<s>
Audio	B-General_Concept
mining	I-General_Concept
is	O
a	O
technique	O
by	O
which	O
the	O
content	O
of	O
an	O
audio	O
signal	O
can	O
be	O
automatically	O
analyzed	O
and	O
searched	O
.	O
</s>
<s>
It	O
is	O
most	O
commonly	O
used	O
in	O
the	O
field	O
of	O
automatic	B-Application
speech	I-Application
recognition	I-Application
,	O
where	O
the	O
analysis	O
tries	O
to	O
identify	O
any	O
speech	O
within	O
the	O
audio	O
.	O
</s>
<s>
The	O
term	O
‘	O
audio	O
mining’	O
is	O
sometimes	O
used	O
interchangeably	O
with	O
audio	B-General_Concept
indexing	I-General_Concept
,	O
phonetic	B-General_Concept
searching	I-General_Concept
,	O
phonetic	B-General_Concept
indexing	I-General_Concept
,	O
speech	B-General_Concept
indexing	I-General_Concept
,	O
audio	B-General_Concept
analytics	I-General_Concept
,	O
speech	B-General_Concept
analytics	I-General_Concept
,	O
word	O
spotting	O
,	O
and	O
information	B-Library
retrieval	I-Library
.	O
</s>
<s>
Audio	B-General_Concept
indexing	I-General_Concept
,	O
however	O
,	O
is	O
mostly	O
used	O
to	O
describe	O
the	O
pre-process	O
of	O
audio	B-General_Concept
mining	I-General_Concept
,	O
in	O
which	O
the	O
audio	O
file	O
is	O
broken	O
down	O
into	O
a	O
searchable	O
index	O
of	O
words	O
.	O
</s>
<s>
Academic	O
research	O
on	O
audio	B-General_Concept
mining	I-General_Concept
began	O
in	O
the	O
late	O
1970s	O
in	O
schools	O
like	O
Carnegie	O
Mellon	O
University	O
,	O
Columbia	O
University	O
,	O
the	O
Georgia	O
Institute	O
of	O
Technology	O
,	O
and	O
the	O
University	O
of	O
Texas	O
.	O
</s>
<s>
Before	O
audio	B-General_Concept
mining	I-General_Concept
became	O
the	O
mainstream	O
method	O
,	O
written	O
transcripts	O
of	O
audio	O
content	O
were	O
created	O
and	O
manually	O
analyzed	O
.	O
</s>
<s>
Audio	B-General_Concept
mining	I-General_Concept
is	O
typically	O
split	O
into	O
four	O
components	O
:	O
audio	B-General_Concept
indexing	I-General_Concept
,	O
speech	O
processing	O
and	O
recognition	O
systems	O
,	O
feature	O
extraction	O
and	O
audio	O
classification	O
.	O
</s>
<s>
The	O
audio	O
will	O
typically	O
be	O
processed	O
by	O
a	O
speech	B-Application
recognition	I-Application
system	O
in	O
order	O
to	O
identify	O
word	O
or	O
phoneme	B-Language
units	O
that	O
are	O
likely	O
to	O
occur	O
in	O
the	O
spoken	O
content	O
.	O
</s>
<s>
This	O
information	O
may	O
either	O
be	O
used	O
immediately	O
in	O
pre-defined	O
searches	O
for	O
keywords	O
or	O
phrases	O
(	O
a	O
real-time	O
"	O
word	O
spotting	O
"	O
system	O
)	O
,	O
or	O
the	O
output	O
of	O
the	O
speech	B-Application
recognizer	I-Application
may	O
be	O
stored	O
in	O
an	O
index	O
file	O
.	O
</s>
<s>
One	O
or	O
more	O
audio	B-General_Concept
mining	I-General_Concept
index	O
files	O
can	O
then	O
be	O
loaded	O
at	O
a	O
later	O
date	O
in	O
order	O
to	O
run	O
searches	O
for	O
keywords	O
or	O
phrases	O
.	O
</s>
<s>
In	O
audio	O
,	O
there	O
is	O
the	O
main	O
problem	O
of	O
