Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol
.
4
,
No
. 5, Oct
o
ber
2
0
1
4
,
pp
. 64
3~
64
7
I
S
SN
: 208
8-8
7
0
8
6
43
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
Voice-Based Door Access Co
ntrol System Using the Mel
Frequency Cepstrum Coefficients
and Gaussian Mixture Model
Ka
yo
de Fr
anc
i
s Aki
n
gb
ade,
Oko
k
o
M
k
p
o
uto
U
m
a
n
n
a
,
Isi
ak
a Ajew
al
e
Al
i
m
i
Department o
f
Electrical and
Elec
tronics Engin
e
ering, Schoo
l of
Engi
neering
and
Engin
eering
Technolog
y
,
Federal University
of
Te
chnolog
y
,
Akure, Nig
e
ria
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
J
u
n 12, 2014
Rev
i
sed
Au
g
20
, 20
14
Accepted Aug 26, 2014
Access to an ar
ea or env
i
ronment ca
n be con
t
rolled b
y
conventional
and
electronic key
s
,
identity
cards, p
e
rsona
l id
entif
ication numbers (
P
INs) pads
and sm
artcards. Due to certain
lim
itat
i
ons of existing door acce
ss
schem
e
s
deplo
y
ed for sec
u
rit
y
in building
s
, this
paper presents speaker recognition fo
r
building s
ecuri
t
y
as
a bett
er m
eans
of adm
i
s
s
i
on into im
portant
places
. Th
is
is proposed due mainly
to
th
e
fact th
at speech
canno
t be stolen,
copied
,
forgotten
,
lost
or guessed with accuracy
. This paper, th
erefo
r
e presen
ts
design of an af
fordable voice
activat
ed door control s
y
s
t
em for buildin
g
security
. The pr
oposed s
y
stem uses
the Mel Frequency
Cepstr
um and the
Gaussian Mixtur
e Model for
feature ex
traction
an
d template patter
n matching
res
p
ect
ivel
y.
Th
e anal
ys
is
of the res
u
lt which is
bas
e
d on the fals
e
accep
tanc
e
and re
je
ction
ra
t
e
s
indi
cat
e
a s
y
s
t
em
ac
curac
y
of
m
o
re than
80%.
Keyword:
Access c
o
ntrol
PINs
Security
Sm
artcard
Voice
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Kayode Fra
n
cis Akingbade
,
Depa
rt
m
e
nt
of
El
ect
ri
cal
and
El
ect
roni
cs
E
n
gi
nee
r
i
n
g,
Sch
ool
o
f
E
ngi
neeri
n
g
an
d E
n
gi
nee
r
i
n
g Tec
h
nol
ogy
,
Fede
ral
U
n
i
v
e
r
si
t
y
of Tec
h
nol
ogy
,
A
k
ure
,
Ni
geri
a
Em
a
il: k
f
ak
ingb
ad
e@fu
ta.edu.ng
1.
INTRODUCTION
The im
ple
m
entation of acce
ss control pre
v
ents una
u
thorized indivi
dua
ls to access s
ecure area
s
,
b
u
ild
i
n
gs,
d
o
c
u
m
en
ts an
d
serv
ices. Th
e con
t
ro
l system
co
n
s
ists o
f
two
m
a
in
stag
es
n
a
m
e
ly, th
e id
en
tificatio
n
an
d v
e
rification
stag
es. Peop
l
e
th
at wan
t
to
access a secu
re
facilities in
tro
d
u
ce th
em
selv
es to
th
e
system
i
n
th
e
id
en
tificatio
n stag
e and
th
e v
e
rificatio
n
stag
e ch
eck
th
e
v
a
lid
ity o
f
t
h
e id
entities o
f
th
e in
t
r
odu
ced
u
s
ers.
