AW(2): adsr and lfo abolish

[Martin Czech]

> Excuse my ignorance, but I still don't understand what the problem
> of ADSR and similar envelope generators is. 
> 

There is no a priori problem.  Depends what you want to do. I also like
the sound of "vintage" machines (especially guitar gear). If you want
to sound like a solo line on a mini, well of course using an adsr is
perfectly ok.

Ok, why all this discussion?  Sometimes I go to the local shop, the
guy has allmost every intersting synth there. After a while you start
thinking: mhhh, these things look quite different, but tend to sound
the same, regardless of synthesis algorithm.  Why is this so?

The human ear is in some way optimised for speech recognition.  If you
take a look at the sensitivity curves from psychoacoustic books, where
is the maximum? Where the main energy of speech is!  The "invention" of
speech is such an enormous evolutionary advantage over other primates,
it simply must be that way.  Now, you all know that speech recognition
is a very difficult task for a machine to do.

If you have catched a bad cold, if we're walking by at a construction site
with a hell lot of noise, if we're talking using a telephone or radio,
if you are using a vocoder or a pitch shifter in order to shock me,
still we are able to understand speech, in spite of the fact that there
is noise and that the spectrum of normal speech is heavily distorted.

How is this possible? The ear/brain does not recognize spectra (not very
good), but it is very good in recognizing the CHANGE of spectra. Speech
is memorized as change, or way of change, otherwise such crude
aproximations vocoders use to do would never work.

Therefore experiments with steady state harmonic waves tend to get boring,
regardless how many harmonics are added.

If this is true then the first goal of synthesis must be to allow for
unlimited freedom in the change of sounds, ie. freedom in the change of
synthesis parameters. The synthesis of complicated steady state spectra
is only a secondary goal.

As things are now, most machines uses a kind of ADSR and LFO for parameter
controll. Ie. you can't controll the synthesis parameters directly.
And since this is the case there is a tendency that most machines sound
quite similar.

Therefore it would be nice if we could controll the key synthesis
parameters in an arbitrary way, e.g. by drawing.  Simple example: Two
operator "FM". You have 4 parameters: carrier and modulator frequency
and amplitude (drop phase information here). Now you take a piece of
paper and draw the evolution of these four simple parameters over time.
I think you can imgaine a lot of interesting curves, eg. the frequencys
come near to integer ratios, then drift away, the ratio changes from
1:1 to 1:7, the carrier level fades in softly  then has a very fast
rise and so on. And all this as a continous process. I think that very
interesting sounds could be achieved, sounds that walk the thin line
between harmonic and "strange" for example. The speed of modulation
change should range from DC to some kHz (20kHz?), and the resolution
should be from 8 to 16 bit, we have to consider the resolution of the
human ear.

A  transient recorder with 16-bit resolution and about 40kHz sampling rate
could do the job. One could think of reducing the data rate by providing
only rate and amplitude information using a interpolation scheme much in
the way that multi-segment-eg do it. I think that digital processing is
necessary in order to keep error and drift down.  A first order recursive
filter should do it, sicne we are interested in the impulse response
and not so much in the frequency response of such an interpolator.

This way one would get more interesting sounds but the price we'd have to
pay is to drop realtime control, at least in the more complicated cases.


If you try now to do such a complicated modulation with some arrangement
fo adsr or even multi-segment-eg, well - it is possible, but lots of
pain. That's why nobody does it.


m.c.