Measuring the
Sound Color
Sound Color = Spectrum
Analysis
The basics for a good sound color detection
lies in spectrum analysis. The spectrum of a song describes how much a
certain frequency is present within the song. However, weighing different
frequencies and making something useful from a spectrum analysis is not
as straight forward as expected.
The Bark Frequency Scale
The first thing to note is that the human
ear is very well suited to detect certain frequencies. Especially in the
range of the human voice we are very capable of hearing well. However,
on the higher frequency ranges (above 11500 Hz), humans are unable to distinguish
different frequencies. Therefore, to correctly describe the sound color
of a song we need to take into account how well the human ear perceives
these frequencies. Luckily, such a scale exists and is called the Bark
Scale.
0-100
100-200
200-300
300-400
400-510
510-630
630-770
770-920
920-1080
1080-1270
1270-1480
1480-1720
1720-2000
2000-2380
2380-2700
2700-3150
3150-3700
3700-4400
4400-5300
5300-6400
6400-7700
7700-9500
9500-12000
12000-15500
Measuring The Spectrum
The spectrum of a song is measured over 10 seconds at the last
used cue-position. The strength of the different frequencies is measured
and exported as 24 bands of the bark-scale.
Similarity Detection
To measure how close two song-)colors are we take the L^2 norm,
which is the sum of the squares of the differences of all different bands.
Determining the Sound
Color
The sound
color itself is created by a principal component analysis. This is
described in more detail in [WVB--REF].
Cluster Analysis
[WVB --TODO]
Copyright (c) Werner
Van Belle December 2001
e-mail: werner.van.belle@vub.ac.be
Tel: +32 486 68 84
48; Fax: +32 2 629 35 25