<s>
Speech	B-Application
coding	I-Application
is	O
an	O
application	O
of	O
data	B-General_Concept
compression	I-General_Concept
of	O
digital	O
audio	O
signals	O
containing	O
speech	O
.	O
</s>
<s>
Speech	B-Application
coding	I-Application
uses	O
speech-specific	O
parameter	O
estimation	O
using	O
audio	B-Algorithm
signal	I-Algorithm
processing	I-Algorithm
techniques	O
to	O
model	O
the	O
speech	O
signal	O
,	O
combined	O
with	O
generic	O
data	B-General_Concept
compression	I-General_Concept
algorithms	I-General_Concept
to	O
represent	O
the	O
resulting	O
modeled	O
parameters	O
in	O
a	O
compact	O
bitstream	O
.	O
</s>
<s>
Some	O
applications	O
of	O
speech	B-Application
coding	I-Application
are	O
mobile	O
telephony	O
and	O
voice	B-Application
over	I-Application
IP	I-Application
(	O
VoIP	B-Application
)	O
.	O
</s>
<s>
The	O
most	O
widely	O
used	O
speech	B-Application
coding	I-Application
technique	O
in	O
mobile	O
telephony	O
is	O
linear	B-Algorithm
predictive	I-Algorithm
coding	I-Algorithm
(	O
LPC	O
)	O
,	O
while	O
the	O
most	O
widely	O
used	O
in	O
VoIP	B-Application
applications	O
are	O
the	O
LPC	O
and	O
modified	O
discrete	B-General_Concept
cosine	I-General_Concept
transform	I-General_Concept
(	O
MDCT	O
)	O
techniques	O
.	O
</s>
<s>
The	O
techniques	O
employed	O
in	O
speech	B-Application
coding	I-Application
are	O
similar	O
to	O
those	O
used	O
in	O
audio	O
data	B-General_Concept
compression	I-General_Concept
and	O
audio	O
coding	O
where	O
knowledge	O
in	O
psychoacoustics	O
is	O
used	O
to	O
transmit	O
only	O
data	O
that	O
is	O
relevant	O
to	O
the	O
human	O
auditory	O
system	O
.	O
</s>
<s>
For	O
example	O
,	O
in	O
voiceband	O
speech	B-Application
coding	I-Application
,	O
only	O
information	O
in	O
the	O
frequency	O
band	O
400	O
to	O
3500Hz	O
is	O
transmitted	O
but	O
the	O
reconstructed	O
signal	O
is	O
still	O
adequate	O
for	O
intelligibility	O
.	O
</s>
<s>
Speech	B-Application
coding	I-Application
differs	O
from	O
other	O
forms	O
of	O
audio	O
coding	O
in	O
that	O
speech	O
is	O
a	O
simpler	O
signal	O
than	O
most	O
other	O
audio	O
signals	O
,	O
and	O
a	O
lot	O
more	O
statistical	O
information	O
is	O
available	O
about	O
the	O
properties	O
of	O
speech	O
.	O
</s>
<s>
As	O
a	O
result	O
,	O
some	O
auditory	O
information	O
that	O
is	O
relevant	O
in	O
audio	O
coding	O
can	O
be	O
unnecessary	O
in	O
the	O
speech	B-Application
coding	I-Application
context	O
.	O
</s>
<s>
In	O
speech	B-Application
coding	I-Application
,	O
the	O
most	O
important	O
criterion	O
is	O
preservation	O
of	O
intelligibility	O
and	O
pleasantness	O
of	O
speech	O
,	O
with	O
a	O
constrained	O
amount	O
of	O
transmitted	O
data	O
.	O
</s>
<s>
Speech	B-Application
coders	I-Application
are	O
of	O
two	O
types	O
:	O
</s>
<s>
The	O
A-law	B-Algorithm
and	O
μ-law	B-Algorithm
