<s>
Voice	B-General_Concept
activity	I-General_Concept
detection	I-General_Concept
(	O
VAD	O
)	O
,	O
also	O
known	O
as	O
speech	O
activity	O
detection	O
or	O
speech	B-General_Concept
detection	I-General_Concept
,	O
is	O
the	O
detection	O
of	O
the	O
presence	O
or	O
absence	O
of	O
human	O
speech	O
,	O
used	O
in	O
speech	B-Algorithm
processing	I-Algorithm
.	O
</s>
<s>
The	O
main	O
uses	O
of	O
VAD	O
are	O
in	O
speaker	B-Application
diarization	I-Application
,	O
speech	B-Application
coding	I-Application
and	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
It	O
can	O
facilitate	O
speech	B-Algorithm
processing	I-Algorithm
,	O
and	O
can	O
also	O
be	O
used	O
to	O
deactivate	O
some	O
processes	O
during	O
non-speech	O
section	O
of	O
an	O
audio	O
session	O
:	O
it	O
can	O
avoid	O
unnecessary	O
coding/transmission	O
of	O
silence	O
packets	B-Protocol
in	O
Voice	B-Application
over	I-Application
Internet	I-Application
Protocol	I-Application
(	O
VoIP	B-Application
)	O
applications	O
,	O
saving	O
on	O
computation	O
and	O
on	O
network	O
bandwidth	O
.	O
</s>
<s>
Voice	B-General_Concept
activity	I-General_Concept
detection	I-General_Concept
is	O
usually	O
independent	O
of	O
language	O
.	O
</s>
<s>
It	O
was	O
first	O
investigated	O
for	O
use	O
on	O
time-assignment	B-Protocol
speech	I-Protocol
interpolation	I-Protocol
(	O
TASI	O
)	O
systems	O
.	O
</s>
<s>
A	O
classification	B-General_Concept
rule	I-General_Concept
is	O
applied	O
to	O
classify	O
the	O
section	O
as	O
speech	O
or	O
non-speech	O
–	O
often	O
this	O
classification	B-General_Concept
rule	I-General_Concept
finds	O
when	O
a	O
value	O
exceeds	O
a	O
certain	O
threshold	O
.	O
</s>
<s>
In	O
these	O
difficult	O
detection	O
conditions	O
it	O
is	O
often	O
preferable	O
that	O
a	O
VAD	O
should	O
fail-safe	B-General_Concept
,	O
indicating	O
speech	O
detected	O
when	O
the	O
decision	O
is	O
in	O
doubt	O
,	O
to	O
lower	O
the	O
chance	O
of	O
losing	O
speech	O
segments	O
.	O
</s>
<s>
VAD	O
is	O
an	O
integral	O
part	O
of	O
different	O
speech	O
communication	O
systems	O
such	O
as	O
audio	O
conferencing	O
,	O
echo	O
cancellation	O
,	O
speech	B-Application
recognition	I-Application
,	O
speech	B-Application
encoding	I-Application
,	O
speaker	B-Application
recognition	I-Application
and	O
hands-free	O
telephony	O
.	O
</s>
<s>
Similarly	O
,	O
in	O
Universal	B-Protocol
Mobile	I-Protocol
Telecommunications	I-Protocol
Systems	I-Protocol
(	O
UMTS	B-Protocol
)	O
,	O
it	O
controls	O
and	O
reduces	O
the	O
average	O
bit	O
rate	O
and	O
enhances	O
overall	O
coding	O
quality	O
of	O
speech	O
.	O
</s>
<s>
In	O
cellular	O
radio	O
systems	O
(	O
for	O
instance	O
GSM	O
and	O
CDMA	B-Protocol
systems	O
)	O
based	O
on	O
Discontinuous	O
Transmission	O
(	O
DTX	O
)	O
mode	O
,	O
VAD	O
is	O
essential	O
for	O
enhancing	O
system	O
capacity	O
by	O
reducing	O
co-channel	O
interference	O
and	O
power	O
consumption	O
in	O
portable	O
digital	O
devices	O
.	O
</s>
<s>
In	O
speech	B-Algorithm
processing	I-Algorithm
applications	O
,	O
voice	B-General_Concept
activity	I-General_Concept
detection	I-General_Concept
plays	O
an	O
important	O
role	O
since	O
non-speech	O
frames	O
are	O
often	O
discarded	O
.	O
</s>
<s>
In	O
VoIP	B-Application
applications	O
,	O
front-end	O
clipping	O
can	O
be	O
reduced	O
by	O
rewinding	O
to	O
shortly	O
before	O
the	O
detection	O
and	O
sending	O
very	O
slightly	O
delayed	O
data	O
.	O
</s>
<s>
The	O
G.729	B-Algorithm
standard	O
calculates	O
the	O
following	O
features	O
for	O
its	O
VAD	O
:	O
line	B-Algorithm
spectral	I-Algorithm
frequencies	I-Algorithm
,	O
full-band	O
energy	O
,	O
low-band	O
energy	O
(	O
<	O
1kHz	O
)	O
,	O
and	O
zero-crossing	B-Algorithm
rate	I-Algorithm
.	O
</s>
<s>
The	O
Speex	B-Application
audio	O
compression	O
library	O
uses	O
a	O
procedure	O
named	O
Improved	O
Minima	O
Controlled	O
Recursive	O
Averaging	O
,	O
which	O
uses	O
a	O
smoothed	O
representation	O
of	O
spectral	O
power	O
and	O
then	O
looks	O
at	O
the	O
minima	O
of	O
a	O
smoothed	O
periodogram	O
.	O
</s>
<s>
Lingua	B-Application
Libre	I-Application
,	O
a	O
Wikimedia	O
tool	O
and	O
project	O
of	O
language	O
documentation	O
,	O
using	O
VAD	O
to	O
allow	O
recording	O
many	O
pronunciations	O
in	O
a	O
short	O
amount	O
of	O
time	O
.	O
</s>
