Indonesi
an
Journa
l
of El
ect
ri
cal Engineer
ing
an
d
Comp
ut
er
Scie
nce
Vo
l.
23
,
No.
1
,
Ju
ly
2021
, p
p.
2
3
7
~
2
46
IS
S
N: 25
02
-
4752, DO
I: 10
.11
591/ijeecs
.v
23
.i
1
.
pp
2
37
-
2
46
237
Journ
al h
om
e
page
:
http:
//
ij
eecs.i
aesc
or
e.c
om
Effici
ent
hearing
aid
alg
orithm
using DCT
with
un
iforml
y
re
-
sampl
ed
and
recu
rsivel
y modi
fied
audio
gram v
alues
K. Ayy
ap
p
a S
w
amy
1
,
Z
achari
ah
C.
Alex
2
1,2
School
of El
ectroni
cs
Engi
ne
er
ing,
Ve
ll
or
e
inst
i
tut
e
of Tec
hno
lo
g
y
,
Ve
l
lore, Ta
m
il
nadu,
Ind
ia
1
Bio
sig
na
l
Rese
arc
h
La
b
,
Sre
eVi
d
y
an
ikethan Eng
ine
er
ing
Col
le
g
e
,
T
irupati
,
Andhr
a
Prade
sh
,
Indi
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
Feb
1
6
, 202
1
Re
vised
Jun
7
,
2021
Accepte
d
J
un
1
7
, 202
1
People
with
the
hea
r
ing
probl
e
m
s
have
diffe
r
e
nt
li
st
eni
ng
pr
ef
ere
nc
es
and
cha
ra
cteri
sti
cs
in
hea
ring
loss.
So,
hea
ring
ai
ds
ne
ed
al
gori
thms
tha
t
provide
amplifi
c
at
ion
b
a
sed
on
fre
quen
c
y
,
so
tha
t
the
h
e
ari
ng
-
impair
ed
p
ersons
ca
n
use
hea
ring
ai
d
s
comforta
bl
y
f
or
a
long
dura
ti
on.
In
thi
s
pape
r,
a
ne
w
al
gorit
hm
is
p
r
oposed
for
hea
ring
ai
ds
in
orde
r
to
compensat
e
fo
r
sensorine
ura
l
an
d
conduc
ti
ve
he
ari
ng
loss
using
discre
te
cosine
tra
nsform
(DCT).
DCT
c
oef
ficien
ts
of
t
he
input
audi
o
signal
are
m
ult
iplied
wit
h
uniforml
y
r
esa
m
ple
d
and
re
cur
sively
m
odi
fie
d
aud
iogra
m
val
ues
t
o
compensat
e
for
hea
ring
loss.
Th
is
al
gori
thm
co
m
prised
of
4
sta
ges
name
l
y
pre
computation
to
ca
l
culate
gai
n
val
u
es
fr
om
audi
ogra
m
,
DCT,
gai
n
adj
ustm
ent
,
an
d
inve
rse
DCT.
In
th
e
ab
ove
stated
sta
ges
exc
ep
t
pre
computation,
ea
ch
stag
e
req
uire
s
onl
y
on
e
m
at
ri
x
m
ult
ipl
i
c
at
ion
,
which
m
ake
s
the
prop
osed
al
gori
thm
computat
ion
al
e
ffic
i
ent
.
Perfor
m
anc
e
of
th
e
proposed
al
gori
t
hm
is
compare
d
with
uniform
fil
t
er
banks,
non
-
un
iform
fil
te
r
banks,
var
ia
b
le
fil
ter
bank
and
rec
onfigur
abl
e
f
ilter
banks.
Th
e
a
lgori
thm
is
te
st
ed
using
aud
iogra
m
s
with
fo
ur
diffe
ren
t
he
ar
ing
loss
ca
ses.
It
is
prove
d
tha
t
the
propose
d
al
gori
thm
provide
s
le
ss
complexi
t
y
,
m
ini
m
ize
d
del
a
y
and
bet
t
er
m
at
ch
ing
with
al
l
t
y
pe
s
of
audi
ogr
a
m
s,
furthe
r,
it
al
so
avoi
ds
degr
adation
of
a
udio
signal
due
t
o
sam
p
li
ng
rat
e
conve
rsions
in
var
ia
b
le
an
d
rec
onfigur
able
fi
lt
er
banks
.
Ke
yw
or
d
s
:
Audiog
ram
r
e
-
sam
pling
Discrete c
os
ine
tran
s
f
or
m
Gain
a
djust
m
e
nt
Hear
i
ng aid
Sensori
neural
hear
i
ng loss
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Zacharia
h
C.
Al
ex
School
of Elec
tro
nics E
ng
i
ne
erin
g
Vell
or
e
instit
ut
e of Tec
hnolog
y, Vell
ore
Tam
il
nad
u, 63
2014, I
nd
ia
Em
a
il
: zac
har
ia
hcalex
@v
it
.a
c.in
1.
INTROD
U
CTION
Elderly
pe
op
le
m
ay
no
t
hear
pro
per
ly
due
to
dam
aged
ne
rv
e
fibe
rs
an
d
sens
or
y
cel
ls
of
the
in
ne
r
ear
[
1].
He
arin
g
ai
d
m
ay
be
us
e
d
to
c
om
pen
sat
e
f
or
t
his
disabili
ty
.
The
hear
i
ng
ai
d
is
an
el
ect
r
o
ac
ou
sti
c
dev
ic
e
th
at
am
plifie
s
sou
nd
s
ign
al
s
to
c
om
pen
sat
e
f
or
heari
ng
lo
s
s.
Howe
ver,
cha
racteri
sti
cs
of
hear
in
g
los
s
var
y
from
per
s
on
to
pe
rson
ba
sed
on
the
he
arin
g
th
res
ho
l
ds.
It
is
co
ns
id
e
red
that,
norm
al
hear
i
ng
is
be
tween
-
10 t
o
20
dB,
the m
il
d
hear
in
g
los
s
occurs
be
tween
20 to
40 dB, m
od
erate
is b
et
w
een
40 t
o
55
dB,
m
oder
at
el
y
sever
e
is
bet
w
een
55
t
o
70
dB
and
seve
re
i
s
70
to
90
dB
or
pro
f
ound
great
er
tha
n
90
dB
[
2
]
,
[
3].
W
it
h
a
sens
or
i
neural
hear
i
ng
l
os
s,
one
m
igh
t
lose
on
ly
a
certai
n
band
of
fr
e
que
ncy
[
4].
Th
us
,
Norm
al
hear
ing
a
i
d
un
i
form
l
y
a
m
p
li
fies
al
l
fr
eq
ue
ncies
in
a
ud
i
o
sign
al
s,
but
it
need
s
to
am
plify
on
ly
the
s
ound
s
t
hat
can
’t
be
he
a
r
by
hear
i
ng
-
im
paire
d,
if
not
the
loude
r
sou
nds
bec
om
e
un
be
arab
le
[
5].
Th
eref
or
e
,
in
hea
rin
g
ai
ds
,
a
pa
r
ti
cular
band
of
f
re
qu
e
ncies
of
au
dio
sign
al
s
are
s
ubj
ect
e
d
f
or
s
ui
ta
ble
gain
a
dju
stm
ent
based
on
a
n
au
diogr
a
m
to
m
ake th
e
per
s
on
unde
rstan
d
t
he
s
peec
h.
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,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
2
3
7
-
2
4
6
238
In
the
pr
ese
nt
scenario,
the
researc
h
is
car
ried
on
in
the
i
m
ple
m
entat
i
on
of
a
sig
nal
pr
oc
essin
g
al
gorithm
to
c
om
pen
sat
e
for
diff
e
re
nt
ty
pes
of
hear
i
ng
l
osse
s.
Cu
rr
e
nt
stud
ie
s
fo
c
us
on
filt
er
bank
st
r
uctu
res
with less com
pu
ta
ti
on
al
co
m
plexity
in
or
der
to r
ed
uce th
e ha
rdwar
e c
om
pl
exity
an
d
al
so
to
increase the
sp
e
e
d
of
operati
on.
Fo
r
em
os
t,
and
m
os
tl
y
us
ed
f
il
te
r
banks
are
un
i
form
filt
er
bank
[
6
]
-
[
8]
and
nonu
nif
orm
filt
er
bank
[
9
]
-
[
11]
.
Ov
e
r
the
past
decad
e
,
the
res
earche
rs
im
ple
m
ented
disti
nguish
e
d
eff
ic
ie
nt
var
ia
ble
filt
er
bank
structu
res [1
2
]
,
[
13]
and r
ec
onfig
ur
a
ble f
il
te
r
banks [14
]
-
[
20
]
to
get b
et
te
r m
at
ching
w
it
h an audio
gram
and
t
o
reduce t
he
c
ompu
ta
ti
onal
c
omplexit
y o
f
filt
er
b
a
nks
.
