Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
8
, No
.
6
,
Decem
ber
201
8
, p
p.
4533
~
4544
IS
S
N: 20
88
-
8708
,
DOI:
10
.11
591/
ijece
.
v8
i
6
.
pp
4533
-
45
44
4533
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/
journa
ls
/i
ndex.
ph
p/IJECE
A Context
-
ba
sed
Nume
ral Readin
g Techni
qu
e for Text to
Sp
ee
ch S
ystems
So
um
ya Pri
yada
rsi
ni P
anda
1
,
A
jit Kum
ar Nay
ak
2
1
Depa
rtment of
CS
E
Sili
con
Inst
it
ute of Te
chnolog
y
Bhubane
sw
ar,
India
2
Depa
rtment of
CS
&
IT
Siks
ha ‘O’ Anusandhan
Univer
sit
y
Bhu
bane
sw
ar,
India
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
6
, 2
01
7
Re
vised
Jun
22
, 201
8
Accepte
d
J
un
30
, 201
8
Thi
s
pape
r
pre
s
e
nts
a
novel
t
ec
h
nique
for
contex
t
base
d
num
era
l
rea
ding
in
India
n
l
angua
g
e
te
xt
to
spee
ch
s
y
stems
.
The
m
odel
uses
a
se
t
of
rule
s
to
det
ermine
the
c
onte
xt
of
th
e
nu
m
era
l
pronunc
iation
and
is
b
ei
n
g
int
egr
ated
with
the
wave
f
orm
conc
at
en
at
i
on
te
chni
qu
e
to
produc
e
spee
c
h
out
of
the
input
t
ext
in
Ind
ia
n
la
ngu
ag
es.
F
or
thi
s
purpose
,
the
thre
e
Indi
an
la
ngu
age
s
Odia,
Hindi
and
Benga
li
ar
e
con
sidere
d.
To
an
aly
z
e
the
per
form
anc
e
of
the
proposed
te
chn
i
que,
a
se
t
of
exp
eri
m
ent
s
are
per
form
ed
conside
r
ing
diffe
r
ent
cont
ex
t
of
num
e
ral
pronun
ci
a
ti
o
ns
and
th
e
resul
t
s
are
compare
d
with
exi
st
ing
s
y
llable
-
base
d
t
ec
hniqu
e.
The
r
esult
s
obtained
from
diffe
ren
t
e
xper
iments
show
s
the
eff
ec
t
ive
ness
of
the
p
roposed
te
chni
q
ue
in
produc
ing
int
el
l
igi
bl
e
spee
ch
out
of
th
e
ent
er
ed
te
xt
ut
te
ran
ce
s
compar
ed
to
the
exi
st
in
g
te
chni
qu
e
eve
n
w
i
th
v
er
y
l
ess stora
ge
and e
xec
ut
ion
t
ime
.
Ke
yw
or
d:
C
on
cat
e
native
synthesis
Indian
la
ngua
ge
s
Sp
eec
h
sy
nth
es
is
Text
-
to
-
S
peec
h
syst
em
s
Wav
e
f
or
m
Con
cat
enati
on
Tech
nique
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
:
Soum
ya
Pr
iy
adar
sini Pa
nda
,
Dep
a
rtm
ent o
f C
SE
,
Sil
ic
on
In
sti
tut
e of Tec
hnolog
y
,
Bhuba
nesw
a
r,
Od
is
ha, I
nd
ia
.
Em
a
il
:
sp
pand
a.cse@
gm
ai
l.c
om
1.
INTROD
U
CTION
The
goal
of
spe
ech
synthesis
is
to
dev
el
op
a
m
achine
hav
i
ng
a
n
intel
li
gib
le
,
natu
ral
soundin
g
voic
e
for
co
nv
ey
in
g
inf
or
m
at
ion
to
the
us
er
s
in
a
desire
d
voic
e,
la
nguag
e
a
nd
acce
nt
[1
]
,
[
2]
.
Re
search
in
th
e
area
of
sp
eec
h
synt
hesis
is
a
m
ulti
-
discipli
na
ry
fiel
d
with
ap
plic
at
ion
s
from
acou
sti
c
phonet
ic
s
(s
peec
h
producti
on
and
pe
rcep
ti
on)
[
3]
over
m
or
ph
ology
(
pro
nunciat
ion)
[4]
and
sy
ntax
(
par
ts
of
sp
eec
h,
gr
am
m
ar)
[
5]
,
to
sp
eec
h
sig
nal
processi
ng
(sy
nth
esi
s
)
[
6]
.
R
ecent
re
searc
h
in
t
he
a
rea
of
Sp
eec
h
an
d
L
angua
ge
P
r
oce
ssing
enab
le
s
m
achines
to
sp
ea
k
na
turall
y
li
ke
hu
m
ans
[7
]
.
A
T
ext
-
to
-
S
peech
(TTS
)
syst
em
i
n
this
as
pect
c
onve
rts
natu
ral
la
ng
ua
ge
te
xt
into
it
s
co
rr
e
spo
nd
i
ng
s
peec
h
[
8
]
.
The
i
ntell
igible
sp
eec
h
synt
hesis
syst
em
s
hav
e
a
wides
pr
ea
d
a
re
a
of
ap
plica
ti
o
ns
in
de
velo
pi
ng
hum
an
–
co
m
pu
te
r
interact
ive
syst
em
[9]
li
ke,
ta
lkin
g
c
om
pu
te
r
syst
e
m
s
[10]
,
ta
lking
t
oys
[11]
,
et
c.
S
pe
ech
synt
hesis,
com
bin
ed
wi
th
sp
eec
h
re
c
ogniti
on,
al
lo
ws
f
or
interact
ion
wit
h
m
ob
il
e
de
vices
via
natu
ral
la
ngua
ge
proce
ssing
inter
face
s.
[
12
]
An
al
yz
ing
the
i
nput
te
xt
an
d
conve
rting
it
into
a
c
om
pu
te
r
rea
dab
le
for
m
fo
r
ob
ta
i
ning
the
a
ppr
opri
at
e
pronu
nciat
ion
play
s
an
im
portant
ro
le
i
n
a
ppr
opr
ia
te
sp
eech
un
i
t
pro
du
ct
io
n
a
nd
f
or
it
s
unde
rstan
dab
il
it
y
by
the
li
ste
ners
[
13
]
.
Te
xt
a
nal
ysi
s
is
the
fr
ont
en
d
la
ngua
ge
pr
oce
sso
r
of
the
TT
S
syst
e
m
[14]
,
wh
ic
h
acce
pts
input
te
xt,
an
al
yz
es
it
and
org
anize
s
into m
anag
ea
ble li
st of
w
ords [1
5
]
.
An
in
put
te
xt
m
ay
con
ta
in
sy
m
bo
ls
(doubl
e
quote,
com
m
a,
repor
t
,
et
c),
nu
m
ber
s,
a
bbr
eviat
ion
s
or
sp
eci
al
sym
bo
ls
[
16
]
.
Te
xt
norm
alizat
ion
in
volves
tra
ns
f
orm
at
ion
of
t
he
r
aw
i
nput
te
xt
into
the
e
quival
ent
of
wr
it
te
n
words
[
17
]
.
It
al
s
o
in
vo
l
ves
c
onve
rting
al
l
le
tt
ers
of
lo
wer
case
or
uppe
r
case,
re
m
ov
ing
pu
nctu
at
ion
s,
acce
nt
m
ark
s,
stop
words
or
too
c
omm
on
words
(like
“
D
on
’t”
vs.
“D
o
no
t”
,
“
I’
m
”
vs
.
“I
am
”,
“C
an’
t”
vs
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4533
-
4544
4534
“cannot”
,
et
c).
Sentences
are
a
gr
oup
of
w
ord
segm
ents
and
these
se
gm
ents
m
ay
be
an
acronym
,
a
sing
le
word
or
a
num
eral
[
18
]
.
Wh
il
e
for
the
ab
bre
viati
on
s
or
acr
on
ym
s,
on
e
ti
m
e
no
rm
al
iz
ed
pro
nunciat
io
ns
m
a
y
be
m
ai
ntained,
the
pro
nuncia
ti
on
of
num
erals
var
ie
s
de
pe
ndin
g
on
the
c
onte
xt
of
it
s
use
in
the
sente
nc
e
or
word [
1
9
]
.