If t
h
e
identity of the
user is
va
lid, then the
user may access sec
u
re a
r
ea
with
t
h
e
assigne
d
permissions. The
access
cont
rol
sy
st
em
i
s
used f
o
r
nu
m
e
rous a
ppl
i
c
a
t
i
ons suc
h
as f
o
r l
o
g
g
i
n
g o
n
ATM
m
achi
n
es, e-ba
n
k
i
n
g ac
cou
n
t
s
or for physical security of a
ro
om
or
bui
l
d
i
n
g
as a
w
hol
e
[1]
.
Access c
o
ntrol for
buildings
is an esse
ntial de
vi
ce for protecting im
portant places i
n
the buildi
ng
th
at h
a
v
e
v
a
luab
le o
r
h
i
gh
ly sen
s
itiv
e m
a
ter
i
als. Serv
er and
strong
ro
o
m
o
f
b
a
nk
s are i
m
p
o
r
tan
t
areas wh
ere
ex
tr
em
ely
ef
f
ectiv
e co
n
t
ro
l syste
m
is r
e
q
u
i
red
.
Th
er
e w
a
ys o
f
secur
ity i
m
p
l
em
en
tatio
n
i
n
a b
u
ild
i
n
g
and
door
access control is an integral part of
t
h
em
. The door acces
s
cont
rol is a m
eans
of sec
u
ri
ng building
by giving
limited access to specific pe
ople and
by
kee
p
ing rec
o
rds of such accesse
s [2].
Sm
artcard
according to [2, 3] is
t
h
e
m
o
st
co
m
m
on aut
h
ent
i
cat
i
on m
e
t
hod
f
o
r t
h
e d
o
o
r acc
ess cont
r
o
l
s
. It
has been
obse
r
ve
d t
h
at
a card-
bas
e
d
access system
can
only c
o
ntrol the
access
of a
u
thorize
d
ca
rd
s
that a
r
e
pie
ces of
plastic,
but
not the
owners
hip
o
f
th
e card
.
It
can
b
e
u
s
ed
illeg
iti
m
a
tel
y
b
y
an un
au
tho
r
ized
p
e
rson
wh
en
i
n
p
o
s
session
o
f
it. Fu
rt
h
e
rm
o
r
e,
sy
st
em
s usi
ng
PIN
s
re
qui
re i
ndi
vi
d
u
al
t
o
ent
e
r spe
c
i
f
i
c
nu
m
b
ers to
g
a
i
n
en
try bu
t the sh
ortco
m
in
g is th
a
t
t
hose
w
h
o real
l
y
ent
e
rs t
h
e c
o
des ca
n
not
be
det
e
rm
i
n
ed sy
s
t
em
.
Th
e limitatio
n
s
o
f
con
v
e
n
tional secu
rity syste
m
s call
for
be
t
t
e
r one
s. T
h
er
e are va
ri
et
i
e
s of
bi
om
et
ri
c
m
e
thods that c
oul
d be em
ployed in access c
ont
rol system
f
o
r
verification
of a
u
thorize
d
person int
o
im
p
o
rta
n
t
or se
nsitive places. An a
u
tom
a
t
i
c verification of ident
ity
in ter
m
s
of
beha
vioral
and/or physiological
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 5
,
O
c
tob
e
r
20
14
:
643
–
6
47
64
4
charact
e
r
i
s
t
i
c
s of a pe
rs
on i
s
carri
ed
out
i
n
t
h
e bi
om
et
ri
c
m
e
t
hods [
2
,
3]
. The
bi
om
et
ri
c devi
ce i
d
e
n
t
i
f
i
e
s
people by cert
a
in unique feat
ures su
c
h
as the finge
rprint,
voice, face a
nd
eye (iris). Additi
onally, the device
can elim
inate the
need for
ca
rd-ba
s
ed acces
s syste
m
. In t
h
e light of
t
h
is, bi
om
etric devices ca
n re
duce the
need for reiss
u
e of lost
or
da
maged
ca
rds
as
the
fingerpri
n
t,
voice,
face a
n
d eye a
r
e ra
rel
y
stolen
or lost
.
