<s>
Speaker	B-Application
recognition	I-Application
is	O
the	O
identification	O
of	O
a	O
person	O
from	O
characteristics	O
of	O
voices	O
.	O
</s>
<s>
The	O
term	O
voice	O
recognition	O
can	O
refer	O
to	O
speaker	B-Application
recognition	I-Application
or	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
Speaker	B-Application
verification	I-Application
(	O
also	O
called	O
speaker	B-Application
authentication	I-Application
)	O
contrasts	O
with	O
identification	O
,	O
and	O
speaker	B-Application
recognition	I-Application
differs	O
from	O
speaker	B-Application
diarisation	I-Application
(	O
recognizing	O
when	O
the	O
same	O
speaker	O
is	O
speaking	O
)	O
.	O
</s>
<s>
Speaker	B-Application
recognition	I-Application
has	O
a	O
history	O
dating	O
back	O
some	O
four	O
decades	O
as	O
of	O
2019	O
and	O
uses	O
the	O
acoustic	O
features	O
of	O
speech	O
that	O
have	O
been	O
found	O
to	O
differ	O
between	O
individuals	O
.	O
</s>
<s>
There	O
are	O
two	O
major	O
applications	O
of	O
speaker	B-Application
recognition	I-Application
technologies	O
and	O
methodologies	O
.	O
</s>
<s>
In	O
a	O
sense	O
,	O
speaker	B-Application
verification	I-Application
is	O
a	O
1:1	O
match	O
where	O
one	O
speaker	O
's	O
voice	O
is	O
matched	O
to	O
a	O
particular	O
template	O
whereas	O
speaker	B-Application
identification	I-Application
is	O
a	O
1:N	O
match	O
where	O
the	O
voice	O
is	O
compared	O
against	O
multiple	O
templates	O
.	O
</s>
<s>
Speaker	B-Application
verification	I-Application
is	O
usually	O
employed	O
as	O
a	O
"	O
gatekeeper	O
"	O
in	O
order	O
to	O
provide	O
access	O
to	O
a	O
secure	O
system	O
.	O
</s>
<s>
Speaker	B-Application
identification	I-Application
systems	O
can	O
also	O
be	O
implemented	O
covertly	O
without	O
the	O
user	O
's	O
knowledge	O
to	O
identify	O
talkers	O
in	O
a	O
discussion	O
,	O
alert	O
automated	O
systems	O
of	O
speaker	O
changes	O
,	O
check	O
if	O
a	O
user	O
is	O
already	O
enrolled	O
in	O
a	O
system	O
,	O
etc	O
.	O
</s>
<s>
In	O
forensic	O
applications	O
,	O
it	O
is	O
common	O
to	O
first	O
perform	O
a	O
speaker	B-Application
identification	I-Application
process	O
to	O
create	O
a	O
list	O
of	O
"	O
best	O
matches	O
"	O
and	O
then	O
perform	O
a	O
series	O
of	O
verification	O
processes	O
to	O
determine	O
a	O
conclusive	O
match	O
.	O
</s>
<s>
Each	O
speaker	B-Application
recognition	I-Application
system	O
has	O
two	O
phases	O
:	O
enrollment	O
and	O
verification	O
.	O
</s>
<s>
Speaker	B-Application
recognition	I-Application
systems	O
fall	O
into	O
two	O
categories	O
:	O
text-dependent	O
and	O
text-independent	O
.	O
</s>
<s>
They	O
are	O
most	O
often	O
used	O
for	O
speaker	B-Application
identification	I-Application
as	O
they	O
require	O
very	O
little	O
if	O
any	O
cooperation	O
by	O
the	O
speaker	O
.	O
</s>
<s>
As	O
text-independent	O
technologies	O
do	O
not	O
compare	O
what	O
was	O
said	O
at	O
enrollment	O
and	O
verification	O
,	O
verification	O
applications	O
tend	O
to	O
also	O
employ	O
speech	B-Application
recognition	I-Application