Con
si
der
a
bly
bette
r
m
a
tc
h
between
the
f
re
qu
e
ncy
res
ponse
of
the
hea
r
ing
ai
d
an
d
aud
i
ogram
is
achieva
ble
if
a
gr
eat
er
num
ber
of
bands
ar
e
assigne
d
for
a
un
if
or
m
and
non
-
unif
or
m
fi
lt
er
bank.
But
a
fe
w
dr
a
w
back
s
can
be
obser
ve
d
s
uch
as
delay
,
powe
r
c
onsu
m
p
ti
on
,
a
nd
the
siz
e
of
the
hea
ring
ai
d
i
ncr
ease
wi
t
h
the
increase
in
the
nu
m
ber
of
bands
.
In
case
of
unif
or
m
and
non
-
un
i
form
f
il
te
r
ban
ks
on
e
sh
ould
com
prom
ise
on
ei
ther
siz
e
a
nd
delay
or
m
a
tc
hin
g
er
r
or
.
T
o
ac
hieve
bette
r
m
at
ching
e
rror
with
sm
al
le
r
delay
a
nd
siz
e
filt
er
bank
str
uctu
re
sh
oul
d
va
ry
wi
th
ty
pe
of
t
he
a
ud
i
ogram
.
Whereas,
in
unif
orm
and
no
n
-
un
i
from
filt
er
ban
ks
the
filt
er
ba
nk
struc
ture
is
fi
xe
d
f
or
al
l
ty
pes
of
aud
i
ogram
s.
To
ove
rco
m
e
these
draw
bac
ks
,
recon
fig
ur
a
ble
filt
e
r
banks
a
re
intr
oduce
d.
I
n
a
re
config
ur
a
ble
fi
lt
er
bank
the
num
ber
of
sub
ba
nds
in
eac
h
ba
nd
va
ries
bas
ed
on
so
m
e
par
am
et
e
rs
wh
ic
h
gi
ves
dif
fer
e
nt
str
uc
tures
for
dif
fere
nt
ty
pes
of
au
diogram
s.
Eve
n
in
rec
onfig
urable
filt
er
ba
nks
a
s
hortcom
ing
is
ob
s
er
ved,
that
it
us
es
inte
rpol
at
io
n
a
nd
deci
m
at
ion
of
filt
er
co
ef
fici
ents
a
nd
/
or
input
sig
nal
to
co
nv
e
rt
the
sa
m
pl
ing
rate,
w
hich
res
ults
in
sig
nal
de
gr
a
da
ti
on
a
nd
al
so,
al
ia
sing
e
ff
ect
m
a
y
occur
du
e
to
s
a
m
pling
rate
c
onve
rsion.
T
o
ov
e
rc
om
e
this
hinderi
ng,
the
pr
e
sent
resear
ch
proposes
a
n
e
w
te
chn
iq
ue
i
n w
hich
t
he gains
are a
dju
ste
d i
n
the freq
ue
ncy
do
m
ai
n
us
i
ng
DCT.
In
this
te
ch
ni
qu
e
,
DCT
c
oe
ff
ic
ie
nts
are
m
ul
ti
plied
with
unif
or
m
ly
r
e
-
sam
pled
and
recu
r
sively
m
od
ifie
d
au
di
ogram
values
t
o
a
dju
st
gain
in
f
re
qu
e
ncy
dom
ai
n,
after
gai
n
a
dju
stm
en
t
the
fr
e
qu
e
ncy
dom
ai
n
sign
al
is
c
onve
rted
back
to
t
i
m
e
do
m
ai
n
us
ing
i
nv
e
rse
D
CT.
A
ud
i
ogra
m
values
are
r
e
-
sam
pled
at
unif
or
m
intervals
of
frequ
e
ncy
an
d
t
hey
are
m
od
if
ie
d
to
get
m
i
nim
u
m
m
at
chi
ng
e
rror.
T
his
pr
ec
om
pu
ta
ti
on
is
perform
ed
for
each
a
ud
i
ogra
m
bef
or
e
loa
di
ng
gai
n
val
ues
into
the
hea
ri
ng
ai
d.
The
pr
opos
e
d
DCT
ba
sed
al
gorithm
is
bette
r
wh
e
n
c
ompare
d
to
fixe
d
filt
er
ba
nk
s
an
d
rec
onfi
gurabl
e
filt
er
ba
nk
s
. Th
is
is bett
er
in
te
rm
s
of
c
om
plexity
as
it
has
on
ly
3
m
a
trix
m
ultip
li
cat
ion
s
to
pe
rfor
m
gain
ad
j
u
stm
ent
in
the
fr
eq
ue
ncy
dom
ai
n.
It
giv
es
bette
r
m
at
ching
er
ror
a
s
it
us
es
rec
urs
ive
m
od
ific
at
ion
s
of
a
ud
i
ogr
a
m
values
bas
ed
on
m
at
ching
er
ror.
This
DCT
bas
ed
te
ch
nique
ne
ed
pre
process
ing
to
get
gain
values
from
a
ud
i
ogram
wh
ic
h
is
not
re
quir
ed
i
n
filt
er
ban
k
str
uc
tures
that
is
the
only
disad
va
ntage
of
the
pro
po
se
d
te
ch
nique.
A
udio
gram
pr
eproces
sing
is
perform
ed
befor
e
loa
ding
the
gain
values
in
to
the
process
or.
So,
it
wo
n'
t
aff
ect
the
sp
ee
d
of
the
hear
in
g
ai
d
syst
e
m
.
The
pro
po
s
ed
al
gorithm
pr
ov
ides
a
sim
ple
so
luti
on
to
com
pensat
e
f
or
t
he
hear
i
ng
loss
without
a
ny
filt
er
banks
a
nd
sam
pling
rate
conversi
ons.
Total
ly
3
sta
ge
s
ar
e
nee
ded
for
the
w
ho
le
process
i)
f
i
nd
i
ng
D
CT
for
the
in
put
aud
i
o
sig
nal,
ii
)
gain
a
dju
stm
e
nt
an
d
ii
i)
i
nv
e
rse
DCT
.
Th
e
pro
po
se
d
al
go
r
it
h
m
is
te
ste
d
us
i
ng
aud
i
ogram
s
with
f
our
differe
nt
hea
rin
g
lo
s
s
cases
s
uch
a
s
m
il
d
hear
i
ng
loss
at
high
f
reque
ncies,
m
i
ld
to
m
od
erate
hear
i
ng
l
os
s
at
lo
w
fr
e
qu
e
ncies,
m
od
e
rate
hea
rin
g
loss
at
m
idd
le
fr
e
qu
e
ncies
and
m
il
d
cond
uctive
hear
i
ng
loss
.
It
is
no
te
d
t
hat
the
pro
posed
al
gorithm
prov
i
de
s
le
ss
c
om
plexity
,
le
ss
delay
an
d
bette
r
m
at
chin
g
with
a
ud
i
ogra
m
with
al
l
ty
pe
s
of
a
ud
i
ogram
s.
It
al
s
o
a
vo
i
ds
de
gr
a
datio
n
of
a
udio
sig
nal
due
to
sam
pli
ng
rat
e
conve
rsions t
ha
t are
us
ed
in v
ariable a
nd r
ec
onfig
ur
a
ble
filt
er
banks.
The
pap
e
r
is
orga
nized
as
f
ol
lows
:
Sect
io
n
2
deals
with
t
he
im
ple
m
enta
ti
on
of
t
he
pro
po
s
ed
DCT
base
d
al
gorith
m
. S
ection
3
di
scusses
t
he
pre
com
pu
ta
ti
on
to
f
in
d
the
gain v
al
ues
f
ro
m
au
di
ogram
. I
n
sect
ion
4
,
desig
n
e
xam
pl
es
an
d
pe
rfo
r
m
ance
evaluat
ion
pro
po
se
d
al
gorithm
are
te
ste
d
with
a
udio
gr
am
s
with
four
diff
e
re
nt
hea
ring
loss
ca
ses.
Sect
ion
5
br
i
ngs
in
ex
per
im
ental
res
ults
an
d
a
naly
sis.
Fin
al
ly
,
the
co
nclusion
is
dr
a
w
n
in
secti
on
6
.
2.
PROP
OSE
D DC
T B
AS
E
D ALGO
RITH
M
In
this
pro
pose
d
te
chn
i
qu
e
,
th
e
aud
io
sig
nal
is
transfor
m
ed
to
the
fr
eq
ue
nc
y
do
m
ai
n
us
i
ng
DCT
to
a
m
plify
the
au
dio
sig
nal
as
pe
r
the
ab
ove
re
qu
i
rem
ent.
The
com
m
on
ly
us
e
d
tra
nsfo
rm
dom
ai
n
ap
proache
s
are
base
d
on
DC
T
[21].
DCT
is
use
d
t
o
c
onve
rt
the
data
int
o
a
su
m
of
c
os
i
ne
wav
e
tray
s
of
diff
e
re
nt
f
requ
encies.
As
the
DCT
c
oe
ff
ic
ie
nts
a
re
a
rr
a
ng
e
d
in
asce
nd
i
ng
orde
r
wi
th
res
pect
to
th
ei
r
corres
pondi
ng
fr
e
quencies
,
it
i
s
ver
y
easy
to
adjust
the
gains.
