Digital Signal Processing Reference
In-Depth Information
Chapter 12
Applications in Intelligent Sound Analysis
If you develop an ear for sounds that are musical it is like
developing an ego. You begin to refuse sounds that are not
musical and that way cut yourself off from a good deal of
experience.
—
John Cage
Apart from the more specific types of sound considered so far—speech and music—
general sound can also carry relevant information. This is, however, a considerably
less researched field up to-date. Most prominent in this area are the tasks of acoustic
event detection (AED) and classification (AEC) [
1
] that can be subsumed under the
area of computational auditory scene analysis (CASA) [
2
]. For these tasks interna-
tional evaluation campaigns exist that have mostly seen HMM and SVM approaches
with various acoustic features [
1
]. Fields of application include media retrieval [
3
]
including affective content analysis [
4
] or human-machine and human-robot interac-
tion [
5
], animal vocalisation recognition [
6
], and monitoring of industrial processes
[
7
]. Mostly, closed-set recognition is addressed, i.e., training and testing classes are
the same. Recently, however, also open-set recognition is faced, the so-called novelty
detection [
8
,
9
].
As before, examples of application have been chosen for illustration of obtainable
performances and methods employed. Three applications have been chosen to cover a
good variety of the above named use cases: Firstly, recognition of animal vocalisation
[
10
], then, acoustic event classification including unsupervised learning to exploit
the availability of sheer infinite amounts of sound on the Internet [
11
], and finally
prediction of the emotion evoked in human listeners of sound [
12
] in analogy to the
sections on speech and music.