A
num
ber
m
ay
be
prono
un
ce
d
di
ff
e
ren
tl
y
in
differe
nt
sit
ua
ti
on
s
a
nd
nee
ds
to
be
co
nvert
ed
int
o
their
appr
opriat
e
prono
un
cea
ble
f
or
m
s
to
pro
d
uc
e
the
desi
red
sp
eec
h
outp
uts
[1
8
]
.
Ta
ble
1
sh
ows
s
om
e
exam
ple
pro
nunciat
ion
o
f
the
E
ngli
sh
n
um
erals
in
dif
fer
e
nt
sit
uations.
In
t
his
aspec
t
m
os
t
of
the
f
or
ei
gn
la
ng
uages
li
ke
En
glish
a
re
w
el
l
researc
hed
[
1
9],
w
her
e
the
pr
onunci
at
ion
r
ules
are
si
m
pler
due
t
o
the
oc
cu
rr
e
nce
of
pro
nunciat
ion
rep
et
it
ion
s
afte
r
20.
(e.
g.
:
Tw
enty
one
,
tw
ent
y
tw
o
,
…,
Th
ir
ty
on
e,
Th
irt
y
t
wo
,…etc)
.
H
oweve
r,
the
Indian
la
ng
uag
e
TT
S
te
ch
niques
sil
l
pr
es
ents
gap
f
or
it
s
acce
ptance
by
the
us
ers
due
to
the
un
a
vaila
bilit
y
of
a
pprop
riat
e
pronu
nciat
ion
ru
le
s.
T
he
pr
ob
a
bili
ty
of
re
petit
ion
of
pro
nunciat
ion
is
r
el
at
ively
ver
y
le
ss
in
Indian
la
ng
uages
at
wo
r
d
le
ve
l
(
e.g
:
pro
nunc
ia
ti
on
s
of
the
nu
m
ber
s
in
H
ind
i:
21
-
“i
k
-
kis”,
22
-
“
baa
-
i
s”
,
23
-
“t
ei
-
s”,
et
c.).
This
inc
reases
the
com
plexity
of
the
n
um
eral
read
in
g
m
od
ule.
The
refor
e
,
m
os
t
of
t
he
researc
hers
us
e
si
m
ple
dig
it
b
ased
rea
ding
m
od
el
s
that
st
or
es
the
rec
ord
ed
unit
s
fo
r
sin
gle
dig
it
s
from
0
to
9
for
pro
du
ci
ng
the
desire
d
outpu
t
sp
ee
ch
but
did
not
addre
ss
the
con
te
xt
base
d
nu
m
eral
read
in
g
.
Howe
ver,
con
te
xt
base
d
nu
m
eral
read
i
ng
play
s
an
im
po
rtant
r
ole
to
en
ha
nce
the
unde
rstan
da
bili
ty
of
the
pr
oduc
e
d
sp
eec
h.
It
is
al
ways
easi
er
to
unde
rstan
d
the
pr
ic
e
of
so
m
e
it
e
m
if
it
is
prono
un
ce
d
base
d
on
posit
ion
base
d
read
i
ng
li
ke
“fi
fty
f
ive thousa
nd
five
hu
ndre
d”
instea
d of
pron
ouncin
g
“fi
ve
five
fi
ve
zer
o
zer
o”.
Th
e
focus
of
this pa
per is t
o
address
the c
onte
xt
base
d num
eral pronu
nci
at
ion
in
In
dian l
anguag
e
sce
na
rio.
Table
1.
E
xam
ple
Pr
on
un
ci
at
ion o
f
a
Nu
m
ber
in
Dif
fer
e
nt S
cenari
os
Exa
m
p
le
Ty
p
e
Pron
u
n
ciatio
n
2015
Date/Quan
tif
ier
Two th
o
u
san
d
f
i
f
teen
2015
Ph
o
n
e nu
m
b
er
Two zer
o
on
e f
iv
e
0
.50
2
Nu
m
b
e
r
Po
in
t f
iv
e
k
n
o
t tw
o
2
0
.15
Deci
m
a
l nu
m
b
er
Twenty
po
in
t on
e f
iv
e
Ther
e
are
on
l
y
few
e
r
m
od
e
ls
docum
ented
f
or
sp
ee
ch
s
ynthesis
i
n
Indian
la
ngua
ge
s
[
20
]
-
[
24
]
,
howe
ver
the
c
on
te
xt
dep
e
nd
ent
num
eral
pronu
nciat
ion
s
ha
s
not
been
w
el
l
addresse
d
[
25
]
-
[
28
]
.
The
dhva
ni
TTS syst
em
f
or
Ind
ia
n
la
ngua
ge
[25
]
, m
a
intai
ns
the
pronu
nc
ia
ti
on
s of nu
m
erals up
to h
undred as
t
he
phonet
ic
represe
ntati
on
and
us
e
the
po
sit
io
n
pronu
nciat
ion
s
for
‘hu
ndre
d’,
‘t
hous
a
nd’
,
et
c
posit
io
ns
at
ta
ched
to
t
he
“
up
to
hundre
d
pro
nunciat
ion
”
for
read
in
g
the
nu
m
erals.
However,
the
co
ntext
dep
e
ndent
nu
m
eral
read
ing
aspect
is
no
t
con
si
dered
in
sp
eec
h
pro
duct
ion.
A
r
ule
-
based
nu
m
eral
read
in
g
m
et
hod
is
pr
ese
nted
in
[
18
]
f
or
the
Od
ia
la
ngua
ge.
In
t
his
pap
e
r,
we
pr
ese
nt
a
pro
nunciat
io
n
r
ule
bas
ed
a
ppr
oach
f
or
the
up
to
hu
ndred
pro
nunciat
io
ns
and
i
ncorpo
rate
it
with
the
wav
e
f
or
m
concat
enati
on
te
c
hn
i
qu
e
(
W
CT
)
[29]
to
pro
duce
ou
t
pu
t
spe
e
ch
f
or
Indian
la
ng
uage
nu
m
erals.
Also
,
the
co
nte
xt
dep
e
nd
e
nt
nu
m
eral
pr
onunc
ia
ti
on
aspects
of
the
num
erals
are
consi
der
e
d
to
pro
du
ce
natu
ra
l
sp
eech
se
gme
nts
to
increas
e
the
underst
and
a
bili
ty
.
A
set
of
ex
per
im
ents
are
perform
ed
to
evaluate
the
pe
rfor
m
ance
of
the
pr
op
os
ed
m
od
el
com
pa
red
to
the
exi
sti
ng
syl
la
ble
base
d
te
chn
iq
ue
with
res
pect
to
di
fferent
c
onte
xts
of
num
eral
pronu
nciat
ion.
A
nd
the
res
ults
ob
ta
ine
d,
s
hows
t
he
eff
ect
ive
ness o
f
the
prop
os
e
d t
echn
iq
ue c
om
par
e
d
t
o
the
exi
sti
ng
tech
niqu
e in
diff
e
re
nt c
on
te
xts.
The
r
em
ai
nd
er
of
t
he
pa
pe
r
i
s
orga
nized
as
fo
ll
ows.
I
n
th
e
nex
t
sect
i
on,
we
discuss
e
d
about
the
wa
ve
form
con
cat
e
natio
n
te
chn
iq
ue
as
t
he
pr
opos
e
d
nu
m
eral
read
i
ng
m
odule
is
inco
rpor
at
e
d
into
the
ru
le
base
d
con
cat
e
native
appr
oach.
Sect
ion
3
descr
i
bes
the
detai
ls
abo
ut
the
pro
pos
ed
m
od
el
and
the
co
ntext
de
pe
nd
e
nt
nu
m
eral
pro
nunciat
ion
r
ules.
The
e
xperim
ental
m
et
ho
do
l
og
y
and
res
ult
an
al
ysi
s
fo
r
our
te
chn
i
qu
e
is
give
n
i
n
sect
ion
4,
sho
wing
the
ef
fec
ti
ven
ess
of
thi
s
te
chn
iq
ue
in
pro
du
ci
ng
inte
ll
igible
sp
eech
.