Un
if
orm
l
y
re
-
sam
pled
and
recursively
m
od
ifie
d
a
udio
gr
am
value
s
are
m
ul
ti
plied
with
the
DCT
c
oeff
ic
ie
nts
of
the
au
dio
sig
nal
to
pe
rform
the
gain
a
dju
stm
ent.
A
udio
gr
am
re
-
sam
pling
is
d
isc
us
s
ed
in
t
he
precom
pu
ta
ti
on
sect
io
n
of
this pap
e
r.
Fig
ure
1
r
ep
rese
nts the
bl
oc
k
diagra
m
of
the
pr
opos
e
d
al
gorithm
.
As
per
the
blo
c
k
diag
ram
,
inp
ut
au
dio
sig
nal
sense
d
by
m
icr
op
hone
is
give
n
t
o
analo
g
t
o
dig
it
al
conve
rter
(
AD
C
)
wh
ic
h
c
onve
rts
the
an
al
og
a
udio
sig
nal
int
o
a
dig
it
al
aud
i
o
si
gn
al
.
T
hen,
that
dig
it
al
sig
nal
is
tra
ns
f
or
m
ed
into
t
he
frequ
e
ncy
do
m
ai
n
us
in
g
DCT
t
ran
s
f
or
m
.
Ou
t
pu
t
of
the
DC
T
bl
ock
is
DCT
coeffic
ie
nts.
Furthe
r,
in
the
gain
a
djust
m
ent
blo
ck
DCT
coeffic
ie
nts
are
m
ultip
li
ed
with
un
i
f
or
m
ly
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
Eff
ic
ie
nt h
eari
ng a
i
d alg
or
it
hm
us
in
g DCT
wi
th unif
or
mly
re
-
sam
pled
an
d
rec
ur
siv
el
y…
(Zac
ha
ri
ah C.
Alex
)
239
re
-
sam
pled
an
d
rec
ur
si
vely
m
od
ifie
d
au
dio
gram
values.
The
n,
in
verse
DCT
tra
ns
f
orm
s
the
a
m
plifi
ed
DC
T
coeffic
ie
nts
int
o
tim
e
do
m
ai
n
.
Finall
y,
dig
it
al
to
analo
g
conv
e
rter
(
DAC
)
is
to
conve
rt
the
dig
it
al
aud
it
ory
com
pen
sat
ed
si
gn
al
t
o
a
nalo
g and gi
ven to t
he
audio
sp
ea
ke
r.
Fig
ure
2
(
a
)
e
xp
la
in
s
the
w
orkin
g
proce
dure
of
the
pr
opose
d
al
gorith
m
.
Con
side
r
the
sam
pling
fr
e
qu
e
ncy
of
t
he
input
aud
i
o
sign
al
Fs=1
6
kH
z
.
In
order
to
app
ly
80
-
point
DCT
on
in
pu
t
sig
nal
80
i
nput
sam
ples
need
t
o
be
st
or
e
d
i
n
the
bu
ff
e
r.
The
n,
a
pply
80
-
po
int
DCT
to
t
he
input
sam
ples
store
d
in
buf
f
er
to
conve
rt
the
ti
m
e
do
m
ai
n
sign
al
i
nto
f
requen
cy
-
dom
ai
n.
Lat
er,
in
Fig
ur
e
2
(
b)
the
DCT
c
oeffici
ents
ar
e
m
ul
ti
plied
with u
nif
orm
l
y
re
-
sam
pled
and
r
e
cur
si
vely
m
od
ifie
d
a
ud
i
ogram
values
to
a
djust
gain
value
s.
Th
us
,
to
co
nv
e
rt
the
a
m
plifie
d
signa
ls
back
to
ti
m
e
-
dom
ai
n,
app
l
y
80
-
point
in
ve
rse
DCT
.
N
ow
the
outp
ut
is
tim
e
do
m
ai
n
au
dio s
ign
al
a
fter
gain
adjust
m
ent.
Figure
1. Bl
oc
k diag
ram
fo
r pro
po
se
d
al
gor
it
h
m
(a)
(b)
Figure
2.
The
s
e figure
s ar
e;
(
a)
w
orki
ng pr
oc
edure
of t
he p
rop
os
ed
alg
or
it
hm
, an
d (
b)
m
od
i
fyi
ng audio
gr
am
values
r
ec
ur
si
ve
ly
3.
PREC
OM
PU
TATION
This
sect
ion
discuss
es
how
a
ud
i
ogram
values
are
re
-
sam
pled
and
m
od
ifi
ed
rec
ur
si
vely
in
order
to
fin
d
the
g
ai
n v
al
ues
that a
re
ne
eded to
be a
dj
us
te
d i
n fr
e
que
ncy dom
ai
n
.
3.1.
Au
di
og
r
am
re
-
s
amp
li
ng
Gr
a
ph
in
Fig
ure
3
(
a
)
represe
nts
the
aud
i
ogr
a
m
values
fo
r
the
hear
i
ng
los
s
case
with
m
i
ld
hear
in
g
los
s
at
m
id
fr
equ
e
ncy.
Gr
a
ph
in
Fig
ure
3
(
b
)
is
the
un
if
or
m
ly
re
-
sa
m
ples
au
diogram
values
at
100
H
z
that
m
ean
sam
plin
g
inter
val
in
f
r
equ
e
ncy
dom
a
in
is
100
Hz
.
Fr
om
the
aud
i
ogram
in
Fig
ure
4
(
a
)
,
it
is
obser
ve
d
that
the
a
udio
gr
am
is
recor
de
d
at
non
-
unif
or
m
fr
e
qu
e
nci
es.
I
n
orde
r
t
o
get
unifo
rm
ly
sam
pled
au
di
ogra
m
values fr
om
the abov
e
audio
gram
, it i
s n
eede
d
to
b
e
re
-
sam
pled
at
unifo
r
m
intervals
of
f
reque
ncy. T
his
can be
i
m
ple
m
ented
by
a
si
m
p
le
com
pu
te
r
app
li
cat
ion
by
the
au
dio
lo
gist
befo
r
e
pr
og
ram
m
ing
the
dig
it
al
hear
in
g
ai
d.
I
nter
po
la
ti
on
is
nee
de
d
f
or
re
-
sam
pling
of
an
a
ud
i
ogr
a
m
and
interp
ol
at
ed
sa
m
ples
sh
oul
d
be
an
a
ver
a
ge
of
previ
ou
s
an
d
ne
xt
sam
ple
values
.
A
s
the
aud
i
ogram
is
reco
r
de
d
at
no
n
-
unif
or
m
fr
eq
ue
ncies
it
m
a
y
req
ui
re
non
-
unif
or
m
i
nter
po
la
ti
on;
f
or
e
xam
ple,
if
the
new
sam
pling
i
nterv
al
in
fr
e
quency
dom
ai
n
is
10
0
Hz
that
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,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
2
3
7
-
2
4
6
240
m
eans
if
the
au
diogram
need
s to
be
re
-
sam
pled
at
ever
y 100
Hz
the
n
betw
een
50
0
Hz
a
nd
10
00
H
z
4
sa
m
ples
need
t
o
be
int
erpolat
ed,
wh
e
reas
bet
ween
4000
Hz
a
nd
8000
Hz
39
s
a
m
ples
are
to
be
inter
po
la
te
d.
Thi
s
non
-
unif
or
m
r
e
-
sam
pling
of t
he
audio
gr
am
is exp
l
ai
ne
d
i
n
t
he
equati
on
.
(a)
(b)
Figure
3.
The
s
e figure
s ar
e;
(
a)
a
ud
i
ogram
values,
a
nd (b
)
un
i
form
l
y
re
-
s
a
m
pled
au
di
ogram
v
al
ues.
In
(
2
)
F(k)
c
onta
ins
the
f
re
quencies
at
w
hi
c
h
the
au
diog
r
a
m
is
reco
rd
e
d.
In
(1)
,
G
(
k)
con
ta
in
s
the
gain
val
ue
s
in
the
a
ud
i
ogra
m
cor
respo
nd
i
ng
to
t
he
fr
e
quencies
de
fine
d
in
the
F(k)
.
N
is
th
e
in
de
x
of
the
current
inter
po
la
ti
ng
sam
ple
a
m
on
g
the
N
nu
m
ber
of
sam
pl
es.
N
is
the
nu
m
ber
of
sam
pl
es
to
be
int
er
pola
te
d
betwee
n
(k
-
1)
th
an
d
k
th
sam
ple
s
of
the
au
di
ogram
.
F
d
is
t
he
fr
e
quency
di
ff
ere
nce
betw
een
tw
o
int
-
e
r
po
la
te
d
sam
ples
(n
ew
un
i
form
sa
m
pl
ing
inte
rv
al
in
fr
e
qu
e
ncy
do
m
ai
n)
F
d
=1
00.
