Realistic Simulations of Acoustic Instruments

In an off-line conversation with one of the tuning-listers, I sent this,
and thought that some of you folks might want to read it too.

A "believable" trumpet, clarinet, or violin is clearly relative. I just
got done spending $2500 to upgrade my alto saxophone. Without a doubt
though, the distinction between the older and upgraded instruments is
beyond the capabilities of even a fairly sophisticated sampling machine to
faithfully reproduce. $2500 for something pretty far down in the realm of
sublties. Musicians are a really picky lot.

As I'm fond of pointing out, when you go from analog synthesis to
additive or sampling, you also go from what our ears perceive as impressive
imitations to really awful-sounding real instruments. So even though in
every clearly-perceptible and measurable way, there is no doubt that the
simulation is more accurate, we've unwittingly hiked the standard by which
it's being judged by an order of magnitude.

Believability is also based upon the listeners' backgrounds: Many
people have vastly more discriminating hearing than the depth of their
experience with certain instruments. Often peoples' beliefs as to what a
certain instrument sounds like isn't really what they actually sound like.
As an example, the recorder simulations in my contribution to the
tuning-list CD, which at least one person thought didn't think sounded very
good, are really pretty darned accurate in side-by-side comparisons to the
instruments they were sampled from. They accurately simulate different
attack hardnesses, slurred vs. tongued attacks, what aspects of the tone
vibrato modulates, noise components in the sound, different timbral
registers, volume/timbre correlations, and of course the differences
between the soprano, tenor (and the others). There were two problems
though:

1. They weren't exactly perfect, and as such they became an insult to a
recorder-fan.
2. My performance is was stiff and unanimated, mostly because I didn't
have my WX-11 working at the time, so I performed them from the
keyboard, which I quite simply can't play.

Along those lines, I've also concluded that, if you're going to simulate
a "real"-instrument ensemble, you're best off sticking with ensembles
people are familiar with, and not tell them that they're a simulation. In
one of my compositions, some people told me that the strings sounded
fakey. In all honesty, they were right; partly because of bad source
sample material. But I was kinda freaked out to find that, for at least
one of these people, it turned out to be because he was trying to liken
them to the sound of an entire string section rather than individual solo
violin, viola, and 'cello! He didn't have any real listening experience
with individual orchestral stringed instruments.

On my ASR-10, I've produced a number of instrument simulations, and
those simulations have improved (most of the improved ones I haven't
recorded anything with yet). But one thing's for sure: if it's
sufficiently accurate, people will judge you against a much higher standard
than if it's a "cartoon", to use W. Carlos' phraseology.