The a
dva
ntage
s
of voice as a
biom
etrics
m
e
thod a
r
e exp
a
tiated
in
[2
] am
o
n
g
wh
ich are si
m
p
l
i
city fo
r
the
use
r
, spee
d
of aut
h
entication and
level of
false
-re
j
ection
rate. To resolv
e prob
lem
s
o
f
th
e
PINs
p
a
ds and
sm
artcards-bas
ed
door access
control
,
this
paper
pres
e
n
ts
voice
-base
d
door ac
cess c
ont
rol system
using the
M
e
l
Freq
ue
ncy
C
e
pst
r
um
and
Gaus
si
an m
i
xt
ure
m
odel
fo
r
bui
l
d
i
n
g
sec
u
ri
t
y
.
The pa
pe
r i
s
o
r
ga
ni
zed as
fol
l
ows.
Sect
i
on
2 desc
ri
bes t
h
e
pr
op
ose
d
sy
st
em
. Sect
i
on 3
foc
u
ses
on
sy
st
em
desi
gn
and i
m
pl
em
ent
a
t
i
on. R
e
s
u
l
t
s
and
per
f
o
r
m
a
nce eval
uat
i
on a
r
e di
sc
u
ssed i
n
s
ect
i
o
n 4
.
C
oncl
u
si
o
n
s a
r
e d
r
aw
n i
n
sect
i
on
5.
In
t
h
e
fo
llowi
n
g
section
s
,
we will qu
ick
l
y go
throug
h featu
r
e ex
tracti
o
n and
Gau
ssi
an
Mix
t
ure
M
odel
.
Ne
xt
,
we l
o
o
k
at
t
h
e
ope
rat
i
o
n a
n
d
i
m
pl
em
ent
a
t
i
on of
t
h
e
v
o
i
ce b
a
sed
do
o
r
c
ont
rol
sy
st
em
and
fi
nal
l
y
,
we prese
n
t perform
a
nce
eval
uation
and res
u
lts.
2.
PROP
OSE
D
SYSTE
M
Research in s
p
eaker rec
o
gniti
on a
nd s
p
eec
h recogn
itio
n
is p
r
esen
tly
m
a
tu
re. Speak
er reco
gn
itio
n
is
essentially used in access c
o
ntrol syste
m
s t
o
gi
ve acce
ss t
o
individuals
whose ide
n
tities are validate
d
from
t
h
ei
r p
r
e
v
i
o
usl
y
st
ore
d
v
o
i
c
e
reco
r
d
s
or m
odel
s
. T
h
i
s
i
n
vol
ves
bot
h s
p
eake
r
i
d
ent
i
f
i
cat
i
on a
n
d
sp
eaker
verification [4]. It is, howe
ver, di
ffere
n
t from speech rec
o
gnition which
re
lies on the share
d
cha
r
acte
r
istic of
wh
at
is said and
wh
at
is stored
i
n
o
r
d
e
r to mak
e
a
decisi
on. Both are
em
ployed
in
sp
eak
e
r id
en
tificatio
n and
veri
fi
cat
i
o
n sy
st
em
s [5]
.
Thi
s
pape
r us
es a t
e
xt
i
nde
pen
d
e
n
t
spea
ker i
d
e
n
t
i
f
i
cat
i
on a
n
d
veri
fi
cat
i
o
n p
r
oces
s
wh
ere t
h
e
p
h
rase or
word to be said
is
no
t kno
wn
t
o
th
e syste
m
.
The
desi
g
n
i
s
im
pl
em
ent
e
d i
n
t
w
o
part
s
nam
e
l
y
t
h
e soft
ware a
n
d t
h
e
har
d
wa
re
p
a
rt
s. F
o
r
t
h
e
soft
ware
, we
use the Mel Fre
que
ncy Cepst
r
al Coeffici
ents
(MFCCs) for
feature e
x
tracti
on a
n
d the
Ga
ussian
M
i
xt
ure M
o
de
l
(GM
M
)
f
o
r
t
e
m
p
l
a
t
e
m
a
t
c
hi
ng.