Ele
m
ents
in
G
n
(
n,k)
a
re
ex
pand
e
d
into a si
ngle
v
e
ct
or
of size
80
for
F
d
=
100
.
(
,
)
=
(
−
1
)
+
[
(
)
−
(
−
1
)
]
(1)
wh
e
re
:
=
[
(
)
−
(
−
1
)
]
(2)
(
0
)
=
(
0
)
=
0
(3)
3.2
.
Recursi
ve
modi
fica
tion o
f audi
og
r
am
values
The
pro
pose
d
al
gorithm
can
be
i
m
ple
m
ented
directl
y
bu
t
the
m
a
tc
hin
g
er
ror
betwee
n
au
diogram
and
fr
e
qu
e
ncy
respon
s
e
of
t
he
de
sign
e
d
syst
em
is
ver
y
high
i
n
so
m
e
cases
li
ke
m
od
erate
s
ens
or
ine
ural
he
arin
g
loss.
T
o
m
ini
m
iz
e
this
m
at
c
hing
er
ror,
m
od
ific
at
ion
of
ga
in
val
ues
(a
udio
gr
am
re
-
sa
m
pled
values)
us
in
g
a
recursive
al
gor
it
h
m
is
pr
opose
d.
I
n
this
al
gorithm
,
the
weigh
te
d
m
at
ching
erro
r
is
recursively
add
e
d
t
o
the
un
i
form
l
y
re
-
sa
m
pled
aud
i
og
ram
values.
Th
is
pr
oce
ss
m
a
y
be
rep
eat
e
d
ti
ll
the
m
at
ching
erro
r
is
redu
ced
to
m
ini
m
u
m
le
vel.
The
num
ber
of
it
erati
ons
a
nd
the
weig
ht
value
s
de
pe
nd
on
the
ty
pe
of
aud
i
ogram
.
By
tria
l
and
er
r
or
,
it
is
ob
s
er
ved
that
t
he
weig
ht
valu
e
is
in
bet
ween
0.1
an
d
1.
Gain
values
are
lo
aded
int
o
th
e
he
ari
ng
ai
d
after
re
-
sa
m
pl
ing
an
d
m
od
i
ficat
ion
.
T
o
ge
ner
at
e
m
o
dified
gai
n
va
lues
for
the
gi
ven
au
diogr
a
m
th
is
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
Eff
ic
ie
nt h
eari
ng a
i
d alg
or
it
hm
us
in
g DCT
wi
th unif
or
mly
re
-
sam
pled
an
d
rec
ur
siv
el
y…
(Zac
ha
ri
ah C.
Alex
)
241
al
gorithm
is
i
m
ple
m
ented
on
PC
.
C
hanges
in
m
at
ching
e
rror
co
nce
rn
i
ng
t
he
nu
m
ber
of
it
erati
on
s
a
nd
weig
ht
is ex
plained
in deta
il
in
the
se
ct
ion
V
e
xp
e
ri
m
ental
r
esults
and analy
sis.
The
wor
king
proce
dure
f
or
re
cur
si
vely
m
od
ify
ing
a
udio
gr
a
m
gain
values
is
show
n
i
n
Fi
g
ure
2
(
b
)
.
In
t
his
pr
oces
s,
gai
n
values
from
aud
io
gra
m
are
subj
e
c
te
d
to
the
uni
form
l
y
re
-
sam
pling
bl
ock
t
o
obta
in
un
i
form
l
y
re
-
s
a
m
pled
a
ud
i
ogram
values.
T
he
se
gai
n
value
s
are
rec
ur
si
ve
ly
m
od
ifie
d
ba
sed
on
the
m
atch
in
g
error,
if
m
at
ch
ing
e
rro
r
is
le
s
s
tha
n
pr
e
def
i
ned
th
reshol
d
value
t
he
n
it
unde
r
go
es
one
m
or
e
it
erati
on
.
Gai
n
m
od
ific
at
ion
in
this alg
or
it
hm
i
m
plies updati
ng g
ai
n values
with the
w
ei
ghte
d
m
at
ching
e
rror
4.
DESIG
N
E
XA
M
PLE
S
AND
PE
RFO
R
M
ANCE E
VA
L
UA
TI
ON
The
idea
of
t
he
pro
pose
d
al
gorithm
fo
r
he
arin
g
ai
d
is
exam
ined
by
us
in
g
so
m
e
exa
m
ples.
Th
e
perform
ance
of
the
pro
pose
d
al
gorithm
is
evaluated
us
in
g
au
diogram
s
w
it
h
f
our
ty
pes
of
hea
rin
g
l
os
s
cases.
Ba
sed
on
le
vels
of
hea
rin
g
th
reshold
s,
hea
ring
los
ses
are
c
at
egorized
as
m
il
d,
m
od
erate,
m
od
eratel
y
severe
,
sever
e
a
nd
pro
fou
nd.
Hea
rin
g
loss
cases
li
ke
sever
e
a
nd
pro
fou
nd
m
ay
no
t
be
com
pen
s
at
ed
us
in
g
a
he
arin
g
ai
d
[
22]
.
4.
1
.
E
xa
m
ple 1
:
A
udio
gr
am
for mil
d
he
aring l
os
s
at
h
igh
freque
ncie
s
The
au
diog
ra
m
fo
r
m
i
ld
hear
in
g
loss
at
hi
gh
f
reque
ncie
s
is
sh
own
in
Figure
4
(a)
.
The
ri
gh
t
ear
hear
i
ng
th
res
holds
re
prese
nted
by
'
O'
are
consi
der
e
d
f
or
com
pen
sat
ion.
Accor
ding
to
the
aud
i
ogra
m
gain
values
a
re
5,
5,
5,
5,
35,
5
dB
.
Gain
values
are
gi
ven
t
o
the
re
-
sam
pling
and
rec
ur
si
vely
m
od
ific
at
ion
bloc
k,
the
n
obta
ine
d 80
un
i
form
ly s
a
m
pled
gain va
lues at t
he
ou
t
put. Thus,
t
hese
80
gain values
are p
r
ovide
d
to
g
ai
n
adjustm
ent
bloc
k.
Gains
are
m
ul
ti
plied
with
the
DCT
c
oe
ff
ic
ie
nts
of
t
he
au
dio
si
gn
al
i
n
the
gain
ad
ju
st
m
ent
blo
c
k.
A
fter
ga
in
ad
justm
ent,
app
ly
in
ve
r
se
DCT
to
c
onvert
a
m
plifie
d
DC
T
coe
ff
ic
ie
nts
back
to
tim
e
do
m
ai
n.
Figure
5
(a)
s
hows
the
a
udio
gr
am
values
a
nd
the
fr
e
quen
cy
respo
ns
e
of
the
hear
i
ng
ai
d
syst
em
.
Figure
6
(
a
)
represe
nts
the
m
at
ching
e
rror
betwee
n
t
he
r
e
-
sam
pled
au
di
ogram
and
th
e
fr
e
quency
re
sp
onse
of
the
hear
i
ng
ai
d
syst
e
m
.
Ma
tc
hin
g
e
rro
r
is
the
di
ff
e
r
ence
bet
wee
n
re
-
sam
pled
a
ud
i
ogram
values
an
d
the
frequ
e
ncy
respo
ns
e
of
th
e
hea
rin
g
ai
d
s
yst
e
m
.
Fr
om
the
a
bove
Fig
ures
5
(a)
an
d
6
(a)
an
d
Table
1,
it
is
cl
ear
t
hat
the
pro
po
se
d
al
go
r
it
h
m
per
f
or
m
s
bette
r
in
te
rm
s
of
m
axim
u
m
m
at
ching
er
r
or
an
d
delay
,
w
hen
c
om
par
ed
with
filt
er
bank
te
chn
i
qu
e
s.
Ma
xim
u
m
m
at
chi
ng
e
rror
is
th
e
m
axi
m
u
m
d
iffer
e
nce
bet
w
een
au
diogra
m
and
fr
e
qu
e
ncy
res
ponse
of
the
he
arin
g
ai
d
syst
e
m
.
Fr
om
Table
2
the
propose
d
DCT
base
d
al
gor
it
hm
giv
es
0.49
dB m
at
ching
e
rror at
20 iterat
ion
s
and
with
weig
ht 0.7.
Table
1.
C
om
par
iso
n of t
he pr
opos
e
d
al
gorithm
w
it
h
the
filt
er b
a
nk str
uct
ur
es
in [1
0
]
,
[
11
]
,
[
15
]
,
[
23
]
,
[
17
]
,
[
19]
Fi
l
t
er
b
a
n
k
E
x
am
p
l
e 1
E
x
am
p
l
e 2
E
x
am
p
l
e 3
E
x
am
p
l
e 4
N
u
m
b
er
o
f
s
i
d
e
b
a
n
d
s
Max
i
m
u
m
Mat
c
h
i
n
g
E
rro
r (
d
B)
D
el
ay
(
m
s
)
N
u
m
b
e
r of
s
i
d
e
b
a
n
d
s
Max
i
m
u
m
Mat
c
h
i
n
g
E
rro
r (
d
B)
D
el
ay
(
m
s
)
N
u
m
b
er
o
f s
i
d
e
b
a
n
d
s
Max
i
m
u
m
Mat
c
h
i
n
g
E
rro
r
(d
B)
De
l
ay
(m
s)
N
u
m
b
er
o
f s
i
d
e
b
a
n
d
s
Max
i
m
u
m
Mat
c
h
i
n
g
E
rro
r (
d
B)
D
el
ay
(
m
s
)
Direct
d
esig
n
8
6
.39
4.
