<s>
Julius	B-General_Concept
is	O
a	O
speech	B-Application
recognition	I-Application
engine	O
,	O
specifically	O
a	O
high-performance	O
,	O
two-pass	O
large	O
vocabulary	O
continuous	O
speech	B-Application
recognition	I-Application
(	O
LVCSR	O
)	O
decoder	O
software	O
for	O
speech-related	O
researchers	O
and	O
developers	O
.	O
</s>
<s>
It	O
can	O
perform	O
almost	O
real-time	B-General_Concept
computing	I-General_Concept
(	O
RTC	O
)	O
decoding	O
on	O
most	O
current	O
personal	B-Device
computers	I-Device
(	O
PCs	B-Device
)	O
in	O
60k	O
word	O
dictation	O
task	O
using	O
word	O
trigram	B-General_Concept
(	O
3-gram	O
)	O
and	O
context-dependent	O
Hidden	O
Markov	O
model	O
(	O
HMM	O
)	O
.	O
</s>
<s>
Standard	O
formats	O
are	O
adopted	O
to	O
cope	O
with	O
other	O
free	B-Application
modeling	O
toolkit	O
.	O
</s>
<s>
The	O
main	O
platform	O
is	O
Linux	B-Application
and	O
other	O
Unix	B-Application
workstations	O
,	O
and	O
it	O
works	O
on	O
Windows	B-Application
.	O
</s>
<s>
Julius	B-General_Concept
is	O
free	B-License
and	I-License
open-source	I-License
software	I-License
,	O
released	O
under	O
a	O
revised	O
BSD	B-Operating_System
style	O
software	B-License
license	I-License
.	O
</s>
<s>
Julius	B-General_Concept
has	O
been	O
developed	O
as	O
part	O
of	O
a	O
free	B-Application
software	I-Application
toolkit	O
for	O
Japanese	O
LVCSR	O
research	O
since	O
1997	O
,	O
and	O
the	O
work	O
has	O
been	O
continued	O
at	O
Continuous	O
Speech	B-Application
Recognition	I-Application
Consortium	O
(	O
CSRC	O
)	O
,	O
Japan	O
from	O
2000	O
to	O
2003	O
.	O
</s>
<s>
From	O
rev.3.4	O
,	O
a	O
grammar-based	O
recognition	O
parser	O
named	O
Julian	O
is	O
integrated	O
into	O
Julius	B-General_Concept
.	O
</s>
<s>
Julian	O
is	O
a	O
modified	O
version	O
of	O
Julius	B-General_Concept
that	O
uses	O
hand-designed	O
type	O
of	O
finite-state	B-Architecture
machine	I-Architecture
(	O
FSM	O
)	O
termed	O
a	O
deterministic	B-General_Concept
finite	I-General_Concept
automaton	I-General_Concept
(	O
DFA	B-General_Concept
)	O
grammar	O
as	O
a	O
language	B-Language
model	I-Language
.	O
</s>
<s>
It	O
can	O
be	O
used	O
to	O
build	O
a	O
kind	O
of	O
voice	B-Application
command	I-Application
system	O
of	O
small	O
vocabulary	O
,	O
or	O
various	O
spoken	O
dialog	B-General_Concept
system	I-General_Concept
tasks	O
.	O
</s>
<s>
To	O
run	O
,	O
the	O
Julius	B-General_Concept
recognizer	O
needs	O
a	O
language	B-Language
model	I-Language
and	O
an	O
acoustic	B-General_Concept
model	I-General_Concept
for	O
each	O
language	O
.	O
</s>
<s>
Julius	B-General_Concept
adopts	O
acoustic	B-General_Concept
models	I-General_Concept
in	O
Hidden	B-General_Concept
Markov	I-General_Concept
Model	I-General_Concept
Toolkit	I-General_Concept
(	O
HTK	B-General_Concept
)	O
ASCII	B-Protocol
format	O
,	O
pronunciation	O
dictionary	O
in	O
HTK-like	O
format	O
,	O
and	O
word	O
3-gram	O
language	B-Language
models	I-Language
in	O
ARPA	O
standard	O
format	O
:	O
forward	O
2-gram	O
and	O
reverse	O
3-gram	O
as	O
trained	O
from	O
speech	O
corpus	O
with	O
reversed	O
word	O
order	O
.	O
</s>
<s>
Although	O
Julius	B-General_Concept
is	O
only	O
distributed	O
with	O
Japanese	O
models	O
,	O
the	O
VoxForge	B-Application
project	O
is	O
working	O
to	O
create	O
English	O
acoustic	B-General_Concept
models	I-General_Concept
for	O
use	O
with	O
the	O
Julius	B-General_Concept
Speech	I-General_Concept
Recognition	I-General_Concept
Engine	I-General_Concept
.	O
</s>
<s>
In	O
April	O
2018	O
,	O
thanks	O
to	O
the	O
effort	O
of	O
Mozilla	B-Operating_System
foundation	O
,	O
a	O
350-hour	O
audio	O
corpus	O
of	O
spoken	O
English	O
was	O
made	O
available	O
.	O
</s>