W
e
use
MFCCs beca
use they a
r
e
ve
ry robust a
n
d
are the
dom
inant features use
d
for speech rec
o
gnition
[6].
Al
s
o
,
G
M
M
s
are usual
l
y
prefer
red
be
cause t
h
ey
of
fe
r hi
g
h
classification accuracy whil
e
still
bei
ng robust to c
o
rruptions
in the
speec
h signal. Also, t
h
ey are very
success
f
ul
when it c
o
m
e
s to
noise
ha
ndling. T
h
is
ha
s
led to the
exte
nsive
use
of
GMM ba
sed s
p
eake
r
reco
g
n
i
t
i
on sy
s
t
em
s. The
har
d
ware
pa
rt
use
s
suc
h
com
p
o
n
en
ts as
d
.
c. m
o
to
rs, th
e L293
B
H-B
r
idg
e
in
teg
r
ated
circu
it, a
p
a
rallel p
o
rt an
d th
e
d
oor stru
cture.
2.
1. Fea
t
ure
E
x
tr
acti
on
The intenti
on
here is to
have a
m
odel of
the sp
eec
h wa
veform
that is
sufficiently an accurate
represe
n
tation
to the s
p
eech. It has
bee
n
observe
d
that
t
h
e s
p
eech si
gnal is a slowly time varying signal
(qu
a
si-station
a
ry). Th
is m
ean
s th
at wh
en ob
serv
ed
ov
er
a su
fficien
tly sh
ort p
e
riod
o
f
ti
m
e
(b
etween 5
and
10
0 m
s
), i
t
s
charact
eri
s
t
i
c
s are fai
r
l
y
st
at
i
onary
but
cha
n
ge
ove
r l
o
n
g
pe
ri
ods
(0
.2s
or m
o
re
) i
n
o
r
de
r t
o
refl
ect
t
h
e di
ffe
rent
s
o
u
n
d
s
bei
n
g s
p
oke
n.
T
h
ere
f
o
r
e, t
o
c
h
a
r
act
eri
ze t
h
e s
p
eec
h
si
gnal
,
t
h
e M
e
l
Fre
que
ncy
C
e
pst
r
al
Co
efficien
ts (MFCCs) wh
ich
is a to
o
l
fo
r sh
ort ti
m
e
sp
ectral an
alysis is e
m
p
l
o
y
ed
.
We refer to
[6]
for a
com
p
l
e
t
e
descri
pt
i
o
n o
f
t
h
e
pr
oce
d
u
r
es f
o
r
obt
ai
ni
ng t
h
e
M
F
C
C
s
feat
ures. I
n
t
h
i
s
w
o
rk
, t
h
e p
r
o
g
r
a
m
m
i
ng
pl
at
fo
rm
used
f
o
r
v
o
i
ce
pr
oces
si
ng
an
d s
o
ft
w
a
re
devel
opm
ent
i
s
M
A
TLA
B
.
2.
2. Ga
ussi
a
n
Mi
x
t
ure M
o
d
e
l
In [
7
-1
0]
, a G
a
ussi
an M
i
xt
ur
e M
odel
i
s
de
scri
be
d as a w
e
i
ght
ed s
u
m
of M
com
pone
nt
Gau
ssi
an
d
e
nsities g
i
v
e
n b
y
th
e equ
a
tion
,
|
|
,
W
h
er
e
i
s
a
D-
di
m
e
nsi
onal
co
nt
i
n
uo
us
-v
al
ued
dat
a
v
e
ct
or
(m
easure
m
ent
o
r
feat
u
r
es)
,
,
1,
…
,
,
are
t
h
e m
i
xture
weights, a
nd
|
,
,
1
,…,
,
are the com
ponent Ga
us
sian de
nsities with
mean vectors
a
n
d
cov
a
r
i
an
ce
ma
tr
ic
e
s
.Each com
p
onent
density is D-variate Gaus
sian function
of t
h
e
fo
rm
,
|
,
1
2
∑
2
⁄
|
|
⁄
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Voice-Base
d
Door Access Control
Sy
stem Using the Mel
Frequency
Ceps
trum C
o
efficients … (Kay
ode
FA)
64
5
The m
i
xture
weights satisfy the constrai
nt that
∑
1
. Th
e
p
a
ram
e
ters o
f
th
e co
m
p
lete
Gau
s
sian
m
o
d
e
l are co
llectiv
ely rep
r
esen
ted
b
y
th
e
no
tatio
n,
,
,
1
,
…
,
.
