
on I
E
EE1
857.
2
L
os
sle
ss
Aud
io
... (
Fath
i
ah A
bdul M
uin
)
183
[3]
Muin
F
.
A
.
,
e
t
al.
,
“
Perform
anc
e
Anal
y
sis
o
f
IEE
E
1857
.
2
Lossless
Audio
Com
pre
s
sion
Li
ne
ar
Predi
ctor
Algorit
hm
,”
ICSIMA
,
p
p
.
6
,
2017
.
[4]
Coal
son J.
,
“
FLAC
-
Free
Lossl
ess Audio
Code
c
,”
2017
.
[5]
Li
eb
che
n
T.
,
“
An
Introduc
ti
on
To
Mpeg
-
4
A
udio
Lossless
Coding
,”
IE
EE
Int.
Conf.
A
cou
st.
Spee
ch
Sign
al
Proce
ss
,
vol
.
3
,
p
p.
1012
–
5
,
200
4
.
[6]
Shahbahr
ami
A
.
,
e
t
al.
,
“
Eva
l
uat
ion
of
Huffm
an
and
Arithmeti
c
A
lgori
th
m
s
for
M
ult
imedia
Com
pre
ss
ion
Standa
rds
,”
Int
J
Comput
Sc
i Eng
Appl
.
,
vol
.
1
,
pp
.
34
–
47
,
2011
.
[7]
Saloman
D
.
M
.
G.
,
“
Handbook of Dat
a
Com
pre
ss
ion
,”
2010
.
[8]
Rez
nik
Y
.
A.
,
“
Coding
of
pre
d
iction
re
sidu
al
in M
PEG
-
4
standa
r
d
for
lossless a
u
dio
codi
ng
(MP
EG
-
4
ALS)
,”
I
E
EE
Int.
Con
f. A
coust
.
Spe
ec
h
,
S
ignal
Proce
ss
,
vol
.
3
,
p
p.
1024
–
7
,
200
4
.
[9]
Moffat
A
.
,
et
a
l.
,
“
Arithmeti
c
Co
ding
Rev
isit
ed
,”
ACM
Tr
ans Inf
S
yst
.
,
vol
.
16
,
pp
.
256
–
94
,
1998
.
[10]
Huang
H
.
,
et
a
l.
,
“
Lossless
audi
o
compress
ion
in
the
new
I
EE
E
St
anda
rd
for
Adv
a
nce
d
Audio
Cod
ing
,”
IEEE
Int
.
Conf.
Ac
oust
.
Sp
ee
ch
Signal P
ro
ce
ss
,
p
p
.
6934
–
8
,
2014
.
[11]
T.
S.
Gunawan,
et
al
.
,
“
Inve
stigation
of
Lossless Audio
Com
pre
ssi
on
using I
E
EE
1857.
2
Advanc
e
d
Audio
Coding
,”
Indone
sian J
our
nal
of
Elec
tric
al
Engi
ne
ering
and
Computer
Sc
ie
n
ce
,
vol
/i
ss
ue:
6(2
)
,
2017
.
[12]
“
IEE
E
1857
.
2
.
I
EE
E
Standa
rd
fo
r
S
y
st
ems
of
Advanc
ed
Audio
an
d
Video
Cod
ing,
”
2013.
[13]
Hara
da
N
.
,
et al
.
,
“
MP
EG
-
4
ALS:
Perform
ance, applicati
ons,
and
re
lated
st
anda
rd
i
za
t
ion
a
ctivit
i
es
,”
2007.
Evaluation Warning : The document was created with Spire.PDF for Python.