Sect
ion
5
co
nclu
des
the
discussi
on,
exp
la
ini
ng
th
e
find
i
ng
s
of
our
ex
per
im
ents
and
the
f
ut
ur
e
directi
ons
of
t
his
w
ork
,
wh
e
r
e
furthe
r work
m
ay
be unde
rtak
en.
2.
WA
VEFO
R
M CO
NCAT
ENATIO
N
T
ECHNIQ
UE (WCT)
As
c
om
par
ed
t
o
E
ng
li
s
h,
m
os
t
of
the
India
n
la
ng
uag
e
s
ha
ve
a
ppr
ox
im
a
te
ly
twic
e
as
m
any
vowel
s
and
c
on
s
onant
s
al
on
g
with
a
nu
m
ber
of
pos
sible
conj
un
ct
char
act
e
rs
for
m
ed
by
com
bi
nation
of
tw
o
or
m
or
e
char
act
e
rs
[
28]
.
The
refo
re,
a
la
rg
e
nu
m
ber
of
sp
eec
h
unit
s
are
nee
ded
t
o
be
sto
re
d
in
the
s
peech
da
ta
bas
e
wh
il
e a con
cat
enati
ve
spe
ech
synthesis t
ech
ni
qu
e is use
d
f
or prod
ucin
g
un
interr
up
te
d
spe
ech. Howe
ver
, W
CT
[29] u
ses
only
3
5 basic s
peec
h
unit
s of the
c
on
s
ona
nt (
C) a
nd
vowel
(V) s
ounds
i
ns
te
a
d of
st
or
i
ng
all
re
qu
i
re
d
sp
eec
h
un
it
s
in
the
database,
and
der
i
ve
al
l
oth
e
r
unit
s
us
ing
a
r
ule
base
d
wav
ef
orm
con
cat
enati
on
te
ch
nique.
The
li
st o
f 3
5 b
asi
c s
peec
h un
i
ts are li
ste
d
i
n Table
2.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A Co
ntext
-
base
d
N
um
er
al Re
adin
g
Tec
hniq
ue
for Tex
t
…
(
Soumy
a
Priy
adarsi
ni P
anda
)
4535
Fo
r
pro
duci
ng
the
ou
t
put
spe
ech
f
or
t
he
r
equ
i
red
s
peec
h
segm
ents,
a
fr
act
io
n
-
base
d
wa
vefo
r
m
con
cat
e
natio
n
te
chn
iq
ue
is
use
d.
The
f
racti
on
dua
rio
ns
a
r
e
determ
ined
dynam
ic
al
l
y
from
the
sp
eec
h
data
base
d
on
the
vowel
on
set
po
i
nt
identific
at
io
n
te
ch
nique
[
29]
.
These
fr
act
i
on
s
durati
ons
a
re
co
ns
ide
re
d
f
or
the
wav
e
f
or
m
concat
enati
on
pro
cess
to
obta
in
the
desire
d
s
peech
unit
s.
Wh
il
e
the
r
ul
e
-
base
d
c
on
cat
enati
ve
te
chn
iq
ue
(RCT)
[
28]
us
es
a
sta
ti
c
fr
act
ion
durati
on
f
or
c
oncat
enati
on
th
e
us
e
of
dyna
m
ic
fr
act
ion
durati
on
s
in
W
CT
[29]
enh
a
nces
t
he
qual
it
y
of
s
peec
h
bei
ng
pro
du
ced.
T
his
f
ract
ion
base
d
co
nc
at
enati
on
pro
cess
is
consi
der
e
d
for
the
de
pe
nd
e
nt
t
ype
of u
nit
pa
irs
s
uch
as
Co
nsona
nts
at
ta
che
d
to
Ma
tra/
F
al
a/
Halant/
Cons
on
a
nts
and
the
w
ho
le
wa
ved
at
a
is
us
e
d
f
or
pro
duci
ng
the
in
de
pende
nt
unit
pairs
li
ke
Co
ns
ona
nts
at
ta
ched
t
o
Con
s
onant/
V
ow
el
,
Vowels
at
ta
cched
to
Con
s
ona
nts/Vowel
s.
Fig
ur
e
1
sh
ows
the
p
ort
ion
based
wa
vefor
m
con
cat
e
natio
n process
to pr
od
uce th
e
s
ound
“
\
re” fro
m
“
\
ra”
an
d
“
\
ae”
us
ing
t
he WCT t
echn
i
qu
e
.
Table
2.
Sp
eec
h un
it
s i
n
Database
Figure
1.
W
a
ve
p
at
te
r
n of
“/r
e”
(C
-
M) s
ound a
fter c
on
c
at
enati
ng po
rtion
s
f
r
om
/ra
an
d
/
ae
s
ound
3.
PROP
OSE
D MO
DEL
In
t
his
sect
io
n,
we
pr
ese
nt
a
pro
nunciat
ion
ru
le
-
base
d
te
c
hn
i
qu
e
f
or
pro
du
ci
ng
sp
eec
h
segm
ents
for
the
Indian
la
ng
uag
e
nu
m
erals
by
identify
ing
the
phon
em
e
l
evel
si
m
il
aritie
s
in
the
nu
m
eral
pr
on
un
ci
at
io
ns
i
n
the
thre
e
co
ns
i
der
e
d
la
ng
ua
ge
s
(
Od
ia
,
Hindi,
a
nd
Be
ngal
i).
Fig
ure
2
shows
the
o
verv
ie
w
of
the
pro
po
s
e
d
nu
m
eral
read
i
ng
m
odule
an
d
the
detai
ls
of
t
he
phases
are
disc
us
se
d
ne
xt.
Howe
ver,
first
a
c
on
te
xt
identific
at
ion p
ro
ces
s is
perfor
m
ed
to ide
ntify
the c
on
te
xt
of
the num
eral pronu
nciat
ion as
discusse
d ne
xt.
Figure
2.
Text
-
to
-
s
peech co
nv
ersio
n process
Set of
sp
eech u
n
its
in th
e datab
ase
\
a
\
o
\
ch
a
\
tth
a
\
th
a
\
pha
\
lla
\
aa
\
ka
\
ch
h
a
\
dda
\
da
\
ba
\
la
\
ee
\
kha
\
ja
\
ddha
\
dha
\
bha
\
sh
a
\
uu
\
ga
\
jh
a
\
nna
\
na
\
ma
\
ha
\
ae
\
gha
\
tta
\
ta
\
pa
\
ra
\
ya
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4533
-
4544
4536
3.1
Conte
xt D
e
pe
ndent N
umer
al Pr
on
u
nciati
on
The
co
ntext
de
pende
nt
num
eral
pronu
ncia
ti
on
is
an
im
portant
issue
for
pro
duci
ng
m
eaning
f
ul
sp
eec
h
sam
pl
es
for
the
nu
m
erals.
The
s
i
m
ple
dig
it
read
in
g
te
ch
nique
m
ay
no
t
pro
vid
e
the
de
sired
unde
rstan
dab
il
it
y
in
al
l
sit
uations
.
F
or
exa
m
ple,
wh
il
e
re
adin
g
a
la
r
ge
r
quantit
y
or
pri
ce
say
1,54,
954
by
si
m
ply
read
in
g
the
dig
it
s
as
“
on
e
-
fi
ve
-
fou
r
-
nin
e
-
five
-
f
our”
m
akes
the
li
ste
ner
thin
k
t
o
r
earra
ng
e
the
num
ber
s
to
unde
rstan
d
the
spok
e
n
pri
c
e
or
quantit
y;
appr
opriat
e
pronu
nciat
ion
as
on
e
-
la
kh,
fifty
fo
ur
-
t
hous
a
nd
,
nin
e
-
hundre
d,
fifty
-
four
m
ay
m
ake
so
m
e
sense
to
the
li
ste
ner
s
.
T
he
sim
il
ar
var
i
at
ion
of
pronu
nciat
ion
al
so
e
xten
ds
to the I
ndia
n
la
ngua
ges.
Ta
ble
. 3
show
s
om
e
exam
ple n
um
e
rals and
thei
r
pro
nunciat
io
n
in
d
iffe
ren
t co
nte
xt in
En
glish a
nd Odia l
an
gu
a
ge
.