In
train
i
ng
th
e GMM, th
ese p
a
ram
e
ters are esti
m
a
ted
su
ch
th
at th
ey b
e
st
m
a
tch
th
e d
i
strib
u
tion
of
t
h
e t
r
ai
ni
ng
ve
ct
ors
[F
uzzy
m
i
xt
ure
M
o
del
f
o
r
Sp
eake
r
R
e
c
o
g
n
i
t
i
on]
.
3.
SYSTE
M
DESIGN AND I
M
PLEME
N
T
A
TIO
N
The
pr
ocess
b
e
gi
ns
wi
t
h
t
h
e
reco
rdi
ng a
n
d
t
r
ai
ni
ng
of
v
o
ice sa
m
p
les o
t
h
e
rwise called en
ro
lm
en
t,
whic
h could
be done either i
n
real tim
e
or using a pr
e-rec
o
rde
d
sam
p
le.
A data
base for each of these
sam
p
les
exists s
u
ch tha
t
any
newly
re
corde
d
s
p
eec
h woul
d
be sa
ved
there and
no
t b
e
lo
st
either
b
e
fo
re or after th
e
reco
g
n
i
t
i
on p
r
ocess. F
o
r
opt
im
al resul
t
s
as in
th
is case,
it is
v
e
ry i
m
p
o
r
ta
nt that the recorde
d
spee
ch be
obtaine
d through the sam
e
means and if possi
ble, pr
ocesses every tim
e
. This
is
because the intrinsic
p
r
op
erties of differen
t
micro
p
h
o
n
e
s
v
a
ry and co
u
l
d
g
r
ea
tly affect th
e qu
ality o
f
th
e sig
n
a
l
an
d
th
e
recogn
ition
syste
m
in gene
ral.
It is in
t
h
is process th
at th
e an
alog
u
e
sp
eech
sign
al
i
s
c
o
n
v
ert
e
d t
o
a di
gi
t
a
l
si
gnal
by
sa
m
p
li
ng.
Th
e
an
alog
u
e
sign
al is cond
itio
n
e
d
with
an
ti-aliasin
g
filtering
(and
add
itio
n
a
l
filtering
if requ
ired
to co
m
p
en
sate
for an
y ch
an
nel i
m
p
a
ir
m
e
n
t
s). Th
e an
ti-aliasin
g
filter li
m
its
th
e b
a
nd
wi
d
t
h o
f
th
e sign
al to
app
r
ox
im
a
t
el
y th
e
Nyqu
ist rate (half th
e sam
p
lin
g
rate)
b
e
fore sa
m
p
lin
g
.
Th
is
d
i
g
itized
sp
eech
is th
en
fu
rth
e
r an
alyzed
t
o
ex
tract
th
e features that wou
l
d
b
e
u
s
ed
fo
r th
e
recog
n
ition
algo
rit
h
m
.
Fig
u
r
e
1
sh
ows th
e series o
f
p
r
o
cesses
th
at th
e
voice
sam
p
les would undergo for a typical c
a
se whe
r
e a
ve
rified I.D is enro
lled and its
m
o
d
e
l is sub
s
eq
u
e
n
tly
com
p
ared
with the
features
of a claim
e
d I.D.