3
-
-
-
-
-
-
8
5
.86
4
.3
[
1
0
]
10
9
.61
1
5
.7
8
3
.2
5
.7
8
9
.2
15
.7
10
3
.67
5
.7
[
1
1
]
16
2
.10
1
2
.8
-
-
-
-
-
-
-
-
-
[
1
5
]
8
4
.82
29
-
-
-
-
-
-
7
2
.67
25
[
2
3
]
7
5
.63
1
2
.1
-
-
-
-
-
-
7
1
.84
2
.1
[
1
7
]
10
2
.84
6
.6
12
1
.51
12
13
2
.72
18
11
1
.49
12
[
1
9
]
6
2
.84
1
5
.7
5
12
1
.49
12
13
2
.72
18
7
1
.36
1
.09
Prop
o
sed
alg
o
rith
m
-
1
7
.4
-
0
.88
7
.4
-
1
.87
7.
4
-
0
.15
7
.4
4.
2
.
E
xa
m
ple 2
:
A
udio
gr
am
for mil
d
to
mod
er
at
e
he
ar
ing loss
at lo
w
f
r
equen
ci
es
Accor
ding
to
the
au
diog
ram
show
n
in
Figure
4
(b)
t
he
gain
val
ues
are
45,
35,
20,
10,
5,
10
resp
ect
ively
.
F
igure
5
(b)
re
pr
ese
nts
the
a
ud
i
ogram
values
and
t
he
f
re
qu
e
ncy
res
ponse
of
t
he
hea
ring
a
i
d
syst
e
m
.
Figu
re
6
(b)
re
pr
ese
nt
s
the
m
at
ching
erro
r
.
From
Figures
5
(b),
6
(
b)
,
a
nd
Ta
bl
e
1
it
is
c
le
ar
t
hat
the
pro
po
se
d
al
go
r
it
h
m
per
f
or
m
s
bette
r
in
te
rm
s
of
m
axim
u
m
m
at
ching
er
r
or
an
d
delay
,
w
hen
c
om
par
ed
with
filt
er
ba
nk
te
c
hn
i
qu
e
s.
DCT
base
d
al
gorit
hm
giv
es
0.8
8
dB
m
at
ching
er
ror
at
20
it
erati
on
s
a
nd
with
weig
ht 0.3.
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,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
2
3
7
-
2
4
6
242
4.
3
.
E
xa
m
ple 3
:
A
udio
gr
am
for m
od
er
ate he
arin
g
los
s
at middle
fre
quencies
Accor
ding
to
t
he
a
ud
i
ogram
sh
ow
n
in
Fig
ure
4
(c
)
the
ga
in
val
ues
ar
e
10,
20,
40,
50,
20,
a
nd
10
resp
ect
ively
.
F
igure
5
(
c)
s
hows
the
a
ud
i
ogr
a
m
values
an
d
the
fr
e
q
uen
cy
r
esp
on
se
of
t
he
hear
i
ng
ai
d
syst
e
m
.
Figure
6
(c)
re
pr
ese
nts
the
m
at
ching
e
rro
r.
Fr
om
the
abov
e
resu
lt
s,
it
is
cl
ear
that
the
pro
po
se
d
DCT
based
al
gorithm
g
ives 1
.
87
dB m
at
c
hing e
rror at
20 iterat
io
ns
a
nd
with
weig
ht
0.4
4.
4
.
E
xa
m
ple 4
:
A
udio
gr
am
fo
r mil
d c
onducti
ve heari
ng
l
os
s
Accor
ding
to
the
au
diogram
sh
ow
n
in
Fig
ure
4
(
d
)
the
ga
in
values
are
25,
25,
25,
35,
25,
an
d
30
resp
ect
ively
.
F
igure
5
(d)
s
ho
ws
t
he
a
ud
i
ogr
a
m
values
an
d
the
f
re
qu
e
ncy
r
esp
on
se
of
t
he
hear
i
ng
ai
d
sys
tem
.
Figure
6
(
d)
re
pr
ese
n
ts
the
m
at
ching
er
ror.
It
is
cl
ear
fro
m
the
above
st
at
ed
Fig
ur
es
4
(d),
5
(
d),
6
(
d),
an
d
Table
1
t
hat
t
he
pro
posed
al
gorithm
per
f
or
m
s
bette
r
wh
e
n
c
om
par
ed
w
it
h
al
l
filt
er
ba
nk
te
ch
niques.
From
Table
2
t
he
propose
d
DCT
ba
sed
al
go
rithm
giv
es
0.1
5
dB
m
at
chin
g
e
rro
r
at
20
it
erati
ons
a
nd
with
w
ei
gh
t
1.
Fr
om
the a
bove
ex
am
ples, it i
s ev
id
ent t
hat the
DCT
based
aud
it
ory
c
om
pen
sat
io
n
is si
m
ple to im
ple
m
e
nt and
has
only
80
m
ulti
pliers
in
ea
ch
of
thre
e
sta
ges
nam
el
y
DCT,
gai
n
a
dju
s
t
m
ent
and
in
ve
rse
DCT.
T
he
m
at
ching
error i
s m
ini
m
um
w
hen com
par
e
d wit
h
t
he fixe
d
filt
er
ba
nk a
nd the
rec
onfig
ur
a
ble f
il
te
r ban
k
(a)
(b)
(c)
(d)
Figure
4. A
udiogram
f
or
;
(a)
Mi
ld h
eari
ng l
os
s at
high
fr
e
quencies
, (b
)
Mi
ld to
m
od
e
rate
hear
i
ng loss
at
low
f
re
qu
e
ncies,
(
c
)
Mo
de
r
at
e h
ea
rin
g
los
s at m
i
d fr
e
qu
e
ncies,
and (
d) Mi
ld c
onduct
ive
heari
ng
loss
[
10]
, [1
7],
[19]
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
Eff
ic
ie
nt h
eari
ng a
i
d alg
or
it
hm
us
in
g DCT
wi
th unif
or
mly
re
-
sam
pled
an
d
rec
ur
siv
el
y…
(Zac
ha
ri
ah C.
Alex
)
243
(a)
(b)
(c)
(d)
Figure
5. A
udiogram
v
al
ues
a
nd the
fr
e
quen
cy
r
esp
onse
of
hear
i
ng a
id
for
;
(a)
m
il
d
hear
i
ng loss
at
highf
reque
ncie
s,
(
b)
m
il
d
to m
od
er
at
e h
ea
ring loss
at lo
w f
reque
ncie
s, (c)
m
od
erate hea
ring loss
at
m
idfr
eq
ue
ncie
s,
a
nd (d)
m
i
ld co
nd
uctive
hea
rin
g
los
s
Table
2.
Ma
xi
m
u
m
m
at
ching erro
r
f
or
diff
e
r
ent h
ea
rin
g
l
oss case
s
for give
n weig
hts a
nd the
nu
m
ber
of
it
erati
ons
Au
d
io
g
ra
m
No
.
o
f
i
terat
-
io
n
s
Maxi
m
u
m
m
atch
in
g
er
ror fo
r
g
iv
en
w
eig
h
t
0
.1
0
.2
0
.3
0
.4
0
.5
0
.6
0
.7
0
.8
0
.9
1
Exa
m
p
le 1
0
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
-
1
1
.27
5
-
1
1
.03
-
1
0
.67
-
10
-
9
.2
-
8
.34
-
7
.4
-
6
.62
-
5
.82
-
5
.1
-
4
.44
10
-
1
0
.63
-
9
.18
-
7
.56
-
6
.05
-
4
.73
-
3
.59
-
2
.6
-
1
.74
-
1
.28
-
0
.95
15
-
9
.95
-
7
.59
-
5
.44
-
3
.7
-
2
.32
-
1
.36
-
0
.93
-
0
.65
-
0
.59
0
.7
20
-
9
.18
-
6
.14
-
3
.76
-
2
.01
-
1
.07
-
0
.67
0
.49
0
.7
0
.76
0
.72
Exa
m
p
le 2
0
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
-
9
.03
5
-
6
.43
-
4
.85
-
3
.63
-
2
.71
2
.54
2
.91
3
.73
4
.69
6
.07
6
.58
10
-
4
.88
-
2
.77
1
.98
2
.45
2
.22
-
1
.74
-
3
-
3
.67
2
.13
6
.17
15
-
3
.71
1
.95
1
.82
0
.97
-
1
.68
-
1
.35
3
.68
5
.52
4
.79
6
.33
20
-
2
.8
1
.68
0
.88
-
1
.06
1
.37
1
.62
-
3
.53
-
3
.74
6
.49
-
3
.64
Exa
m
p
le 3
0
-
1
4
.48
-
1
4
.48
-
1
4
.48
-
1
4
.48
-
14.