Table
3.
Pronu
nciat
ion
of a
N
um
ber
in
Dif
fe
ren
t
Scena
rios
Exa
m
p
le
Ty
p
e
Eng
lish
Pron
u
n
ciatio
n
Od
ia Pr
o
n
u
n
ciatio
n
୦୨
-
୦୭
-
୨୦୧୫
or
୦୨
/
୦୭
/
୨୦୧୫
Date
Ju
ly
two two
tho
u
san
d
f
if
teen
“
Du
i
-
saat
-
d
u
i hajaar
p
an
d
ara”
୦୨
:
୪୦
Ti
m
e
Two f
o
rty
Du
i
-
ta
-
ch
aalish
୦୬୭
୪
-
୨୦୧୫
Ph
o
n
e
n
u
m
b
er
Zer
o
six
sev
en
f
o
u
r
two
ze
ro
o
n
e f
iv
e
Su
n
-
ch
h
a
-
saat
-
ch
a
ari
-
dui
-
su
n
-
aek
-
p
aanch
୨୦୧୫
Qu
an
tif
ier
two
tho
u
san
d
and
fif
teen
Du
i
-
h
ajjar
-
p
an
d
ara
୨୦୧୫
.
୧୫
Nu
m
b
e
r
with
Deci
m
a
l po
in
ts
Two th
o
u
san
d
f
i
f
teen
po
in
t
o
n
e f
iv
e
Du
i
-
h
ajjar
-
p
an
d
ara
-
d
as
m
ic
-
p
an
d
ara
A
num
ber
in
diff
e
ren
t
I
nd
ia
n
la
ngua
ges
m
ay
be
pro
nounce
d
by
si
m
ply
read
in
g
the
dig
it
s
wh
il
e
m
ean
for
a
qu
a
ntit
y
[21],
phone
nu
m
ber
or
cred
it
card
num
ber
,
et
c.;
the
nu
m
ber
m
ay
be
read
by
the
relat
ion
s
hi
p
with
it
s
po
sit
ion
s
w
hile
m
eant
fo
r
a
pr
ic
e
in
di
cat
or
or
ye
ar.
In
case
of
a
fr
a
ct
ion
value
the
le
ft
par
t
bef
or
e
the
decim
al
po
int
is
read
ba
sed
on
the
relat
ionship
bet
ween
the
posit
ion
of
the
cha
racter
a
nd
the
num
ber
s
after
the
pe
rio
d
are
read
as
si
ngle
dig
it
s.
Wh
il
e
r
eadin
g
a
date
people
al
ways
read
as
“ae
k
-
tin
-
dui
-
ha
j
aa
r
-
s
ohal
a”
for
the
date
“
01
-
03
-
2016”
in
triple
t
form
at
(d
d
-
mm
-
yyyy
or
dd/m
m
/y
yyy)
.
Als
o,
for
rea
ding
a
ti
m
e
int
erv
al
separ
at
e
d
by
a
colo
n
the
f
or
m
at
is
diff
eren
t
f
or
t
he
num
ber
befor
e
the
colo
n
an
d
after
the
col
on.
T
o
inco
rpor
at
e
al
l
the
con
sid
e
re
d
var
ia
ti
on
of
a
nu
m
eral
pr
onunci
at
ion
a
set
of
m
anu
al
ly
cod
ed
ru
le
s
are
pr
e
par
e
d.
T
he
con
te
xt
ide
ntifi
cat
ion
ru
e
s
a
re
presente
d
belo
w,
w
her
e
n
is
t
he
nu
m
ber
of
dig
it
s
in
the
nu
m
ber
and d
i
is t
he
it
h digit
in
the
nu
m
ber
.
Conte
xt de
p
endent
pronuncia
ti
on
rules:
Rul
e
1:
IF n
>=
10
AN
D no
sepa
rati
on
in
b
et
we
en
THEN
per
form
digi
t rea
d
ing
Rul
e
2:
IF n
>=
10
AN
D d
i
sepa
ra
ted b
y
“
,
”
THEN
per
form
positi
on
b
ase
d
di
git
r
ea
ding
Rul
e
3:
IF di
git
s sep
arat
ed
b
y
“
-
“
or
“
/” i
n
a tri
pl
et format
THEN
per
form
dat
e
form
at
d
igi
t
rea
d
ing
Rul
e
4:
IF
digi
ts se
p
arat
ed
b
y
“
-
“
THEN
per
form
digi
t rea
d
ing
Rul
e
5:
IF di
git
s sep
arat
ed
b
y
“
:
”
THEN
per
form
t
ime
form
at
dig
it
rea
d
ing
(dig
it
r
ea
ding
for
dig
it
s
bef
ore
“
:”
and
positi
on
base
d
d
igi
ts
re
adi
ng
for
digi
ts
aft
er
“
:”
Rul
e
6:
IF di
git
s sep
arat
ed
b
y
“
.
”
THEN
Perform
positi
on
b
ase
d
di
git
r
ea
ding
for
di
git
s be
for
e”.” a
n
d
digi
t
re
adi
ng
f
or
digits
after
“
.
”
Rul
e
7:
IF num
ber
foll
o
wed
b
y
pri
ce i
nd
ic
a
tor
THEN
per
form
positi
on
b
ase
d
di
git
r
ea
ding
Rul
e
8:
IF rul
e
no
t
found
for
th
e
d
ig
it for
m
at
THEN
per
form
digi
t rea
d
ing
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A Co
ntext
-
base
d
N
um
er
al Re
adin
g
Tec
hniq
ue
for Tex
t
…
(
Soumy
a
Priy
adarsi
ni P
anda
)
4537
3.2
.
Pron
uncia
tion Rules
As
the
pr
opos
e
d
te
ch
nique
f
or
sp
eec
h
sy
nth
e
sis
stores
only
so
m
e
basic
sp
e
ech
un
it
s
a
nd
pro
duce
al
l
the
sou
nds
f
rom
these
basic
un
it
s
base
d
on
so
m
e
sp
eci
fied
r
ules,
t
he
prono
un
cea
ble
unit
s
f
or
num
er
al
s
ar
e
need
e
d
to
be
identifie
d
a
nd
m
app
ed
to
the
res
pecti
ve
ch
aracte
r
e
quival
ents
for
t
he
s
ounds
to
pro
du
ce
the
desire
d
outp
ut
sp
eec
h.
Also
,
there
is
no
ge
ner
al
iz
ed
r
ule
avail
able
f
or
t
he
pr
onunci
at
ion
of
num
ber
s
up
t
o
100.
H
ow
e
ver,
for
num
ber
s
gr
eat
er
t
han
100,
a
re
petit
ion
of
pr
onunci
at
ion
m
ay
occu
r
(e.
g.
:
122
“e
k
sou
-
baais”,
12
3
-
“
ek
sou
-
te
is”
,
e
tc
in
Hind
i
la
ngua
ge.
T
her
e
fore,
the
nu
m
erals
after
10
0
m
ay
be
fo
rm
ed
by
con
cat
e
natin
g
the
10
0th
10
00th,
…et
c
place
pro
nunciat
io
ns
with
t
heir
r
especti
ve
up
t
o
100
pro
nunc
ia
ti
on
s.
We
pr
e
pa
re
a
set
of
pro
nunci
at
ion
ru
le
s
f
or
ob
ta
ini
ng
the
up
to
100
pr
onunci
at
ion
s
.
The
pro
nounceable
un
it
identific
at
ion p
ro
ces
s is
discu
ssed belo
w.
The
nu
m
ber
s
from
1
-
9
a
nd
al
l
10
th
posi
ti
on
pronu
nciat
ion
s
a
re
nee
de
d
to
be
m
ain
ta
ined
f
or
perform
ing
sing
le
di
git
read
in
g.
T
he
pro
nunciat
ions
of
the
nu
m
erals
fr
om
1
-
9
a
nd
10th
posit
io
ns
are
pr
ese
nted
in
T
able
4
and
Ta
bl
e
5
resp
ect
ive
ly
fo
r
the
cons
idere
d
l
angua
ge
s.