Th
e h
a
rdw
a
re o
f
th
is pro
j
ect
is d
e
sig
n
e
d
and
bu
ilt u
s
in
g
a si
m
p
le d
o
o
r pro
t
o
t
yp
e m
a
d
e
w
ith
w
ood
(pl
y
w
o
od
)
hav
i
ng t
w
o
DC
m
o
t
o
rs.
The
DC
m
o
t
o
rs are
l
i
g
ht
wei
g
ht
a
n
d c
ons
um
e l
e
ss p
o
we
r,
w
h
i
c
h
i
m
pli
e
s
that the batteries would last
m
u
ch longe
r.
These m
o
to
rs provide
the ne
eded rota
tional
displacem
ent
for the
door to open a
nd they are controlled by
an
H-Bri
d
ge IC (
L
2
93B
). T
h
is IC
is in
tu
rn
driv
en
d
i
rectly b
y
th
e
parallel port of the system
connecte
d
via a
para
llel p
o
rt cable an
d con
t
ro
lled
thro
ugh
M
A
TLAB.
Fi
gu
re 1.
O
p
er
at
i
ons o
n
a typi
cal analogue si
gnal
Verified I
D
Feature
Extr
acti
on
Enrollment
Pattern
Ma
tchin
g
Claimed ID
Speaker
Mo
dels
Feature
Extr
acti
on
Filtering
and A/D
Fea
t
ure
Ex
tra
c
tio
n
Pattern
Matching
Decision
Clai
m
e
d
I
D
Verified
ID
Enroll
m
e
n
t
Speak
e
r
Models
Accept
Rejec
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
I
J
ECE Vo
l. 4
,
N
o
. 5
,
O
c
tob
e
r
20
14
:
643
–
6
47
64
6
4.
R
E
SU
LTS AN
D PER
F
ORM
ANC
E EVALU
A
T
ION
A to
tal
o
f
seven
(7)
vo
ice sa
m
p
les fro
m
ten
(10
)
diffe
re
nt s
p
eake
r
s t
h
at are
recorde
d
through the
sam
e
process
a
n
d at a sam
p
ling rate
of
88.2KHz
i
s
used
f
o
r t
h
e
pe
rf
orm
a
nce e
v
al
uat
i
o
n.
Si
nce t
h
i
s
sy
st
em
i
s
not
a t
e
xt
-
d
epe
nde
nt
sy
st
em
, the
voi
ce sam
p
l
e
s are
vari
e
d
f
r
o
m
nam
e
s t
o
n
u
m
b
ers de
pe
nd
i
ng
o
n
t
h
e c
h
oi
ce o
f
the speake
r
.
Furt
herm
ore, i
n
assessing t
h
e
syste
m
pe
rformance with re
spect to
accuracy and reliabi
lity, we
use the
false a
ccept rate a
n
d
the false
reject
rate. T
h
ere
f
ore, out of te
n (10)
ve
rification
trials each for
every
i
ndi
vi
dual
set
,
≡
13.27%
This inva
riantly
m
eans that the ge
nui
ne
acce
ptance
proba
b
ility of the syste
m
is;
100
1
3.27
.
%
The figures obtained for the
FAR an
d Ge
n
u
ine Acce
pt Rate (GAR)
of
t
h
i
s
s
y
s
t
em
cl
ea
rl
y
i
ndi
c
a
t
e
s
th
at th
e system b
a
sed
on
th
is
t
e
st, has
an efficiency of m
o
re
th
an 80
% so far.
Similarly,
≡
18.5%
The lower t
h
e
False reject
rate, the hi
ghe
r the effi
ciency of
any biom
etric
syste
m
. Ad
d
itio
n
a
lly, th
is
test also prove
s
the efficacy of the gi
ven sy
ste
m
. Th
e perform
a
nce of the
Autom
a
ted Speaker Recognition is
summ
arized in Table
1.
Tabl
e
1.
Ge
ner
a
l
Perf
o
r
m
a
nce A
u
t
o
m
a
t
e
d Sp
eaker
R
eco
gni
t
i
on
FAR (%
)
FRR (%)
Speaker
1
20
10
Speaker
2
0
20
Speaker
3
30
0
Speaker
4 10
20
Speaker
5
50
40
Speaker
6
20
30
Speaker
7
10
10
5.
