[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Dear Dr. Husain,
I often employ a very primitive technique to generate noise, which might
be criticized in the field of acoustic engineering. For people in the
field of auditory psychology, however, the technique seems practical
(People who are not studying practical things are sometimes practical.).
1. Divide the frequency band into N cells (= subbands). The cells are
numbered from 1 to N according to their center frequencies. The cells
must be small enough compared with critical bands (or auditory filters).
2. Choose a representative component, whose frequency and initial phase
are determined randomly, from the cells 1 and 2. Choose a
representative in the same way from the cells 2 and 3 (Notice that the
cell 2 has two chances to offer a representative component.). --- Choose
a representative from the cells N-1 and N. Choose a representative
from the cells N and 1. Thus each cell had two chances to offer a
representative, but each time there was a probabilistic competition
between two neighboring or farthest cells. All the representative
components have the same level.
I use HP-BASIC to generate sound, and the above procedure is written
as a simple FOR...NEXT loop with two random numbers. The probabilistic
competition between cells seems necessary to avoid a perceptual
impression of beats, which could be an interesting topic itself. This
technique is convenient to control spectral patterns intuitively without
difficult mathematics. The perceptual quality of the generated noises
was really noise-like. One of my students dared to employ this technique
even to synthesize simple Japanese speech. Of course I admit that
we sometimes need more rigorous techniques.