Also,
there
m
ay
be
a
si
m
i
la
rit
y
no
ti
ced
in
t
he pr
onunci
at
ion
s
of the
num
erals in t
he
th
ree c
on
si
der
e
d
la
nguag
e
s.
Table
4.
Pronu
nciat
ion
of
Num
erals
up to 1
0
in
the t
hr
ee
Con
si
der
e
d
La
ngua
ges
Nu
m
eral
Od
ia
Hin
d
i
Ben
g
ali
1
Aek
Aek
Aek
2
Du
i
Do
Du
i
3
Tin
Tin
Tin
4
ch
aari
Ch
aar
Ch
aar
5
Paan
ch
Paan
ch
Paach
6
Ch
h
a
Ch
h
e
Ch
h
o
y
7
saate
Saat
Sh
aat
8
Aatth
e
Aath
Aat
9
na
n
au
noy
Table
5.
Pronu
nciat
ion
of
Num
erals
for 10
t
h
Po
sit
io
ns
i
n
th
e thr
ee
Co
ns
id
ered Lan
guag
e
s
Nu
m
eral
Od
ia
Hin
d
i
Ben
g
ali
0
Su
n
Su
n
y
a
Sh
o
o
n
n
o
10
Dasa
Das
Do
sh
20
Ko
d
iae
Bis
h
Ku
ri/bis
h
30
Tir
ish
Tish
Tir
ish
40
Ch
aalish
Ch
aallish
ch
aallish
50
p
achaas
h
p
achaas
h
p
o
n
ch
aash
60
saath
iae
saatth
Sh
aat
70
saath
iae
sattar
sh
o
tto
r
80
asi
ash
i
ash
i
90
n
ab
e
n
ab
e
n
o
b
b
o
i
100
sah
e
sau
Sh
o
1000
h
ajaar
e
h
ajar
h
ajaar
1
0
0
0
0
0
lak
h
y
a
laak
h
lak
sh
1
0
0
0
0
0
0
0
k
o
ti
k
arod
k
o
ti
As
in
I
nd
ia
n
la
ngua
ges,
t
he
pro
ba
bili
ty
of
re
petit
ion
of
pro
nunciat
ion
is
r
el
at
ively
ver
y
le
ss
at
w
ord
le
vel,
we
t
ry
to
de
rive
t
he
pronu
nciat
ion
si
m
il
arities
at
phon
em
e
le
vel
for
the
up
to
100
pro
nunciat
ions.
F
or
exam
ple,
w
hen
the
num
eral
2
is
pr
e
sent
at
un
it
or
10th
place
it
has
one
ty
pe
of
pro
nunciat
ion
at
be
ginnin
g
or
end.
T
he
pro
nunciat
io
n
sim
i
la
riti
es
in
the
three
c
on
si
de
red
la
ngua
ges
for
2
at
un
it
and
10th
p
la
ce
are
pr
ese
nted
in
Table
6
a
nd
Table
7
res
pe
ct
ively
.
Consi
der
i
ng
s
uch
si
m
il
arities
in
pro
nunciat
io
ns
a
set
of
si
m
il
arity r
ules
are pre
pa
red f
or the
pro
nunci
at
ion
s
for n
ume
rals f
ro
m
1
1
-
99.
Table
6.
E
xam
ple
SCR
I
PT
s a
nd
Pro
nunciat
ion R
epeti
ti
ons
for
“
2”
at
te
nth
Plac
e
Nu
m
eral
Od
ia
Hin
d
i
Ben
g
ali
21
Eko
-
is
Ik
-
k
is
Aek
22
Baa
-
ish
Baa
-
ish
Baa
-
ish
23
Te
-
ish
Te
-
ish
Te
-
ish
…
…
…
…
29
An
a
-
tir
-
ish
Un
-
t
-
ish
u
n
o
tirish
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4533
-
4544
4538
Table
7.
E
xam
ple
Sc
ripts
a
nd
Pronu
nciat
ion
Re
petit
ion
s
for
“2
” at
unit
Pla
ce
Nu
m
eral
Od
ia
Hin
d
i
Ben
g
ali
12
Baa
-
ra
Baa
-
ra
b
aro
22
Baa
-
ish
Baa
-
ish
b
aaish
32
Ba
-
tish
Bat
-
ish
b
o
ttrish
…
…
…
…
92
Baya
-
n
ab
e
b
ay
an
ab
e
Bira
-
n
o
b
b
o
i
The
pro
nuncia
ti
on
s
of
num
erals
with
sta
rtin
g
a
nd
e
nd
i
ng
s
i
m
i
la
riti
es
in
the
th
ree
c
onsi
der
e
d
la
ngua
ge
s
are
pr
ese
nted
in
T
able
8,
Ta
ble
9
an
d
Ta
ble
10
resp
ect
ively
for
O
dia,
Hindi,
and
Be
ngal
i
la
ngua
ge
a
nd
f
or
ea
c
h
of
the
la
ngua
ge
and
sim
il
arity
three
sta
te
s
are
m
ai
ntained
at
phon
em
e
lev
el
as
sho
wn
i
n
Fig
ur
e
3
a
nd
base
d
on
prop
e
r
m
atch
the
res
pecti
ve
pro
nouncea
ble
un
it
s
are
e
xtracted
f
r
om
the
sp
eec
h
dat
abase
to
pro
du
ce
th
e
ou
t
pu
t
sp
eec
h.
Table
8.
Pronu
nciat
ion
of
Od
i
a
N
um
erals
with
Sta
rting (C
olu
m
n W
ise
)
a
nd
En
ding
(Row
W
i
se) Si
m
il
arity
Pron
u
n
ciatio
n
aek
b
aa
te
ch
au
p
an
ch
h
au
sat
ath
an
a
Ra
aeg
-
aa
-
ra
Baa
-
ra
Te
-
ra
Ch
au
-
da
p
an
-
da
-
ra
so
-
ha
-
la
sat
-
a
-
r
a
Ath
-
a
-
ra
Un
e
-
ish
Is
aek
-
oi
-
s
Baa
-
is
te
-
tis
Ch
au
-
b
is
p
an
-
ch
is
ch
h
a
-
b
is
sat
-
e
-
is
ath
-
e
-
is
An
a
-
tiris
Ris
aek
-
ti
-
ris
ba
-
tis
te
-
tis
ch
au
-
ti
-
ris
p
ain
-
tiris
ch
h
a
-
tis
sain
-
tiris
-
ath
-
a tir
is
An
a
-
ch
alis
Lis
aek
-
ch
alis
ba
-
ya
-
lis
te
-
ya
--
lis
ch
au
-
ra
-
lis
p
ain
-
ch
alis
ch
h
a
-
ya
-
lis
sat
-
ch
alis
ath
-
a
-
ch
alish
An
a
-
ch
as
Ban
aek
-
b
an
ba
-
b
an
te
-
p
an
Ch
au
--
b
an
p
an
-
ch
aaban
-
Ch
h
a
-
p
an
sat
-
aa
-
b
an
ath
-
aa
-
b
an
An
a
-
sath
i
Sath
i
aek
-
a
-
sath
i
ba
-
sath
i
te
-
sath
i
ch
au
-
sath
i
p
an
-
sath
i
ch
h
a
-
sath
i
sat
-
sath
i
ath
-
a
-
sath
i
An
a
-
sto
ri
Stari
aek
-
a
-
stari
b
aa
-
stari
te
-
stari
ch
au
-
stari
p
an
-
ch
a
-
stari
ch
h
a
-
stari
sat
-
a
-
stari
ath
-
a
-
stari
An
a
-
asi
Asi
aek
-
a
-
asi
ba
-
ya
-
asi
te
-
ya
-
asi
ch
au
-
raa
-
asi
p
an
-
ch
aa
-
asi
ch
h
a
-
y
aa
-
asi
sat
-
aa
-
asi
ath
-
aa
-
asi
An
aa
-
n
ab
e
Nab
e
aek
-
aa
-
n
ab
e
ba
-
ya
-
n
ab
e
te
-
ya
-
n
ab
e
ch
au
-
ra
-
n
ab
e
p
an
-
ch
aa
-
n
ab
e
ch
h
a
-
y
aa
-
n
ab
e
sat
-
aa
-
n
ab
e
ath
-
aa
-
n
ab
e
an
e
-
sat
Table
9.