algorithms	I-Algorithm
(	O
G.711	B-Algorithm
)	O
used	O
in	O
traditional	O
PCM	B-Algorithm
digital	B-General_Concept
telephony	I-General_Concept
can	O
be	O
seen	O
as	O
an	O
earlier	O
precursor	O
of	O
speech	B-Application
encoding	I-Application
,	O
requiring	O
only	O
8	O
bits	O
per	O
sample	O
but	O
giving	O
effectively	O
12	O
bits	O
of	O
resolution	O
.	O
</s>
<s>
Although	O
this	O
would	O
generate	O
unacceptable	O
distortion	O
in	O
a	O
music	O
signal	O
,	O
the	O
peaky	O
nature	O
of	O
speech	O
waveforms	O
,	O
combined	O
with	O
the	O
simple	O
frequency	O
structure	O
of	O
speech	O
as	O
a	O
periodic	O
waveform	O
having	O
a	O
single	O
fundamental	O
frequency	O
with	O
occasional	O
added	O
noise	O
bursts	O
,	O
make	O
these	O
very	O
simple	O
instantaneous	O
compression	B-General_Concept
algorithms	I-General_Concept
acceptable	O
for	O
speech	O
.	O
</s>
<s>
A	O
wide	O
variety	O
of	O
other	O
algorithms	O
were	O
tried	O
at	O
the	O
time	O
,	O
mostly	O
delta	B-General_Concept
modulation	I-General_Concept
variants	O
,	O
but	O
after	O
careful	O
consideration	O
,	O
the	O
A-law/μ	O
-law	O
algorithms	O
were	O
chosen	O
by	O
the	O
designers	O
of	O
the	O
early	O
digital	B-General_Concept
telephony	I-General_Concept
systems	O
.	O
</s>
<s>
In	O
2008	O
,	O
G.711.1	O
codec	O
,	O
which	O
has	O
a	O
scalable	O
structure	O
,	O
was	O
standardized	O
by	O
ITU-T	O
.	O
</s>
<s>
At	O
the	O
same	O
time	O
,	O
far	O
more	O
processing	O
power	O
was	O
available	O
,	O
in	O
the	O
form	O
of	O
VLSI	O
circuits	O
,	O
than	O
was	O
available	O
for	O
earlier	O
compression	B-General_Concept
techniques	I-General_Concept
.	O
</s>
<s>
As	O
a	O
result	O
,	O
modern	O
speech	O
compression	B-General_Concept
algorithms	I-General_Concept
could	O
use	O
far	O
more	O
complex	O
techniques	O
than	O
were	O
available	O
in	O
the	O
1960s	O
to	O
achieve	O
far	O
higher	O
compression	O
ratios	O
.	O
</s>
<s>
The	O
most	O
widely	O
used	O
speech	B-Application
coding	I-Application
algorithms	O
are	O
based	O
on	O
linear	B-Algorithm
predictive	I-Algorithm
coding	I-Algorithm
(	O
LPC	O
)	O
.	O
</s>
<s>
In	O
particular	O
,	O
the	O
most	O
common	O
speech	B-Application
coding	I-Application
scheme	O
is	O
the	O
LPC-based	O
code-excited	B-Algorithm
linear	I-Algorithm
prediction	I-Algorithm
(	O
CELP	B-Algorithm
)	O
coding	O
,	O
which	O
is	O
used	O
for	O
example	O
in	O
the	O
GSM	O
standard	O
.	O
</s>
<s>
In	O
CELP	B-Algorithm
,	O
the	O
modeling	O
is	O
divided	O
in	O
two	O
stages	O
,	O
a	O
linear	O
predictive	O
stage	O
that	O
models	O
the	O
spectral	O
envelope	O
and	O
a	O
code-book-based	O
model	O
of	O
the	O
residual	O
of	O
the	O
linear	O
predictive	O
model	O
.	O
</s>
<s>
In	O
CELP	B-Algorithm
,	O
linear	O
prediction	O
coefficients	O
(	O
LPC	O
)	O
are	O
computed	O
and	O
quantized	O
,	O
usually	O
as	O
line	B-Algorithm