48
-
1
4
.48
-
1
4
.48
-
1
4
.48
-
1
4
.48
-
1
4
.48
5
-
1
0
.9
-
8
.88
-
7
.3
-
6
.14
-
5
.08
-
4
.22
-
3
.45
4
.6
7
.11
9
.55
10
-
8
.87
-
6
.06
-
4
.08
2
.75
4
.53
5
.43
5
.19
-
6
.16
-
9
.53
-
1
2
.5
15
-
7
.34
-
4
.06
2
.97
3
.73
2
.95
-
5
.31
-
6
.69
-
6
.38
1
6
.17
-
1
1
.32
20
-
6
.03
-
2
.52
3
.14
1
.87
-
4
.53
4.
12
1
2
.53
1
5
.9
20
-
199
Exa
m
p
le 4
0
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
-
2
.89
5
1
.89
1
.9
1
.86
1
.77
1
.64
1
.48
1
.3
1
.1
0
.92
0
.75
10
1
.89
0
.75
1
.47
1
.16
0
.88
0
.68
0
.54
0
.45
0
.4
0
.35
15
1
.83
1
.47
1
.04
0
.71
0
.52
0
.41
0
.34
0
.28
0
.24
0
.21
20
1
.73
1
.18
0
.73
0
.49
0
.37
0
.29
0
.24
0
.2
0
.17
0
.15
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,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
2
3
7
-
2
4
6
244
(a)
(b)
(c)
(d)
Figure
6. Ma
tc
hing e
rror f
or
;
(a)
m
il
d
hear
i
ng lo
ss at
high
f
reque
ncies,
(b)
m
il
d
to m
od
er
at
e h
eari
ng los
s
at
low fr
e
qu
e
nc
ie
s,
(c
)
m
od
era
te
h
eari
ng loss
at
m
id f
re
quen
ci
es,
an
d (d) m
il
d
co
nductive
hear
i
ng loss
.
5.
E
X
PERI
MEN
TAL RES
UL
TS A
ND AN
A
LYSIS
Ma
tc
hin
g
er
rors
for
dif
fer
e
nt
ty
pes
of
a
udio
gr
am
s
with
the
dif
fer
e
nt
num
ber
of
it
erati
on
s
a
nd
weig
hts
are
s
how
n
in
the
Ta
ble
2.
From
the
Table
2,
it
is
cl
ear
that
th
e
m
a
tc
hin
g
e
rro
r
c
ha
ng
e
s
wit
h
th
e
nu
m
ber
of
it
erati
on
s
an
d
we
igh
t.
I
n
the
case
of
m
i
ld
hear
in
g
loss
at
high
fr
e
quenc
y
(ex
am
ple
1)
bette
r
m
at
ching
can be
ob
se
r
ved
at
w
ei
ght
gr
eat
e
r
than
or
e
qu
al
s to
0.7.
In
th
e
case
of
m
il
d
to
m
od
erate
hear
i
ng
l
os
s
at
low
fr
e
quen
cy
and
m
o
der
a
te
hear
i
ng
loss
at
m
idd
le
f
re
quency
(e
xam
pl
e
2
a
nd
3)
bett
er
m
at
ching
ca
n
be
ob
s
er
ved
at
w
ei
gh
t
betwee
n
0.3
an
d
0.7.
Wh
e
reas
i
n
co
nductive
hear
i
ng
l
os
s
case
,
be
tt
er
m
at
ching
can
be
ob
s
er
ved
at
w
ei
gh
t
eq
uals
to
1.
I
n
al
l
ca
ses
m
at
ching
error
re
duces
wit
h
an
increa
se
in
the
nu
m
ber
of
it
erati
on
s.
By
analy
zi
ng
Table
2
we
ca
n
obser
ve
t
hat
for
al
l
cases
m
at
ching
er
r
or
is
bette
r
at
20
it
erati
on
s.
I
f
hear
i
ng
lo
ss
is
at
hig
h
fr
e
qu
e
nc
y
weigh
t
s
houl
d
be
bet
ween
0
.
7
an
d
0.9
.
I
n
case
of
lo
w
f
re
qu
e
ncy
hea
rin
g
los
s
w
ei
ght
sh
oul
d
be
betwee
n
0.3
an
d
0.6
.
From
exa
m
ple
3
in
case
of
m
idd
le
fr
e
quenc
y
weigh
t
shoul
d
be
betwee
n 0.4 a
nd
0.7. Fo
r
c
onduct
ive
hear
i
ng
loss
weig
ht s
houl
d be
1.
The
pr
opos
e
d
al
gorithm
is
tested
us
i
ng
a
n
a
ud
i
o
sig
nal
f
or
the
au
diogram
with
m
i
ld
hea
rin
g
loss
in
high
fr
e
quency
as
sh
own
in
F
ig
ure
4.
T
he
frequ
e
ncy
-
dom
a
in
represe
ntati
on
of
the
in
pu
t
aud
io
sig
nal
and
th
e
a
m
plifie
d
signa
ls
are
sh
own
in
Fig
ure
7
(
a
)
.
Thus,
it
is
clear
that
the
gain
is
m
axi
m
u
m
at
the
fr
eq
ue
ncies
gr
eat
er
t
han
2000
Hz
an
d
the
gai
n
is
ch
ang
i
ng
with
t
he
f
reque
ncy
con
ce
r
ning
th
e
aud
i
ogram
.
Fr
om
tim
e
-
do
m
ai
n
wav
ef
or
m
s
show
n
in
Fi
g
ure
7
(
b
)
,
it
is
cl
ear
th
at
sign
al
s
with
high
fr
e
quency
are
am
plifie
d
wit
h
high
gain val
ue
s and l
ow
-
f
re
qu
e
ncy c
om
po
nen
ts
are
am
plifie
d wit
h
sm
al
l
er
gain valu
es
.
5.
1
.
Dela
y an
aly
sis
The
pro
posed
al
gorithm
co
m
pr
ise
s
th
ree
sta
ges
DCT,
gain
ad
j
us
tm
ent
and
in
verse
D
CT.
Be
fo
r
e
loading
gai
n
va
lues
int
o
t
he
hear
i
ng
ai
d,
A
ud
i
ogram
is
re
-
sam
pled
an
d
m
od
ifie
d.
T
he
pro
posed
al
gorithm
requires
a
buf
f
er
w
hich
is
ne
eded
t
o
c
om
pu
te
the
80
-
po
i
nt
DCT.
T
otal
de
la
y
is
the
delay
du
e
t
o
bu
ffer
pl
us
delay
of
t
hr
ee
m
at
rix
m
ulti
pli
cat
ion
s.
Acc
ordin
g
to
[24],
m
at
rix
m
ulti
pli
cat
ion
wit
h
le
ngth
80
ta
kes
0.8
m
s,
three
s
uc
h
m
ulti
plica
ti
on
s
are
nee
ded
f
or
the
pro
posed
al
go
rithm
,
so
total
d
el
ay
due
to
m
at
rix
m
ulti
plications
is
t
m
=2.
4
m
s.
Accor
ding
to
the
(
4
)
propose
d
al
gorithm
tak
es
5
m
s
delay
du
e
to
the
buf
fer
siz
e
of
80
at
the
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
Eff
ic
ie
nt h
eari
ng a
i
d alg
or
it
hm
us
in
g DCT
wi
th unif
or
mly
re
-
sam
pled
an
d
rec
ur
siv
el
y…
(Zac
ha
ri
ah C.
Alex
)
245
input
be
f
or
e
a
pply
ing
D
CT
a
s
show
n
i
n
Fi
g
ure
2
(
a
)
.
From
(
5
)
the
total
d
e
la
y
between
in
pu
t
a
nd o
ut
pu
t si
gn
al
is 7.
4
m
s
.
=
(4)
w
he
re
N
is
th
e
siz
e
of
the
buff
e
r
in
this
case
N=80,
a
nd
F
s
sam
pling
fr
e
qu
e
ncy=1
6000
Hz.
T
herefo
re
,
T
b
=5m
s
total
d
el
ay
.
=
+
=
5
+
2
.
4
=
7
.
4
(5)
5.
2
.
Co
m
pu
t
at
i
onal ef
fici
ency
To
re
duce
the
power
c
onsum
pt
ion
an
d
de
la
y
hear
ing
a
id
al
gorithm
s
hould
be
com
pu
ta
ti
onal
ly
eff
ic
ie
nt
[25].
In
the
pro
pose
d
al
go
rithm
DCT
an
d
ID
CT
ta
kes
m
or
e
c
om
pu
ta
ti
on
s.
O
ne
dim
ension
a
l
DCT
and
I
DCT
re
quires
2Nlo
g2(
N)
nu
m
ber
of
a
ddit
ion
s
a
nd
m
ulti
plica
ti
on
s
,
as
sh
ow
n
Fig
ure
7
[
26]
.
