[sdiy] String Filter - Answers (different than NM post)

[Kenneth Elhardt]

My internet computer is now out of my studio and up connected to the net on
a more permanent basis, so I'll be online most of the time. So there should
be these long delays in my catching up on 9000+ posts.

yves usson writes:

>>Very impressive Ken!<<



Peter Forrest writes:

>>Fantastically impressive, Ken!<<

Thanks for the compliments. Since I'm writing this from the future,
hopefully you've also listened to my second batch under the "String filter
adjusted to real instruments" thread. They are more accurate than the first
batch. I still need to make a couple of more passes though to get even
closer to the real thing.



Gene Stopp writes:

>>So many times I find out I do not "get" something even though I have read
about it for years. In this case, finally I "get" the importance of having
the formant and resonances correct for the wood body of a string instrument
for imitative synthesis. I can hear it. <<

Yeah, typical synth filtering doesn't allow this kind of stuff to be done.
That's why the need for dense bandpass filter bands with sharps resonances.
There's nothing better than hearing it rather than just reading about it.



Dino Leone writes:

>>You know, you're making it a lot worse, my obsession and desire for such a
stringfilter!! ;-) Could we kindly ask you to go a bit more into details
regarding the peaks and notches per octave? What are the exact numbers?
Would that mean you make a peak at, say, C4, then a notch at D#4,
another peak at F#4 and so on? Are these the same numbers you used to
generate the filter plot? How steep would the filterbands need to be? How
big a Q? What's the "height" (in dB) of the peaks and notches?<<



The numbers for these demos are pretty much those in the chart. I was using
anyware from 36 to 42 bands depending how low or high a range I needed to
cover. But that's kind of a generic wood filter and not yet modified to
match actual specific instruments. Normally the string filter is just
something to give you a generic wood-like sound for improved string
emulations. But I don't want to stop there. A real violin is much more
random than some evenly spaced filters. So what's up there is obsolete
already after my second batch of demos. And I now have different filter
setups for violin, viola, and cello, so it's a whole bunch of different
numbers now. But just so you understand what's going on up there and by my
comments, when I say I'm alternating filter spacing between 2 and 3
semitones, that means for example, one at G3, then A#, then C4, then D#,
etc. Those are where the peaks are at. The notches just happen between those
positions. You can also look at my chart to see the dB heights of those
peaks. It varies depending upon distance between bands, but I'm averaging
about 15dB. That's what Bob Moog stated in his string filter writeup. The
steepness or Q needs to be enough to give you whatever height of peaks or
notches you want. In my digital world I have numbers between 0 and 127. That
really doesn't mean anything outside the Nord modular.


Ian Fritz writes:

>>Thanks for putting all that up. I'm looking forward to studying it in
detail. I made a 36-band filter bank many years ago using the Burhans
(2.1)^(1/5) spacing. It was in Electronotes, if you ever looked at any of
that old stuff.<<

I haven't seen that one. I have Bob Moog's article and have heard about the
Matthews filter bank. But none I've seen give absolute numbers. Moog just
talked in general terms. The thing is I've spent days trying all kinds of
configurations, and they can sound radically different. Even just moving the
entire filterbank up or down one semitone can go from something pretty good
to something that sounds awful. I don't know why it is so touchy. Also, I
know the kind of low, grinding, cello type wood sound I want, and that seems
to require diverging from what others say.



Barry Klein writes:

>>Has this sort of thing been played with on the Capybara 320? I have one
but do not know how to use it yet. Have you used one? Judging by your studio
photos you have lots of toys!<<



No I haven't used one, but I believe the Capybara does just about every kind
of function known to man. I bet it would be great for massive filterbanks.
I'm thinking of moving mine over to Reaktor on the Mac so I can continue to
improve the driver portion of my patch. I've run out of Nord DSP power.

-----------------------

Posts that came after my second batch of demos:

Matt Simpson writes:

>>These are absolutely incredible.<<

Thanks for the compliments.



Tim Parkhurst writes:

>>Absolutely FANTASTIC! I'm thinking that the realism of your work is due
not only to the string filter, but also to your playing technique. Do you
have more info or hints on technique and any special approaches you use? Of
course, I'd like to see more info on the filter and the hardware you use
also.<<



Well hopefully more of it is the sound of the patch because that's what I'm
comparing to real samples, but it is programmed with velocity and pedal
control of dynamics with some other details to make it expressive to play.
The filter and the hardware is the Nord Modular and its software based
modules. I can cut from a post I made to the Nord list which describes some
of the details others have asked:



"It takes days to try different filter configurations to even get something
that sounds somewhat like what I want. Then it's another long miserable
period of time analysing real string samples, taking FFT's (sometimes
animated ones of instruments sliding up a string), trying to determine all
the formant peaks and dips in the harmonic spectrum, and trying to set up
filters that mimic them, and sweeping sine waves through my filter bank to
see what it looks like and why so many unpredictable things seem to happen
when moving the filters around which I have to fight with.



(regaring bowed attacks) My slot 1 string generator patch uses two
oscillators. You're always hearing a hardsync'd oscillator, but since most
of the time it's in perfect sync with the master oscillator (that you don't
hear) it just generates a sawtooth wave. But on the attacks, I hit that
sync'd oscillator with a burst of filtered noise which causes it to bounce
around, yet at the same to try to sync to the master oscillator. This is
kind of what a string does as the bow starts to pull it, but it randomly
slips prematurely before its cycle is fully complete generating smaller
shorter waveforms, but still overall kind of syncing to the overall pitch of
the master osc. The freq of the attack noise, the length, and the depth can
be adjusted to taste. I believe the depth of it is under velocity sensitive
control.



As for additional bow noise, to get a coarser sound overall, I'm using a
wavewrapper, and modulating the CV input of that with some noise. This also
adds brightness to the sound like an exciter and gives a more complex and
moving higher freq content to hide the potential synthetic sounding high
freqs. That and the sweeping delay lines animate the higher harmonics to act
like a real instrument up there. If both of these were taken out, the upper
end would sound too sawtooth/synthetic, or if my filters are too resonant up
there, it sounds too digital, like a digital wavetable oscillator. But the
wavewrapper's sound is very dependent on what's fed into it, and that takes
fooling around with so as not to sound brittle and digital. In my patch, it
comes after two LPF's (combined for 48dB cutoff). All this is very
sensitive, as changing one thing can affect others. For instance, if my
sweeping delay lines are too fast, it diffuses my vibrato, and I can't get a
strong vibrato. The driver patch is kind of balancing on a razor's edge."



Hopefully that will give you some info.



-Elhardt