Pronu
nciat
ion
of
Hindi
N
um
erals
w
it
h
Starti
ng
(Colu
m
n W
ise
)
a
nd
E
nd
i
ng (
R
ow
W
i
se) Si
m
ilarity
Pron
u
n
ciatio
n
ik
b
aa
te
ch
au
p
an
ch
h
a
sat
ath
un
Raa
Gy
a
-
raa
b
aa
-
raa
te
-
raa
Ch
au
-
d
aa
p
an
-
d
-
ra
So
-
la
sat
-
raa
ath
-
aa
-
raa
un
-
ish
Is
ik
-
is
-
is
Baa
-
is
te
-
is
Ch
au
-
b
is
pa
-
ch
-
is
Ch
h
a
-
b
is
sat
-
aa
-
is
ath
-
aa
-
is
Un
-
ti
-
sh
Tis
ik
-
a
-
tis
ba
-
tis
te
-
tis
ch
au
-
tis
p
ain
-
tis
Ch
h
a
-
tis
sain
-
tis
ath
-
tis
un
-
ch
alis
Lis
ik
-
ch
a
-
-
lis
ba
-
ya
-
lis
te
-
ya
-
lis
ch
au
-
ra
-
lis
p
ain
-
ta
-
lis
Ch
h
a
-
ya
-
lis
sat
-
ch
a
-
lis
ath
-
ch
a
--
lish
Un
-
ch
as
Ban
ik
-
ya
-
b
an
ba
-
b
an
te
-
p
an
Ch
au
-
b
an
pa
-
ch
-
p
an
Ch
h
a
-
p
an
sat
-
aa
-
b
an
ath
-
aa
-
b
an
un
-
sath
Sath
ik
-
sath
ba
-
sath
te
-
sath
ch
au
-
sath
p
ain
-
sath
ch
h
a
-
sath
sat
-
sath
-
ath
-
a
-
sath
Un
-
atar
Tar
ik
-
a
-
tar
b
aa
-
tar
te
-
ta
r
ch
au
-
tar
pa
-
ch
a
-
ta
r
Ch
h
a
-
tar
sat
-
a
-
ta
r
ath
-
a
-
tar
una
-
asi
Asi
ik
-
ya
-
asi
ba
-
y
a
-
asi
te
-
ya
-
asi
ch
au
-
raa
-
asi
p
an
-
ch
aa
-
asi
Ch
h
a
-
y
aa
-
asi
sat
-
aa
-
asi
ath
-
aa
-
asi
-
un
-
ya
-
n
ab
e
Nb
e
ik
-
ya
-
n
b
e
ba
-
y
a
-
n
b
e
te
-
ya
-
n
b
e
ch
au
-
ra
-
n
b
e
p
an
-
ch
aa
-
n
b
e
ch
h
a
-
y
aa
-
n
b
e
sat
-
aa
-
n
b
e
ath
-
aa
-
n
b
e
ni
-
n
y
a
--
nbe
To
der
i
ve
al
l
t
he
pro
nunciat
ion
s
f
or
the
num
erals,
we
ha
ve
prepa
re
d
different
gro
ups
c
on
si
der
i
ng
th
e
above
disc
us
se
d
sim
il
arities.
We
ha
ve
cl
assi
fied
t
he
pro
noun
cea
ble
unit
s
to
be
int
o
t
hr
e
e
sta
te
s
of
gro
up
s
as:
Be
gin
sta
te
(B
)
,
Mi
dd
le
sta
te
(M
)
an
d
E
nd
st
at
e
(E)
.
De
pe
ndin
g
on
t
he
posit
ion
of
th
e
nu
m
ber
i.e.
unit
or
10
th
,
the
sta
te
s
are
de
te
rm
ined
a
nd
the
pr
onunci
at
ion
is
de
rive
d.
Fo
r
exam
ple
f
or
obta
inin
g
t
he
pro
nunciat
io
n
of
a
nu
m
ber
N
ha
vi
ng
le
ng
t
h
L,
a
s
{n
1
,
n
2
,
…
n
L
},
The
re
exist
3
s
ta
te
s
re
pr
es
entat
ives
of
th
e
pro
nunciat
io
ns
,
{B
,
M,
E}
fo
r
the
un
it
an
d
10
th
posit
ion
s
,
n
L
an
d
n
L
-
1
re
sp
ect
i
vely
and
the
unit
s
from
n
1
to
n
L
-
1
m
a
y
be
der
ive
d
us
in
g
the
com
m
on
pr
onunci
a
ti
on
ru
le
s
by
con
cat
e
natin
g
100
th
,
1000
th
,
et
c
po
sit
ion’s
pr
onunci
at
ion
wi
th
the
up
t
o
10
0
pr
onunci
at
ion.
F
or
exam
ple,
in
pro
du
ci
ng
t
he
pro
nunciat
ion
of
t
he
num
eral
11
a
s
“
ek
-
ga
-
ra
”
in
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A Co
ntext
-
base
d
N
um
er
al Re
adin
g
Tec
hniq
ue
for Tex
t
…
(
Soumy
a
Priy
adarsi
ni P
anda
)
4539
Od
ia
la
ngua
ge
the
un
it
s
invo
lved
in
the
pro
nunciat
ion
a
re
“
\
ek”
from
“B
set
”,
“
\
ga”
fro
m
“M
se
t”
and
“
\
r
a”
from
“E
se
t”
a
s
sh
ow
n
in
Fig
ur
e
3.
i
.e
B(L)
-
M(L
-
1)
-
E
(L
-
1).
H
ow
e
ve
r
the
above
re
petit
i
ve
pro
nunciat
ion
i
s
no
t
sam
e
for
num
ber
s
ha
ving
a
9
or
0
at
the
unit
place.
T
o
overc
om
e
this
we
se
pa
rate
the
nu
m
ber
s
of
these
cat
egory
f
ro
m
the
gr
oups
a
nd
pro
duce
thei
r
pro
nunciat
ion
by
m
ai
ntaining
sp
eci
al
c
ases
of
pro
nunciat
i
on
as
{9:
“
na”,
19
:
“une
-
is”
,
29
:
“
ana
-
ti
ris
”,
et
c},
{2
0:
“k
o
-
di
e”
,
30
:
“ti
ris”
,
et
c}
H
oweve
r,
for
s
om
e
unit
s,
the
pro
nunciat
ion
ru
le
s
do
e
s
not
include
t
he
m
i
dd
le
sta
te
,
f
or
exam
ple
fo
r
t
he
nu
m
eral
two
at
the
un
it
place,
the
m
app
in
g
m
ay
be
“
\
ba”
f
r
om
t
he
B st
at
e
an
d
the
ne
xt one
is from
the
E
sta
te
as
“
\
ra”
to
f
orm
the
pro
nunc
ia
ti
on
“ba
-
ra”
for
th
e
nu
m
eral “1
2”.