So,
80
-
po
i
nt
DCT
an
d
ID
C
T
ta
kes
11
20
num
ber
of
m
ulti
plica
ti
on
s
a
nd
add
it
io
ns
.
T
o
a
dju
st
the
gain
values
i
n
f
requ
ency
do
m
ai
n
80
m
ul
ti
plica
ti
on
are
nee
ded.
To
pe
rfor
m
the
pro
po
se
d
al
gorithm
on
80
sam
pl
es,
12
00
m
ul
ti
plica
ti
on
s
and
1120
a
dd
i
ti
on
s
are
re
qu
ir
ed.
T
otal
14
ad
diti
on
s
a
nd
15
m
ul
ti
plica
ti
on
s
are
nee
de
d
f
or
one
sam
ple.
In
[
19
]
,
the
num
ber
of
m
ulti
pliers
are
67
i
nclu
ding
al
l
sub
bands
,
from
this
it
is
cl
ear
that
m
ulti
plier
com
plexity
is red
uce
d by
77.61%
(a)
(b)
Figure
7. I
nput
and
ou
t
pu
t
of
pro
po
se
d h
ea
ring aid al
gorith
m
;
(a)
freq
ue
nc
y do
m
ai
n
re
presentat
ion an
d
(b)
ti
m
e d
om
ain
represe
ntati
on
6.
CONCL
US
I
O
N
In
the
prese
nt
resea
rch,
a
DCT
base
d
a
ud
it
ory
c
om
pen
sat
io
n
us
in
g
un
i
form
l
y
re
-
s
a
m
pled
a
nd
recursively
m
od
ifie
d au
diogra
m
v
al
ues
is imple
m
ented.
The pr
opose
d al
gorithm
p
r
ov
i
de
s a sim
ple so
lu
ti
on
t
o
com
pen
sat
e
for
the
hear
i
ng
l
os
s
with
out
an
y
filt
er
ban
ks
a
nd
sam
pling
ra
te
con
ve
rsions
.
Total
ly
three
sta
ges
are
nee
de
d
f
or
the
w
ho
le
pr
ocess
:
i)
Fin
di
ng
D
CT
f
or
th
e
input
au
dio
sign
al
,
ii
)
Gain
ad
justm
ent
a
nd
ii
i)
Inverse
DCT.
DCT
coe
ff
ic
ie
nts
of
the
a
ud
i
o
sign
al
are
m
ulti
plied
with
un
i
form
l
y
re
-
sa
m
pled
and
re
cur
si
vely
m
od
ifie
d
au
dio
gram
values
to
ad
j
us
t
the
gains
in
fr
e
qu
ency
dom
ai
n.
The
pe
rfo
rm
a
nce
of
the
propose
d
al
gorithm
is
com
par
ed
with
diff
e
ren
t
ty
pes
o
f
filt
er
banks
nam
el
y
un
if
or
m
filt
er
ban
k
no
nunif
or
m
filt
er
bank
,
var
ia
ble
filt
er
bank
an
d
rec
onfi
gurab
le
filt
er
ba
nk
str
uctu
r
es.
The
pr
opose
d
al
gorithm
i
s
te
ste
d
fo
r
dif
fer
e
nt
ty
pes
of
hea
rin
g
loss
cases
li
ke
m
il
d
hear
in
g
loss
at
high
fr
e
qu
e
ncies,
m
il
d
to
m
od
erat
e
hear
in
g
loss
at
low
fr
e
qu
e
ncies,
m
od
e
rate
hea
rin
g
loss
at
m
idd
le
fr
e
qu
e
ncies
a
nd
m
il
d
conduct
ive
hear
i
ng
l
os
s.
From
the
abov
e
te
st,
it
is
i
ll
us
trat
ed
that
the
propose
d
DCT
ba
sed
al
gorithm
pr
ovid
es
bette
r
m
at
ching
bet
ween
the
f
re
quenc
y
respo
ns
e
of
he
arin
g
ai
d an
d
a
ud
i
ogram
. I
t i
s
achieve
d wit
h m
ini
m
u
m
d
el
ay
an
d com
pu
ta
ti
on
al
co
m
plex
it
y
.
ACKN
OWLE
DGE
MENTS
The
resea
rch
le
adin
g
t
o
these
res
ults
recei
ve
d
fun
ding
f
r
om
Scie
nce
for
Eq
uity
,
Em
po
wer
m
ent
an
d
Dev
el
op
m
ent
Divisio
n
unde
r
Tec
hnol
og
y
In
te
r
ve
ntio
ns
f
or
Disa
bled
a
nd
Elde
rly
(T
I
DE),
Dep
a
rtm
ent
of
Scie
nce a
nd Te
chnolo
gy, G
overn
m
ent o
f
In
di
a u
nde
r Gra
nt
Agreem
ent N
o SEE
D/TI
DE/0
15
/
2017/G
.
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,
Vo
l.
23
, N
o.
1
,
Ju
ly
2021
:
2
3
7
-
2
4
6
246
REFERE
NCE
S
[1]
N.
Ti
war
i
and
P.
C.
Pand
e
y
,
“
Slidi
ng
-
band
d
ynamic
ra
ng
e
co
m
pre
ss
ion
for
use
in
he
ari
ng
aids
,
”
Int
ernati
on
al
Journal
of
Spe
ech Tec
hnolog
y
,
v
ol.
22
,
no
.
4
,
pp
.
911
-
926,
2
019
,
doi:
10
.
1007/s10
772
-
019
-
09635
-
4
.
[2]
W
.
B.
Alshuaib,
J.
M.
Al
-
Kanda
ri,
and
S.
M.
Ha
san,
“
Cla
ss
ifi
cat
ion
of
hea
ring
lo
ss
,
”
In
Update
on
Hearing
Los
s
,
p.
29
,
2015
.
[3]
Y.
D.
H
y
t
er
and
M.
B.
Sala
s
-
Pr
ovanc
e
,
Cult
ur
all
y
r
esponsive
pr
ac
t
ic
es
in
spe
ech,
la
ngu
age,
and
hea
ring
sc
ie
nc
e
s.
Pl
ural
Publ
ishin
g
,
2018
.
[4]
M.
C.
Kill
ion
and
S.
Fikret
-
P
asa
,
“
The
3
t
ypes
of
sensorine
ura
l
h
ea
ring
l
oss
:
Loud
ness
and
int
e
ll
ig
ibi
l
i
t
y
conside
ra
ti
ons,
” Hear
ing journa
l,
vol.
46
,
pp
.
31
–
3
1,
1993
.
[5]
A.
R.
Mølle
r
,
H
e
ari
ng:
anatom
y
,
ph
y
siolog
y
,
and disorders
of the audi
tor
y
s
y
st
em.
Pl
ural
Publishi
ng
,
2012
.
[6]
Y.
Li
m
,
“
A di
git
al
fil
t
er
bank
for digit
a
l
audi
o
s
y
s
te
m
s
,”
IEEE
Tr
a
nsacti
ons on
Circui
ts and
Syste
m
s
,
vol.
33,
no.
8,
pp.
848
–
849
,
19
86
,
doi
:
10
.
1109
/T
CS
.
1986.
1085
988
.
[7]
T.
Lunne
r
and
J.
Hell
gre
n,
“
A
digi
ta
l
fi
lt
e
rba
nk
h
ea
ring
a
id
-
desig
n,
implementa
t
io
n
and
eva
luation
,
”
in
ICASSP
91:
1991
Inte
rnat
ion
al
Conf
er
enc
e
o
n
Ac
oust
ic
s,
Spe
ec
h,
and
S
ignal
Proce
ss
ing.
IE
E
E
,
1991
,
pp
.
366
1
–
3664
.
[8]
E.
Onat,
M.
Ah
m
adi
,
G.
Julli
en
,
and
W
.
Mill
er
,
“
Optimize
d
del
a
y
cha
r
ac
t
eri
st
i
cs
for
a
hea
ring
instrumentfi
lter
bank,
”
in
Proc
e
edi
ngs
of
the
43rd
IEE
E
Midwe
st
Symposium
on
Circu
it
s
and
Syste
ms
(
Cat.
No.
C
H37144)
,
vol.
3
,
2000,
pp
.
1074
–
1077
,
doi
:
10
.
11
09/MW
SC
A
S.2000.
951401
.
[9]
Y.
W
ei
and
Y.
Li
an
,
“
A
computat
ion
al
l
y
ef
ficie
nt
non
-
uniform
digi
tal
f
ir
f
il
t
er
bank
for
hea
r
in
g
ai
d
,
”
in
IEEE
Inte
rnational
Workshop
on
Bi
o
medic
al
C
ircui
t
s
and
Syste
ms
,
2004,
p
p.
S1
–
3
,
doi:
10
.
1109/BI
OCA
S.2004.
1454116.
[10]
Y.
Li
a
n
and
Y.
