Page 214 - Foundations of Cognitive Psychology : Core Readings

The Auditory Scene 219

Figure 9.4
A spectrogram of a mixture of sounds (containing the word ‘‘shoe’’).

what like a picture created by making a spectrogram of each of the individual
sounds on a separate piece of transparent plastic, and then overlaying the in-
dividual spectrograms to create a composite. The spectrogram of the word shoe
is actually oneofthe componentspectrogramsofthe mixture.
Although the theorist has the privilege of building the composite up from the
pictures of its components, the auditory system, or any machine trying to imitate
it, would be presented only with the spectrogram of the mixture and would
have to try to infer the set of pictures that was overlaid to produce it.
The recognizer would have to solve the following problems: How many
sources have created the mixture? Is a particular discontinuity in the picture
a change in one sound or an interruption by a second one? Should two dark
regions, one above the other in the picture (in other words, occurring at the
same time), be grouped as a single sound with a complex timbre or separated
to represent two simultaneous sounds with simpler timbres? We can see that if
we look at a spectrogram representing a slice of real life, we would see a com-
plex pattern of streaks, any pair of which could have been caused by the same
acoustic event or by different ones. A single streak could have been the sum-
mation of one, two, or even more parts of different sounds. Furthermore, the
frequency components from one source could be interlaced with those of an-
other one; just because one horizontal streak happens to be immediately above
another, it does not mean that they both arose from the same sonic event.
We can see that just as in the visual problem of recognizing a picture of
blocks, there is a serious need for regions to be grouped appropriately. Again,
it wouldbeconvenienttobeable tohandthe spectrogramovertoamachine
that did the equivalent of taking a set of crayons and coloring in, with the same
color, all the regions on the spectrogram that came from the same source. This
‘‘coloring problem’’ or ‘‘auditory scene analysis problem’’ is what the rest of
this chapter is about.