Table
10. Pron
un
ci
at
io
n of B
eng
al
i
Nu
m
erals with
Starti
ng (
Col
um
n W
ise
)
and E
nd
i
ng (
R
ow
W
ise
)
Sim
i
la
rity
Pron
u
n
ciatio
n
aek
b
aa
te
ch
o
pon
ch
h
o
sh
at
ath
una
ra
Aeg
-
aa
-
ro
Baa
-
ro
te
-
ro
ch
o
-
ddo
pon
-
e
-
ro
so
-
ha
-
la
sh
at
-
a
-
ra
ath
-
a
-
ra
Un
e
-
ish
ish
aek
-
u
sh
b
aa
-
ish
te
-
ish
ch
o
-
b
ish
po
-
ch
-
ish
ch
h
o
-
b
ish
sh
at
-
aa
-
sh
ath
-
aa
-
sh
Un
o
-
tirish
rish
aek
-
ti
-
rish
bo
-
t
-
rish
te
-
ti
-
rish
ch
o
u
-
ti
-
rish
poy
-
ti
-
rish
ch
h
o
-
ti
-
rish
sh
ai
-
ti
-
rish
-
att
-
i r
ish
Un
o
-
ch
o
lish
lish
aek
-
ch
o
-
lish
bi
-
ya
-
lish
te
-
ta
-
lish
ch
u
-
ya
-
lish
poy
-
ta
-
lish
ch
h
e
-
ch
h
o
-
lish
sh
at
-
ch
o
-
lish
-
att
-
ch
o
lish
uno
-
pon
-
ch
aash
nno
aekaa
-
nno
Bah
a
-
nno
ti
-
p
an
n
o
ch
u
-
ya
-
nno
pon
-
ch
a
-
nno
ch
h
a
-
p
an
n
o
sh
at
-
aa
-
nno
att
-
aa
-
nno
uno
-
sh
at
sh
o
tti
aek
-
sh
o
tti
ba
-
sh
o
tti
te
-
sh
o
tti
ch
o
u
-
sh
o
tti
poy
-
sh
o
tti
ch
h
e
-
sh
o
tti
sh
at
-
sh
o
tti
att
-
sh
o
tti
Un
o
-
sh
o
tto
r
tto
r
aek
-
a
-
tto
r
b
ah
a
-
tto
r
ti
-
y
a
tto
r
ch
u
-
ya
-
tto
r
po
-
ch
a
-
tto
r
ch
h
i
-
ya
-
tto
r
sh
at
-
aa
-
tto
r
att
-
aa
-
t
to
r
Un
o
-
ash
i
ash
i
aek
-
aa
-
ash
i
bi
-
r
-
ash
i
tir
-
ash
i
ch
u
-
ra
-
ash
i
po
-
ch
a
-
ash
i
ch
h
i
-
ya
-
ash
i
sh
at
-
a
-
ash
i
ata
-
ash
i
uno
-
no
-
bboi
n
o
b
b
o
i
aek
-
aa
-
n
o
b
b
o
i
bi
-
ra
-
n
o
b
b
o
i
tir
-
a
-
n
o
b
b
o
i
ch
u
-
ra
-
n
o
b
b
o
i
po
-
ch
aa
-
n
o
b
b
o
i
ch
h
i
-
y
aa
-
n
o
b
b
o
i
sh
at
-
aa
-
n
o
b
b
o
i
ataa
-
n
o
b
b
o
i
n
ira
-
n
o
b
b
o
i
Be
gin
Mi
ddle
E
nd
Figure
3. P
os
si
ble stat
es of a
nu
m
eral i
n Od
i
a langua
ge
The
upto
10
0
pr
on
un
ci
at
io
n
m
ay
fail
fo
r
certai
n
num
er
al
s,
e.g
:
co
ns
ider
the
num
eral
14
pro
noun
ced
as
“chau
-
da”.
T
his
do
es
not
fo
ll
ow
the
pro
nunc
ia
ti
on
si
m
il
ar
it
ie
s.
An
obvi
ou
s
(br
ute
forc
e)
w
orkaro
und
is
to
hav
e
a
sm
al
l
dicti
on
ary
of
suc
h
dis
-
sim
il
ar
un
it
s,
an
d
c
hec
k
wh
et
her
a
giv
en
num
ber
m
at
c
hes
a
ny
of
t
hem
at
the
be
ginnin
g
of
te
xt
a
naly
sis
phase.
I
f
s
o,
break
it
up
int
o
the
co
rr
e
spo
nding
pro
nounce
able
unit
s
se
pa
ratel
y
and
par
se them
to
the n
ext p
ha
se sep
aratel
y. This works sati
sfacto
rily
, an
d
we'
ve
i
m
ple
m
e
nted
this with
a few
nu
m
erals (
14
-
“
ch
au
-
da”, 1
6
-
“
so
-
ha
-
la
”,
35
-
“
pain
-
tir
-
is”
, 5
3
-
“t
e
-
pan
”
, 56
-
“chh
a
-
pan
”
, et
c).
3.3.
Speech
D
atabase
M
ap
pin
g And
W
av
e
for
m Conc
at
e
nat
ion
As
the
m
od
el
us
es
the
W
CT
te
chn
iq
ue
to
pro
du
ce
t
he
de
sired
ou
t
pu
t
s
peech,
the
re
s
pecti
ve
ba
se
so
un
d
unit
s
in
the
s
peec
h
da
ta
base
ar
e
nee
ded
to
be
obta
ined
f
or
pe
rfo
rm
ing
wa
veform
con
cat
enati
on
to
pro
du
ce
the
outp
ut
sp
eec
h.
The
s
peech
da
ta
base
m
app
in
g
phase
ide
ntifie
s
the
res
pec
ti
ve
sp
eech
da
ta
base
un
it
s
to
pe
rform
ru
le
base
d
w
aveform
con
ca
te
nation.
The
W
CT
te
ch
niqu
e
is
then
us
e
d
to
pr
oduce
the
desir
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4533
-
4544
4540
ou
t
pu
t
s
peec
h
for
the
Od
ia
num
eral.
Figu
re
4
sho
ws
the
portio
n
c
on
cat
e
nation
proces
s
for
the
num
eral
“1”
pro
nounced
as
“ae
-
ka”
in O
dia langua
ge fr
om
the two s
pee
ch data
base
un
it
s“
\
ae” and
“
\
ka”.
Portion
from
“
\
a
e”
sound
Portion
from
“
\
ka
”
s
ound
Figure
4.
W
a
ve
p
at
te
r
n of
nu
m
eral “1”
(one
)
in
Odia (
‘
୧
’
)
pro
nounced
as
“ae
-
ka”
3.4
.
ILL
US
TR
AT
ION
The
co
ntext
id
entifi
cat
ion
pr
ocess
f
or
a
n
O
dia
la
ngua
ge
num
eral
is
pr
es
ented
in
Fig
ur
e
5
and
the
sp
eec
h
unit
identific
at
ion
/m
a
pp
i
ng
ste
p
in
volve
d
f
or
producin
g
the
num
eral
pro
nunciat
ion
f
ro
m
the
base
35
sp
eec
h
un
it
s
is
presente
d
i
n
Figure
6.
I
n
producin
g
t
he
ou
tpu
t
s
peec
h
for
the
nu
m
erals,
the
discu
ssed
up
to
100
pro
nunci
at
ion
ru
le
is
us
e
d
t
o
fin
d
t
he
e
qu
i
valent
c
har
act
er
unit
s
in
volv
ed.
The
sam
e
port
i
on
con
cat
e
natio
n m
et
ho
d
is
u
se
d t
o pro
du
ce
the
f
inal
outp
ut speec
h.
Figure
5
.
N
ume
ral co
ntext
i
de
ntific
at
ion
f
or
input n
um
eral 943
435245
4
in
Odia l
an
gu
a
ge
Figure
6.
N
ume
ral sou
nd pr
oductio
n f
or
i
nput num
eral 4
in
Od
ia
la
ngua
ge
4.
RESUL
T A
NALYS
IS
The
pro
posed
nu
m
eral
rea
ding
te
c
hn
i
qu
e
an
d
th
e
W
C
T
te
c
hn
i
qu
e
is
im
ple
m
ented
in
C/
C+
+
an
d
is
bein
g
te
ste
d
for
pro
du
ci
ng
dif
fer
e
nt ty
pes
of
nu
m
erals in d
if
fer
e
nt co
nte
xt in
the conside
re
d
India
n
la
ngua
ges.
To
a
naly
ze
the
qual
it
y
of
the
pro
du
ce
d
s
pee
ch,
t
he
Me
an
Op
i
nion
Sc
ore
(MOS)
te
st
[
30
]
is
con
si
der
e
d
al
ong
with
the
stora
ge
an
d
exec
utio
n
ti
m
e
with
res
pect
to
t
he
exi
sti
ng
syl
la
ble
ba
sed
te
xt
t
o
s
pe
ech
te
ch
nique
[19].
The det
ai
ls o
f
t
he results
obta
ined
are
d
isc
us
s
ed belo
w.
4.1
.