W
ei
,
“
A
computat
ion
al
l
y
eff
ic
i
ent
nonuniform
fir
digita
l
filte
r
bank
for
h
ea
r
ing
ai
ds,
”
IE
E
E
Tr
ansacti
ons
on
Circui
ts
&
Syste
ms
I:
Re
gular
P
aper
s
,
vol.
52,
n
o.
12,
pp.
2
75
4
–
2762,
200
5
,
doi:
10
.
1109/T
C
SI.2005.
857871
.
[11]
Y.Wei
and
Y.
Li
an
,
“
A
16
-
ba
nd
nonuniform
fir
digi
tal
fil
t
er
bank
for
hea
rin
g
ai
d,
”
in
2006
IEE
E
Bi
omed
i
cal
Circui
ts and
Sys
te
ms
Confe
ren
ce
,
2006
,
pp
.
186
–
189
.
[12]
T.
-
B.
Deng
,
“
Thre
e
-
ch
annel
va
ria
bl
e
fil
t
er
-
ban
k
for
digi
ta
l
he
ari
ng
ai
ds,
”
IET
signal
proce
ss
ing
,
vol.
4,
no.
2,
pp.
181
–
196
,
20
10
,
doi
:
10
.
1109
/ISP
ACS
.
2009.
4806730
.
[13]
N
.
I
t
o
a
n
d
T
.
-
L
.
D
e
n
g
,
“
V
a
r
i
a
b
l
e
-
b
a
n
d
w
i
d
t
h
f
i
l
t
e
r
-
b
a
n
k
f
o
r
l
o
w
-
p
o
w
e
r
h
e
a
r
i
n
g
a
i
d
s
,
”
i
n
2
0
1
0
3
r
d
I
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
g
r
e
s
s
o
n
i
m
a
g
e
a
n
d
s
i
g
n
a
l
p
r
o
c
e
s
s
i
n
g
,
v
o
l
.
7
,
2
0
1
0
,
p
p
.
3
2
0
7
–
3
2
1
1
[14]
R.
Mahe
sh
and
A.
P.
Vinod,
“Re
conf
igur
abl
e
low
are
a
complexi
t
y
filter
bank
arc
hitect
ur
e
ba
sed
on
fre
quency
response
m
asking
for
nonunifor
m
cha
nneliz
at
io
n
in
softwar
e
r
ad
io
re
ce
iv
ers,
”
IE
EE
Tr
ansacti
ons
on
Ae
ros
pace
an
d
El
e
ct
ronic
Syst
e
ms
,
vol. 47, no.
2,
pp
.
1241
–
125
5,
2011
,
doi
:
10
.
1109/T
AES.201
1.
5751255
.
[15]
Y.
W
ei
and
D.
Li
u
,
“
A
rec
onf
igura
bl
e
dig
it
a
l
fil
terbank
for
h
ea
ring
-
ai
d
s
y
ste
m
s
with
a
var
iet
y
of
soundw
av
e
dec
om
positi
on
pla
ns,”
I
E
EE
tr
ansac
ti
ons
on
Bi
omedi
cal
Eng
ine
ering
,
vol.
6
0,
no.
6,
pp.
1
628
–
1635,
2013
,
doi:
10
.
1109/T
B
ME.
2013.
22406
81.
[16]
Am
ir,
A.,
Rak
esh
Ina
ni
,
and
El
i
za
be
th
E
li
a
s.
,
“
Rec
onfigur
abl
e
low
complexi
t
y
hea
r
ing
ai
d
s
y
s
te
m
usin
g
adj
ustablefi
lt
er
bank,
”
in
2016
IEE
E
Re
gion
10
C
onfe
renc
e
(
TENCON
)
,
2016,
pp.
2684
–
2688
,
doi:
10
.
1109/TE
NCO
N.2016.
7848526
.
[17]
A.
Am
ir,
T
.
Bi
ndi
y
a,
and
E.
E
li
as,
“
Design
an
d
implementatio
n
of
re
conf
igurable
fi
lter
bank
struct
ur
e
for
low
complexi
t
y
h
ea
r
ing
ai
ds
using
2
-
le
ve
l
sound
wave
de
compos
it
i
on,
”
B
iomedical
Signal
Proc
essingand
Control
,
vol.
43
,
pp
.
96
–
1
09,
2018
,
doi
:
10
.
1016/j.bspc.
201
8.
02.
020
.
[18]
Y.
W
ei
,
T
.
Ma
,
B.
K.
Ho,
and
Y.
Lian,
“
The
d
esi
gn
of
low
-
power
16
-
band
nonun
i
form
fil
te
r
bank
for
hearing
ai
ds,
”
IEE
E
transact
i
ons
on
biom
edi
ca
l
ci
rcui
ts
and
systems
,
vol.
13,
n
o.
1,
pp.
1
12
–
123,
2018
,
doi:
10
.
1109/T
B
CAS
.
2018.
2888860
.
[19]
A.
Am
ir,
T.
Bin
di
y
a,
and
E.
El
i
a
s,
“
Low
-
complexi
t
y
implementation
of
eff
i
cient
rec
onfigur
able
struct
ure
for
cost
-
eff
ective
he
ari
n
g
ai
ds
using
fra
ct
ion
al
int
erp
o
la
ti
on,
”
Computer
s
&
El
ec
tric
al
E
ngine
ering
,
vol
.
74,
pp.
391
–
41
2
,
2019
,
doi
:
10
.
10
16/j
.
compelece
n
g.
2019.
02
.
008
.
[20]
T.
Ma,
C
.
Shen,
and
Y.
W
ei
,
“
Adjustabl
e
f
il
t
er
b
ank
design
for
h
ea
ring
ai
ds
s
y
ste
m
,
”
in
2019
IEEE
Inte
rnationa
l
Symposium on
C
ircui
ts and
Sy
ste
ms
(
ISCAS
)
,
2019,
pp
.
1
–
5
,
doi
:
1
0.
1109/ISCAS
.
2019.
8702314.
[21]
R.
S.
Rashid
a
nd
J.
R.
Moham
m
ed,
“Sec
urin
g
spee
ch
signals
by
wa
te
rm
ark
ing
bina
r
y
imag
es
in
the
wave
l
e
t
dom
ai
n,
”
Indon
esian
Journal
of
Elec
tri
cal
E
ngine
ering
and
Computer
Scienc
e
(
IJEECS)
,
v
ol.
18
,
no.
2,
pp.
1096
–
1103,
2
020
,
doi
:
10
.
115
91/i
jeec
s.v18
.
i2
.
pp1096
-
1103
.
[22]
M.
C.
Fl
y
nn
,
R.
C.
Dow
el
l,
and
G.
M.
Cla
rk,
“
Aided
spee
ch
rec
o
gnit
ion
ab
il
i
ti
es
of
adul
ts
with
a
seve
reo
r
seve
r
e
-
to
-
profound
hearing
loss,”
Jour
nal
of
spee
ch
,
l
anguage,
and
h
earing
research
,
vol.
41,
no
.
2,
p
p.
285
–
299,
199
8
,
doi:
10
.
1044/j
slh
r.
4102.
285
.
[23]
Y.
W
ei
and
Y.
W
ang,
“
Design
of
low
complexit
y
adj
ust
abl
e
f
il
t
er
bank
for
p
ersona
lized
he
ari
ng
ai
d
soluti
ons
,
”
IEE
E
/A
CM
Tr
ansacti
ons
on
audio,
spee
ch
,
a
nd
language
proce
ss
i
ng
,
vol.
23,
no.
5,
pp.
923
–
931,
2015
,
doi:
10
.
1109/T
A
SLP
.
2015.
24097
74
.
[24]
P.
Saha,
A.
Banerje
e
,
P.
Bhat
t
ac
h
ar
y
y
a
,
and
A.
Danda
pat,
“
Im
prov
ed
m
at
rix
m
ult
ip
li
er
design
for
hi
gh
spee
d
digi
t
a
l
signal
pro
ce
ss
in
g
applications,”
IET
ci
rcu
it
s,
devic
es
&
systems
,
vol.
8
,
no
.
1
,
pp
.
27
–
37,
2014
.
[25]
S.
As
hra
f,
M.
Gao,
Z.
Chen
,
H.
Nae
em,
A.
Ahm
ad,
and
T.
Ahm
ed,
“
Under
wate
r
pra
gm
at
i
c
routi
ng
appr
oa
ch
through
pac
k
et
rev
erb
er
at
ion
m
ec
hani
sm
,
”
IEE
E
A
ccess
,
vol.
8,
p
p.
163
091
–
163
114,
2020
,
doi:
10
.
1109/AC
CESS
.
2020.
3
02
2565.
[26]
M.
St´
epha
ne,
“
Chapt
er
8
-
wav
elet
pac
k
et
and
l
oca
l
cosin
e
bas
es,
”
A
Wav
elet
Tour
of
Signal
Proce
ss
ing(
Thir
d
Edi
ti
on)
,
pp
.
377
–
434
,
2009
.
Evaluation Warning : The document was created with Spire.PDF for Python.