ELECTRIC OPRPHEUS ACADEMY
SPILLING THE BEANS #3 SQUEEZE
Each operation in a limited buffer, regardless of
how large this is, allows itself, as we know, to also be implemented granularly:
A sound will be disassembled into small, overlapping portions, these will
be processed and then reassembled. A confusing variety of
techniques, tricks and secrets (not to say 'myths') is centered on this
simple method.
A lot of this can be tried out step-by-step
in VASP. Firstly, the disassembly and resynthesis are regulated by so-called
'structures'. These can be generated or also analytically derived from
the sound content itself - and different structures can also be applied
for analysis and resynthesis. This alone already opens a plethora of experimental
possibilities.
Beyond this, the grains can also be defined in each domain – in
the time domain or in the frequency domain. Then the grains themselves
can likewise be processed in each domain again. (For effects, as an example,
which are crudely denoted as 'pitch shift' or 'time stretch', numerous
methodic variants can be found).
In addition, there are operations that admittedly treat many adjacent
samples as a unit, but nonetheless do not work granularly, because the
window segment is pushed further on sample by sample. (A luxury if the
segments are large).
One of these is the automodulation squeeze. As in the
other classic automodulations, it creates combination frequencies of all
the frequencies contained in the original, however, with a strong stress
on the doubled frequencies. Applied to language, a squeezed, falsetto-like
octavation arises (hence the term). The level drop outs are not as glaring
as in other automodulations, and can still be compensated by an additional
standardization.
demo04_squeeze.mp3
This audio example shall be understood as the processing
of an excerpt from published recordings. With the possibility to listen
it and study it here for demonstration purposes, there are no further
exploitation rights associated with it.
Once again, Oskar Werner, who we will
encounter more often in various transformations. The corresponding VASP
script looks like this:
sfload gelassen.wav
hilb
squeeze.norm
opt
play
If the application of squeeze in the time domain results in an octavation upwards - a 'pitch shift' - then the application in the frequency domain (in the total spectrum) results in a type of 'time stretch'. And this is also the case:
demo04_squeeze_FFT.mp3
This
audio example shall be understood as the processing of an excerpt from
published recordings. With the possibility to listen it and study it here
for demonstration purposes, there are no further exploitation rights associated
with it.
The corresponding VASP script looks like this:
sfload gelassen.wav
squeeze.norm
FFT-
opt
play
Where, however, is the complex processing here?? – Very simple,
while half of the buffer is empty. That is also evident from another view:
If the resulting sound becomes twice as long, it is clear that there must
be space in the buffer for it. [As I said, I would like
to begin the workshop with an introduction to complex audio].
The method is also available as an intermodulation: vsqueeze.
(In VASP, all operations that merge the two buffer contents in parallel
with each other begin with a 'v' for 'vector').
The effects with it, however, are less spectacular. Most of the time,
one component dominates and the others rather seem to be irritations.
A related method, which one can get more nuances out of, is filtering
the sound with segments from another one: vfir. fir stands
for 'finite impulse response', a standard among filters. Any arbitrary
segment of a sound can be selected as a filter function for another sound
with it. If the segment is very large, then the recommended method is
a convolution, because filtering, resonance and reverberation are principally
from the same root methodically, but differ solely in the character of
the components.
With vfir.run, however, the selected segment can be continuously
moved further.
In the following example white noise is filtered by a relatively large
segment (2047 samples) of the poem, whereby the segment (beginning at
0) is moved along at a speed of 0.01 (1/100 of the original speed).
demo05_vfir.mp3
This audio example
shall be understood as the processing of an excerpt from published recordings.
With the possibility to listen it and study it here for demonstration
purposes, there are no further exploitation rights associated with it.
One already nicely hears the transition of the vowels
from E to A and the hiss of the S in "gelassen..."
The script looks like this:
A:
sfload gelassen.wav
B:
colnoise.white
vfir.run2047 0,0.1 (win=bell)
opt
play
Such a 'running filter' can be solved in no other, quicker manner (convolution, etc.). The whole process lasts about one minute on an averagely fast computer. But we have time, don’t we? ...
akueto
G.R.
(c) Günther
Rabl 2010