CO
NCL
USI
O
N
Thi
s
pa
per
has
descri
be
d t
h
e
desi
g
n
of a
v
o
i
ce act
i
v
at
ed do
o
r
co
nt
rol
s
y
st
em
. W
e
ha
ve use
d
t
h
e
MFCCs fo
r featu
r
e ex
traction
wh
ile th
e GMM is u
s
ed
fo
r
p
a
ttern
m
a
t
c
h
i
ng
.
W
e
h
a
ve also
sh
own
th
at th
e
do
o
r
co
nt
r
o
l
s
y
st
em
coul
d e
a
si
l
y
be assem
b
l
e
d
usi
n
g c
h
e
a
p a
nd ea
si
l
y
avai
l
a
bl
e m
a
t
e
ri
al
s. A
n
al
y
s
i
s
of t
h
e
results
using s
t
anda
rd
pe
rformance m
e
trics suc
h
as
FAR
and
FRR produced ac
cu
racy (ge
n
uine acce
ptance
probability) of
m
o
re than
80%
,
which is
high
whe
n
c
o
m
p
are
d
with e
x
isting access c
ont
rol
schem
e
s.
REFERE
NC
ES
[1]
E Dovgan, B Kaluža, T Tušar an
d M Gams. Agent-ba
sed Security
S
y
stem fo
r User Verification.
I
n
ternational Joint
Conference on
Web Intelligen
ce
and Intelligen
t
Agent Technolog
y
. 2009
: 331-334
.
[2]
W
A
W
a
h
y
udi a
nd M Sy
a
z
il
awa
ti. Int
e
ll
igent Vo
ice-B
a
sed Door Access Control S
y
stem
Using Adaptiv
e-Network
-
based Fuzzy
Inf
e
rence S
y
s
t
ems (
ANFIS) for Building Secur
i
ty
.
Jo
urnal of Computer Science
. 2007
; 3(5): 274-280.
[3]
SY Kung, MW Mak and SH
Lin
.
B
i
ometr
i
c Au
th
enti
cation:
Machine Learning
Approach
. Prentice Hall. 2004.
[4]
S Furui. Recen
t
advances
in
speaker recognition.
Patter
Recognition Letters
. 1997; 18: 859-872
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Voice-Base
d
Door Access Control
Sy
stem Using the Mel
Frequency
Ceps
trum C
o
efficients … (Kay
ode
FA)
64
7
[5]
JP Campbell. “
Speaker
Recognition: a Tutorial
”.
Proceedings of
t
h
e IE
EE
. 1997;
85(9): 1437-146
2.
[6]
Sirko Molau,
Micha
e
l Pi
tz,
Ra
lf
Schl
¨
u
ter, and Hermann
Ney
.
Computing Mel
Frequency
C
e
pstral Co
efficien
ts
on
the P
o
wer S
p
ect
rum
.
ICASSP
. 2
001.
[7]
D Tran and M Wagner. “
Fuzzy Normalization Methods
for Speaker Verification
”. In Proc. IC
SLP2000, Beijin
g,
China. 2000; 1:
446-449.
[8]
D Tran and M
Wagner. “
A Proposed Likelihoo
d Transfo
rma
tion for Speaker Verifica
tion
”
.
In Proc
.
ICASSP20
00,
Turkey
. 2000; 2: 1069-1072.
[9]
J
C
Bezdek
. "
Pa
ttern Recognition
with
Fuzzy
Objective Function Algorithms
". Plen
um Press, New York. 1987.
[10]
JM Me
nde
l.
Uncertain Ru
le-bas
ed Fuzzy Logic Sy
stems: Introduction and New
Directions
. Pre
n
tic
e-Hall
, Upper
Saddle R
i
ver
,
NJ. 2001
.
Evaluation Warning : The document was created with Spire.PDF for Python.