information	B-Library
retrieval	I-Library
-	O
there	O
is	O
a	O
need	O
to	O
locate	O
the	O
text	O
documents	O
that	O
contain	O
the	O
search	O
key	O
.	O
</s>
<s>
Hence	O
,	O
audio	B-General_Concept
indexing	I-General_Concept
allows	O
efficient	O
search	O
for	O
information	O
by	O
analyzing	O
an	O
entire	O
file	O
using	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
It	O
is	O
done	O
through	O
mainly	O
two	O
methods	O
:	O
Large	O
Vocabulary	O
Continuous	O
Speech	B-Application
Recognition	I-Application
(	O
LVCSR	O
)	O
and	O
Phonetic-based	O
Indexing	O
.	O
</s>
<s>
In	O
text-based	O
indexing	O
or	O
large	O
vocabulary	O
continuous	O
speech	B-Application
recognition	I-Application
(	O
LVCSR	O
)	O
,	O
the	O
audio	O
file	O
is	O
first	O
broken	O
down	O
into	O
recognizable	O
phonemes	B-Language
.	O
</s>
<s>
It	O
is	O
then	O
run	O
through	O
a	O
dictionary	B-Operating_System
that	O
can	O
contain	O
several	O
hundred	O
thousand	O
entries	O
and	O
matched	O
with	O
words	O
and	O
phrases	O
to	O
produce	O
a	O
full	O
text	O
transcript	O
.	O
</s>
<s>
If	O
the	O
text	O
or	O
word	O
could	O
not	O
be	O
found	O
in	O
the	O
dictionary	B-Operating_System
,	O
the	O
system	O
will	O
choose	O
the	O
next	O
most	O
similar	O
entry	O
it	O
can	O
find	O
.	O
</s>
<s>
On	O
the	O
other	O
hand	O
,	O
LVCSR	O
is	O
susceptible	O
to	O
common	O
issues	O
of	O
speech	B-Application
recognition	I-Application
.	O
</s>
<s>
Another	O
problem	O
with	O
LVCSR	O
is	O
its	O
over	O
reliance	O
on	O
its	O
dictionary	B-Operating_System
database	O
.	O
</s>
<s>
LVCSR	O
only	O
recognizes	O
words	O
that	O
are	O
found	O
in	O
their	O
dictionary	B-Operating_System
databases	O
,	O
and	O
these	O
dictionaries	B-Operating_System
and	O
databases	O
are	O
unable	O
to	O
keep	O
up	O
with	O
the	O
constant	O
evolving	O
of	O
new	O
terminology	B-Language
,	O
names	O
and	O
words	O
.	O
</s>
<s>
Should	O
the	O
dictionary	B-Operating_System
not	O
contain	O
a	O
word	O
,	O
there	O
is	O
no	O
way	O
for	O
the	O
system	O
to	O
identify	O
or	O
predict	O
it	O
.	O
</s>
<s>
Audio	B-General_Concept
mining	I-General_Concept
systems	O
try	O
to	O
cope	O
with	O
OOV	O
by	O
continuously	O
updating	O
the	O
dictionary	B-Operating_System
and	O
language	B-Language
model	I-Language
used	O
,	O
but	O
the	O
problem	O
still	O
remains	O
significant	O
and	O
has	O
probed	O
a	O
search	O
for	O
alternatives	O
.	O
</s>
<s>
This	O
makes	O
LVCSR	O
an	O
expensive	O
approach	O
to	O
audio	B-General_Concept
mining	I-General_Concept
.	O
</s>
<s>
Phonetic-based	O
indexing	O
also	O
breaks	O
the	O
audio	O
file	O
into	O
recognizable	O
phonemes	B-Language
,	O
but	O
instead	O
of	O
converting	O
them	O
to	O
a	O
text	O
index	O
,	O
they	O
are	O
kept	O
as	O
they	O
are	O
and	O
analyzed	O
to	O
create	O
a	O
phonetic-based	O
index	O
.	O
</s>
<s>
It	O
begins	O
by	O
converting	O
the	O
input	O
media	O
into	O
a	O
standard	O
audio	O
representation	O
format	O
(	O
PCM	B-Algorithm
)	O
