[Ardour-Users] Analog synths (was: Ardour-Users Digest, Vol 167, Issue 13)

Gordonjcp gordonjcp at gjcp.net
Fri Dec 22 16:45:21 PST 2017


On Fri, Dec 22, 2017 at 10:53:27PM +0100, David Kastrup wrote:
> Adriano Petrosillo <ampetrosillo at gmail.com> writes:
> 
> >> We don't need analogue synths any more.  They were a shitty solution
> >> to a problem we no longer have.  It's better to do it all digitally.
> >
> > Oscillators may be relatively easy to model.
> 
> Sine is easy to do "correct" in the digital domain.  But everything else
> is much much more complicated than one would think because of the
> interaction with the antialiasing and interpolating filters and the
> sampling.
> 

It's not that hard to do.  Polybleps are probably the simplest way to do
antialiasing, and provide perfectly acceptable performance - in
isolation you may hear a little aliasing in pathological cases but in a
mix it disappears.

> > It's quite hard to recreate the character certain filters have, on the
> > other hand. When you push filters into non-linear operation, they do
> > not behave as in the textbooks.  Can you do this digitally? Of course
> > you can, provided you put in the effort, you have the CPU power to
> > model their behaviour well enough, etc.
> 
> Again, nonlinearities have large interactions with sampling frequencies
> and their interpolating filters.
> 
> Even if you want to do this "exactly", it is a whole lot of effort.  It
> may be easier to do this in a seriously oversampled domain and apply
> low-order lowpass filtering on the signals.

Exactly, but why bother?  The filters don't sound *that* different.  You
can accurately model different types of filters but no-one will ever
hear a difference for the most part.  One trick that *is* worth doing is
introducing a bit of the audio-rate distortion that Sallen-Key filters
like the Korg MS20 filters suffer from, which gives that nice "twin
peaks" resonance as it pulls itself out of tune.

Back in the day, there was this craze for building TB303 clones where
people obsessed about getting exactly the right kind of shitty 1980s
floor-sweepings transistors for the filter, and all they'd go on about
was how exact the filter was (bonus prize for saying it's 18dB/octave
response, which is so astonishingly wrong you could probably start a
religion around it).  Every single one of these clones "improved" the
VCO (losing the character of the badly-done squarewave) and "improved"
the envelope (totally losing the point of the entire project).  But hey,
they sure did spend a lot of time on that filter.

Everyone seems to miss this important point.  The oscillators and
filters don't really contribute much to the overall sound, they just
make a beeping noise.  It's how you modulate them that governs how
synthesizers sound.  That's why it's difficult to accurately model an
existing synthesizer - a sawtooth is a sawtooth is a sawtooth, but the
way the envelopes respond is absolutely crucial.


> Oh, and seriously?  Never ever ever ever reply to a digest, particularly
> not quoting the _whole_ _darn_ _digest_ at the bottom.  Your quotation
> of the entire digest will appear as part of the next digest.  And
> imagine someone then doing the same?

This list is mercifully free of such nonsense.

-- 
Gordonjcp




More information about the Ardour-Users mailing list