St
or
age
Req
uire
ment
Wh
il
e
the
syl
la
ble
ba
sed
te
c
hn
i
qu
e
s
re
quir
es
ar
ound
800
sp
eec
h
un
it
s
of
syl
la
ble
unit
s
requiri
ng
a
m
e
m
or
y
of
ar
ound
1
-
2
MB
in
com
pr
essed
form
at
,
the
W
CT
te
ch
niqu
e
that
produce
s
the
sp
eec
h
s
egm
ents
from
the
basic
35
s
peech
un
it
s
re
qu
i
rin
g
a
m
e
m
or
y
of
a
r
ound
23
5
KB
only
with
ou
t
f
ur
th
er
c
om
pr
e
ssio
ns.
N
o
oth
e
r
unit
s
are
require
d
to
be
add
e
d
to
t
he
database
for
pro
du
ci
ng
t
he
nu
m
eral
pronu
nciat
ion
in
dif
fer
e
nt
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
A Co
ntext
-
base
d
N
um
er
al Re
adin
g
Tec
hniq
ue
for Tex
t
…
(
Soumy
a
Priy
adarsi
ni P
anda
)
4541
con
te
xt.
Assu
m
ing
the
t
otal
stora
ge
requi
red
by
the
syl
la
ble
-
base
d
te
chn
i
qu
e
to
be
100%
,
the
pr
opos
e
d
te
chn
iq
ue
ac
hi
eves
81%
re
du
ct
ion
in
t
otal
s
tora
ge
re
quire
m
ent
and
91%
reducti
on
in
num
ber
of
un
it
s
in
the
sp
eec
h databas
e com
par
ed
to t
he
existi
ng tec
hn
i
qu
e
.
4.2.
Execu
tion T
im
e
To
a
naly
ze
the
perform
ance
of
t
he
pr
opos
e
d
te
ch
nique
in
te
rm
s
of
execut
ion
ti
m
e
co
m
par
e
d
to
t
he
syl
la
ble
based
te
chn
iq
ue,
different
te
xt
file
s
are
pr
e
par
e
d
c
on
ta
ini
ng
nu
m
erals
in
diff
e
re
nt
con
te
xt
of
it
s
us
e
.
By
var
yi
ng
the
nu
m
ber
of
nu
m
erals
in
each
file
fr
om
10
t
o
100,
the
exe
cution
ti
m
e
(in
m
s)
is
m
easur
ed
by
bo
t
h
the
te
c
hniqu
es
.
Fig
ure
10,
Fi
gure
11
and
Fig
ur
e
12
s
hows
the
a
ve
rag
e
e
xec
utio
n
ti
m
e
for
bot
h
the
te
chn
iq
ues
.
Th
e
resu
lt
s
al
l
th
e
exp
e
rim
ents
perform
ed
sh
ows
the
e
xpone
ntial
increase
of
e
xec
ution
ti
m
e,
du
e
to
the
inc
reas
e
in
num
ber
of
deco
m
pr
essi
on
to
the
.
gsm
file
s
in
the
syl
la
ble
base
d
te
c
hn
i
qu
e
,
wh
il
e
the
pro
po
se
d
a
pp
r
oach s
hows rel
at
ively
v
ery lo
w growt
h rate
in all
the
scena
r
ios
te
ste
d
.
Figure
10. E
xe
cution t
im
e fo
r
syl
la
ble
-
base
d an
d pro
po
se
d
t
echn
i
qu
e
w
it
h resp
ect
t
o
in
cre
asi
ng num
ber
of
words in
Odia
la
nguag
e
Figure
11. E
xe
cution t
im
e fo
r
syl
la
ble
-
base
d an
d pro
po
se
d
t
echn
i
qu
e
w
it
h resp
ect
t
o
in
cre
asi
ng num
ber
of
words in
H
i
nd
i
langua
ge
Figure
12. E
xe
cution t
im
e fo
r
syl
la
ble
-
base
d an
d pro
po
se
d
t
echn
i
qu
e
w
it
h resp
ect
t
o
in
cre
asi
ng num
ber
of
words in
Ben
ga
li
lang
ua
ge
0
20
40
60
80
100
10
20
30
40
50
60
70
80
90
100
Ex
ecution
Tim
e
(
in m
s)
Num
ber
o
f
num
eral
s
(O
dia
)
Sy
l
l
ab
l
e
-
b
a
s
ed
pr
opos
ed
0
20
40
60
80
100
10
20
30
40
50
60
70
80
90
100
Ex
ecution
Tim
e
(in
m
s)
Num
ber
o
f
num
eral
s
(H
indi)
Sy
l
l
ab
l
e
-
b
a
s
ed
pr
opos
ed
0
20
40
60
80
100
10
20
30
40
50
60
70
80
90
100
Ex
ecution
Tim
e
(in
m
s)
Num
ber
o
f
num
eral
s(Beng
a
li)
Sy
l
l
ab
l
e
-
b
a
s
ed
pr
opos
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
8
, N
o.
6
,
Dece
m
ber
2
01
8
:
4533
-
4544
4542
4.1
.
Sub
jecti
ve
Me
as
ure
For Speech
Qu
al
ity
Fo
r
perf
or
m
ing
the
MOS
te
sts,
a
set
of
ra
ndom
nu
m
eral
s,
N
1
,
N
2
….
N
8
are
sel
ect
ed
r
epr
ese
ntin
g
diff
e
re
nt
cat
eg
or
y
of
pro
nunc
ia
ti
on
s
for
t
he
sp
eci
fied
r
ules.
The
ou
t
pu
t
spe
ech
is
ge
nerat
ed
by
the
propos
e
d
te
chn
iq
ue
as
well
as
by
the
syl
la
ble
-
base
d
nu
m
eral
read
i
ng
te
c
hniq
ue.
A
gr
oup
of
15
nati
ve
sp
ea
ke
rs
ar
e
sel
ect
ed
from
each
la
ng
uag
e
to
pe
rfor
m
the
li
ste
ner
s
te
st
and
are
as
ke
d
to
giv
e
t
heir
feedback
on
the
ba
sis
of
ease
of
unde
rstand
a
bili
ty
on
outp
ut
s
peech
pro
duced
by
t
he
two
te
ch
nique
s
in
a
5
point
s
cal
e
(1
-
ve
ry
lo
w,
2
-
low,
3
-
a
ver
a
ge
,
4
-
high,
5:
very
hig
h)
.
All
th
e
te
sts
wer
e
pe
rfor
m
ed
with
a
headpho
ne
set
.
Figure.
7,
Fi
g.
8
an
d
Fig.
9
show
s
the
ave
rag
e
M
OS
te
st
res
ults
by
al
l
li
s
te
ners
for
dif
fer
e
nt
nu
m
erals
resp
e
ct
ively
fo
r
the
t
hr
ee
consi
der
e
d
la
ngua
ges
.
The
re
su
lt
s
of
al
l
the
exp
e
rim
ents
per
f
or
m
ed
sh
ow
the
eff
ect
ive
ne
ss
of
the
pro
po
s
ed
te
chn
iq
ue
i
n producin
g
c
om
par
able
res
ults wi
th the e
xisti
ng
te
chn
iq
ue e
ve
n wit
h
a
v
e
ry s
m
al
l database.
Figure
7
.
A
verage M
OS
for Odia la
ngua
ge
Figure
8
.
A
verage M
OS
for Hin
di lan
gu
a
ge
Figure
9
.
A
verage M
OS
for
B
eng
al
i l
a
ngua
ge
1
1.5
2
2.5
3
3.5
4
4.5
N1
N2
N3
N4
N5
N6
N7
N8
MO
S
for
Odia
Un
it
Ty
pe
(Nu
me
ral
s)
Pr
o
p
o
sed
Sy
llab
le-
b
ased
1
1.5
2
2.5
3
3.5
4
4.5
N1
N2
N3
N4
N5
N6
N7
N8
MO
S
for
Hi
nd
i
Un
it
ty
pe(Nu
me
ral
s)
Pr
o
p
o
sed
Sy
llab
le-
b
ased
1
1.5
2
2.5
3
3.5
4
4.5
N1
N2
N3
N4
N5
N6
N7
N8
MO
S
for
Bengali
Un
it
Ty
pe(Nu
me
ral
s)
Pr
o
p
o
sed
Sy
llab
le-
b
ased
Evaluation Warning : The document was created with Spire.PDF for Python.