.	O
</s>
<s>
The	O
user	O
's	O
search	O
query	B-Library
term	O
is	O
parsed	O
into	O
a	O
possible	O
phoneme	B-Language
string	O
using	O
a	O
phonetic	O
dictionary	B-Operating_System
.	O
</s>
<s>
Then	O
,	O
multiple	O
PAT	O
files	O
can	O
be	O
scanned	O
at	O
high	O
speed	O
during	O
a	O
single	O
search	O
for	O
likely	O
phonetic	O
sequences	O
that	O
closely	O
match	O
corresponding	O
strings	O
of	O
phonemes	B-Language
in	O
the	O
query	B-Library
term	O
.	O
</s>
<s>
Phonetic	B-General_Concept
indexing	I-General_Concept
is	O
most	O
attractive	O
as	O
it	O
is	O
largely	O
unaffected	O
by	O
linguistic	O
issues	O
such	O
as	O
unrecognized	O
words	O
and	O
spelling	O
errors	O
.	O
</s>
<s>
That	O
makes	O
it	O
particularly	O
useful	O
for	O
searching	O
specialized	O
terminology	B-Language
or	O
words	O
in	O
foreign	O
languages	O
that	O
do	O
not	O
commonly	O
appear	O
in	O
dictionaries	B-Operating_System
.	O
</s>
<s>
Furthermore	O
,	O
in	O
contrast	O
to	O
LVCSR	O
,	O
it	O
can	O
process	O
audio	O
files	O
very	O
quickly	O
as	O
there	O
are	O
very	O
few	O
unique	O
phonemes	B-Language
between	O
languages	O
.	O
</s>
<s>
However	O
,	O
phonemes	B-Language
cannot	O
be	O
effectively	O
indexed	O
like	O
an	O
entire	O
word	O
,	O
thus	O
searching	O
on	O
a	O
phonetic-based	O
system	O
is	O
slow	O
.	O
</s>
<s>
An	O
issue	O
with	O
phonetic	B-General_Concept
indexing	I-General_Concept
is	O
its	O
low	O
accuracy	O
.	O
</s>
<s>
Phoneme-based	O
searches	O
result	O
in	O
more	O
false	O
matches	O
than	O
text-based	O
indexing	O
.	O
</s>
<s>
Deemed	O
as	O
the	O
most	O
critical	O
and	O
complex	O
component	O
of	O
audio	B-General_Concept
mining	I-General_Concept
,	O
speech	B-Application
recognition	I-Application
requires	O
the	O
knowledge	O
of	O
human	O
speech	O
production	O
system	O
and	O
its	O
modeling	O
.	O
</s>
<s>
The	O
electrical	O
speech	O
production	O
system	O
converts	O
acoustic	O
signal	O
into	O
corresponding	O
representation	O
of	O
the	O
spoken	O
through	O
the	O
acoustic	O
models	O
in	O
their	O
software	O
where	O
all	O
phonemes	B-Language
are	O
represented	O
.	O
</s>
<s>
A	O
statistical	B-Language
language	I-Language
model	I-Language
aids	O
in	O
the	O
process	O
by	O
identifying	O
how	O
likely	O
words	O
are	O
to	O
follow	O
each	O
other	O
in	O
certain	O
languages	O
.	O
</s>
<s>
Put	O
together	O
with	O
a	O
complex	O
probability	O
analysis	O
,	O
the	O
speech	B-Application
recognition	I-Application
system	O
is	O
capable	O
of	O
taking	O
an	O
unknown	O
speech	O
signal	O
and	O
transcribing	O
it	O
into	O
words	O
based	O
on	O
the	O
program	O
's	O
dictionary	B-Operating_System
.	O
</s>
<s>
ASR	O
(	O
automatic	B-Application
speech	I-Application
recognition	I-Application
)	O
system	O
includes	O
:	O
</s>
<s>
Acoustic	O
model	O
:	O
establishes	O
relationship	O
between	O
speech	O
signal	O
and	O
phonemes	B-Language
,	O
pronunciation	O
model	O
and	O
language	B-Language
model	I-Language
.	O
</s>
