[A6] RE: Neuron (ot)

[UnderTow]

On Wed, 12 Mar 2003, Rhen, Kris wrote:

> >>>> Sampler - no.  The neuron is closer to something that
> >>>> does physical modeling, like the Z1/Prophecy
> >
> > The principle behind the neuron is NOT like physical modelling!!
> > While pm generates sounds after modeled instruments neuron
> > makes analysis of
> > a sample and uses a special model the keep as "an algo" and
> > afterwards can
> > be tweaked (paramters are def'd by that model)..
>
> It most certainly is (at the 'what you see' level).  What do you think a
> phys model is but an algorythm that can be tweaked? :-)  The difference is
> that the models are 'hard coded' for phys modeling synths like the z1, but
> the models are 'figured out' by the modelmaker software (neural nets) for
> the Neuron.
> KRIS

I don't think the Neuron uses models in the way physical modelling synths
do. With a physicall model, you model the effect of modifying certain
"real"  attributes of the physical object you are modelling. For instance
on a string model, you can vary the length of the string(s) or you can
vary the angle of the pick you use to hit the strings etc ...

The Neuron on the other hand has no sense of what is being modelled. It
takes a sound as input. (Or several sounds. I think). And it analyses
certain properties of the input data. (Probably after some FFT maths).
This way the Neuron builds a "model" of that/those input sound(s) and you
can tweak some of the parameters of that "model". (Model is between quotes
as it isn't really the correct word).

Two entirely different approaches which will have two entirely different
results.

You compared the Neuron to a Z1 with an infinite number of models. I don't
agree at all. Unless you can take a large number of samples which are
recordings of the original sound with variations on certain aspects of the
sound and map these variations to controls on the synth, you can't make
that comparison. What I mean in practise is that you would, for instance,
take a PW oscilator of the A6 (to stay on-topic ;) and tweak the Pulse
Width one unit at a time and feed each sample you take to the Neuron and
assigning that to one parameter on the Neuron. (Which you would call PWM
:) Then you could redo the whole thing but changing a different parameter
(Hmmm ... pitch maybe) and assign that in the Neuron Etc.

Not only do I think that isn't actually possible with the Neuron but it
would also take so much work that it would be completely pointless. (It
isn't for nothing that Korg arn't releasing new models for the Z1 every
month or so ...)

Now on a different but related note, has anyone ever played with the kind
of software used to create reverb algorythms in those expensive impulse
modelling reverbs from Roland/Lexicon/Yamaha (and I think TrueVerb) but
fed completely different sounds to it so as not to get reverb models but
... something else? :)

UnderTow