spectral	I-Algorithm
pairs	I-Algorithm
(	O
LSPs	O
)	O
.	O
</s>
<s>
In	O
addition	O
to	O
the	O
actual	O
speech	B-Application
coding	I-Application
of	O
the	O
signal	O
,	O
it	O
is	O
often	O
necessary	O
to	O
use	O
channel	O
coding	O
for	O
transmission	O
,	O
to	O
avoid	O
losses	O
due	O
to	O
transmission	O
errors	O
.	O
</s>
<s>
In	O
order	O
to	O
get	O
the	O
best	O
overall	O
coding	O
results	O
,	O
speech	B-Application
coding	I-Application
and	O
channel	O
coding	O
methods	O
are	O
chosen	O
in	O
pairs	O
,	O
with	O
the	O
more	O
important	O
bits	O
in	O
the	O
speech	O
data	O
stream	O
protected	O
by	O
more	O
robust	O
channel	O
coding	O
.	O
</s>
<s>
The	O
modified	O
discrete	B-General_Concept
cosine	I-General_Concept
transform	I-General_Concept
(	O
MDCT	O
)	O
,	O
a	O
type	O
of	O
discrete	B-General_Concept
cosine	I-General_Concept
transform	I-General_Concept
(	O
DCT	B-General_Concept
)	O
algorithm	O
,	O
was	O
adapted	O
into	O
a	O
speech	B-Application
coding	I-Application
algorithm	O
called	O
LD-MDCT	O
,	O
used	O
for	O
the	O
AAC-LD	B-Algorithm
format	O
introduced	O
in	O
1999	O
.	O
</s>
<s>
MDCT	O
has	O
since	O
been	O
widely	O
adopted	O
in	O
voice-over-IP	B-Application
(	O
VoIP	B-Application
)	O
applications	O
,	O
such	O
as	O
the	O
G.729.1	B-Algorithm
wideband	B-Algorithm
audio	I-Algorithm
codec	O
introduced	O
in	O
2006	O
,	O
Apple	O
's	O
FaceTime	B-Operating_System
(	O
using	O
AAC-LD	B-Algorithm
)	O
introduced	O
in	O
2010	O
,	O
and	O
the	O
CELT	B-Algorithm
codec	O
introduced	O
in	O
2011	O
.	O
</s>
<s>
Opus	B-Algorithm
is	O
a	O
free	B-Application
software	I-Application
audio	O
coder	O
.	O
</s>
<s>
It	O
combines	O
both	O
the	O
MDCT	O
(	O
CELT	B-Algorithm
)	O
and	O
LPC	O
(	O
SILK	O
)	O
audio	O
compression	B-General_Concept
algorithms	I-General_Concept
,	O
using	O
the	O
former	O
for	O
speech	O
.	O
</s>
<s>
It	O
is	O
widely	O
used	O
for	O
VoIP	B-Application
calls	I-Application
in	O
WhatsApp	B-Application
.	O
</s>
<s>
The	B-Operating_System
PlayStation	I-Operating_System
4	I-Operating_System
video	O
game	O
console	O
also	O
uses	O
Opus	B-Algorithm
for	O
its	O
PlayStation	B-Device
Network	I-Device
system	O
party	O
chat	O
.	O
</s>
<s>
Codec2	B-Algorithm
,	O
which	O
operates	O
at	O
bit	O
rates	O
as	O
low	O
as	O
450bit/s	O
,	O
sees	O
use	O
in	O
amateur	O
radio	O
.	O
</s>
<s>
NATO	O
currently	O
uses	O
MELPe	B-Algorithm
,	O
offering	O
legible	O
speech	O
600bit/s	O
(	O
with	O
one	O
nonstandard	O
variant	O
halving	O
the	O
number	O
)	O
.	O
</s>
<s>
Lyra	B-Application
by	O
Google	O
takes	O
an	O
unusual	O
machine	O
learning	O
approach	O
,	O
giving	O
"	O
almost	O
eerie	O
"	O
quality	O
at	O
3kbit/s	O
.	O
</s>
<s>
Microsoft	O
’s	O
Satin	B-Algorithm
also	O
uses	O
machine	O
learning	O
,	O
but	O
uses	O
a	O
higher	O
tunable	O
bitrate	O
and	O
is	O
wideband	O
.	O
</s>