to	O
determine	O
what	O
the	O
user	O
is	O
saying	O
at	O
the	O
point	O
of	O
authentication	O
.	O
</s>
<s>
Speaker	B-Application
recognition	I-Application
is	O
a	O
pattern	O
recognition	O
problem	O
.	O
</s>
<s>
The	O
various	O
technologies	O
used	O
to	O
process	O
and	O
store	O
voice	O
prints	O
include	O
frequency	O
estimation	O
,	O
hidden	O
Markov	O
models	O
,	O
Gaussian	O
mixture	O
models	O
,	O
pattern	B-Language
matching	I-Language
algorithms	O
,	O
neural	B-Architecture
networks	I-Architecture
,	O
matrix	B-Algorithm
representation	I-Algorithm
,	O
vector	O
quantization	O
and	O
decision	B-Algorithm
trees	I-Algorithm
.	O
</s>
<s>
Linear	B-Algorithm
predictive	I-Algorithm
coding	I-Algorithm
(	O
LPC	O
)	O
is	O
a	O
speech	B-Application
coding	I-Application
method	O
used	O
in	O
speaker	B-Application
recognition	I-Application
and	O
speech	B-General_Concept
verification	I-General_Concept
.	O
</s>
<s>
Due	O
to	O
the	O
introduction	O
of	O
legislation	O
like	O
the	O
General	O
Data	O
Protection	O
Regulation	O
in	O
the	O
European	O
Union	O
and	O
the	O
California	O
Consumer	O
Privacy	O
Act	O
in	O
the	O
United	O
States	O
,	O
there	O
has	O
been	O
much	O
discussion	O
about	O
the	O
use	O
of	O
speaker	B-Application
recognition	I-Application
in	O
the	O
work	O
place	O
.	O
</s>
<s>
In	O
September	O
2019	O
Irish	O
speech	B-Application
recognition	I-Application
developer	O
Soapbox	O
Labs	O
warned	O
about	O
the	O
legal	O
implications	O
that	O
may	O
be	O
involved	O
.	O
</s>
<s>
Between	O
1996	O
and	O
1998	O
,	O
speaker	B-Application
recognition	I-Application
technology	O
was	O
used	O
at	O
the	O
Scobey	O
–	O
Coronach	O
Border	O
Crossing	O
to	O
enable	O
enrolled	O
local	O
residents	O
with	O
nothing	O
to	O
declare	O
to	O
cross	O
the	O
Canada	O
–	O
United	O
States	O
border	O
when	O
the	O
inspection	O
stations	O
were	O
closed	O
for	O
the	O
night	O
.	O
</s>
<s>
In	O
May	O
2013	O
it	O
was	O
announced	O
that	O
Barclays	O
Wealth	O
was	O
to	O
use	O
passive	O
speaker	B-Application
recognition	I-Application
to	O
verify	O
the	O
identity	O
of	O
telephone	O
customers	O
within	O
30	O
seconds	O
of	O
normal	O
conversation	O
.	O
</s>
<s>
The	O
system	O
used	O
had	O
been	O
developed	O
by	O
voice	O
recognition	O
company	O
Nuance	O
(	O
that	O
in	O
2011	O
acquired	O
the	O
company	O
Loquendo	O
,	O
the	O
spin-off	O
from	O
CSELT	O
itself	O
for	O
speech	O
technology	O
)	O
,	O
the	O
company	O
behind	O
Apple	O
's	O
Siri	B-Application
technology	O
.	O
</s>
<s>
The	O
private	O
banking	O
division	O
of	O
Barclays	O
was	O
the	O
first	O
financial	O
services	O
firm	O
to	O
deploy	O
voice	B-Application
biometrics	I-Application
as	O
the	O
primary	O
means	O
to	O
authenticate	O
customers	O
to	O
their	O
call	O
centers	O
.	O
</s>
<s>
Speaker	B-Application
recognition	I-Application
may	O
also	O
be	O
used	O
in	O
criminal	O
investigations	O
,	O
such	O
as	O
those	O
of	O
the	O
2014	O
executions	O
of	O
,	O
amongst	O
others	O
,	O
James	O
Foley	O
and	O
Steven	O
Sotloff	O
.	O
</s>