<s>
Training	O
algorithms	O
are	O
applied	O
to	O
the	O
speech	O
database	O
to	O
create	O
statistical	O
representation	O
of	O
each	O
phoneme	B-Language
,	O
thus	O
generating	O
an	O
acoustic	O
model	O
with	O
a	O
set	O
of	O
phonemes	B-Language
and	O
their	O
probability	O
measures	O
.	O
</s>
<s>
Some	O
applications	O
of	O
speech	O
processing	O
includes	O
speech	B-Application
recognition	I-Application
,	O
speech	O
coding	O
,	O
speaker	O
authentication	O
,	O
speech	O
enhancement	O
and	O
speech	O
synthesis	O
.	O
</s>
<s>
Prerequisite	O
to	O
the	O
entire	O
speech	B-Application
recognition	I-Application
process	O
,	O
feature	O
extraction	O
must	O
be	O
established	O
first	O
within	O
the	O
system	O
.	O
</s>
<s>
Non-stationary	O
signals	O
can	O
be	O
analyzed	O
using	O
Fourier	B-Algorithm
and	O
short-time	B-Algorithm
Fourier	I-Algorithm
,	O
while	O
time-varying	O
signals	O
are	O
analyzed	O
using	O
Wavelet	B-Algorithm
and	O
Discrete	O
wavelet	B-Algorithm
transform	I-Algorithm
(	O
DWT	O
)	O
.	O
</s>
<s>
Audio	O
classification	O
is	O
a	O
form	O
of	O
supervised	B-General_Concept
learning	I-General_Concept
,	O
and	O
involves	O
the	O
analysis	O
of	O
audio	O
recordings	O
.	O
</s>
<s>
There	O
are	O
several	O
methods	O
to	O
audio	O
classification	O
using	O
existing	O
classifiers	O
,	O
such	O
as	O
the	O
k-Nearest	B-General_Concept
Neighbors	I-General_Concept
,	O
or	O
the	O
naïve	B-General_Concept
Bayes	I-General_Concept
classifier	I-General_Concept
.	O
</s>
<s>
There	O
has	O
also	O
been	O
research	O
into	O
using	O
deep	O
neural	O
networks	O
for	O
speech	B-Application
recognition	I-Application
and	O
audio	O
classification	O
,	O
due	O
to	O
their	O
effectiveness	O
in	O
other	O
fields	O
such	O
as	O
image	O
classification	O
.	O
</s>
<s>
Audio	B-General_Concept
mining	I-General_Concept
is	O
used	O
in	O
areas	O
such	O
as	O
musical	O
audio	B-General_Concept
mining	I-General_Concept
(	O
also	O
known	O
as	O
music	B-Application
information	I-Application
retrieval	I-Application
)	O
,	O
which	O
relates	O
to	O
the	O
identification	O
of	O
perceptually	O
important	O
characteristics	O
of	O
a	O
piece	O
of	O
music	O
such	O
as	O
melodic	O
,	O
harmonic	O
or	O
rhythmic	O
structure	O
.	O
</s>
<s>
Within	O
the	O
field	O
of	O
linguistics	O
,	O
audio	B-General_Concept
mining	I-General_Concept
has	O
been	O
used	O
for	O
phonetic	O
processing	O
and	O
semantic	O
analysis	O
.	O
</s>
<s>
The	O
efficiency	O
of	O
audio	B-General_Concept
mining	I-General_Concept
in	O
processing	O
audio-visual	O
data	O
lends	O
aid	O
in	O
speaker	O
identification	O
and	O
segmentation	O
,	O
as	O
well	O
as	O
text	O
transcription	O
.	O
</s>
<s>
In	O
particular	O
,	O
this	O
has	O
been	O
used	O
for	O
speech	B-General_Concept
analytics	I-General_Concept
.	O
</s>
<s>
Further	O
use	O
has	O
been	O
seen	O
in	O
areas	O
of	O
speech	B-Application
recognition	I-Application
and	O
text-to-speech	O
applications	O
.	O
</s>
