Modal synthesis is one of my favorite jams: an additive group of noise burst --> resonator chains.

The catch is the resonator. TL;DR Which object(s) should I use for filtering here?

More:

In SuperCollider, there's a lovely resonator called Ringz, which specifies bandwidth in terms of 60 dB decay time (and it's "constant skirt gain," meaning that it doesn't lose volume as the band gets narrower). This is especially useful because you can control the amount of the exciter's color that leaks through using Ringz.ar(exciter, freq, decayTime) - Ringz.ar(exciter, freq, attackTime) -- longer attack time = less "kch" from the noise burst.

I've gone into SC's source code and tried to backtrack through the math, but I can't replicate the sound using SOS (second-order section, basically [biquad~] in Pd). I think the problem is that Ringz internally uses double precision coefficients, but I can get only single-precision from audio math units.

So that's another question... is there a way to feed doubles into [biquad~]? (I guess, no.)

Anyway, for my own work I'll continue to use Ringz in SC, but I wanted to have a class session this term on modal synthesis, but I'll have to drop it if there's no good filter.

Thanks,
hjh

@ddw_music Intriguing! Would love to hear an example of how this sounds in SuperCollider if you've uploaded a recording somewhere.

I'm not the best one to speak to your filter question (although I know the iemlib library has quite a few filters, you could check those out).

As for double precision: yes! It can be done with the iem_dp library. However, you will have to roll up your sleeves to use it—documentation is sparse, and you'll have to compile from source. (unless you happen to be on a recent MacOS, in which case you could try the version I compiled & uploaded here)

@beep.beep "Would love to hear an example of how this sounds in SuperCollider if you've uploaded a recording somewhere."

Most of these sounds are made by modal synthesis. (This piece peaks at 15000 UGens!)

(I guess I should clarify -- SC's audio processing is single-precision. Language-side math is double-precision.)

iem_dp, interesting but probably more trouble than I can manage in the near term.

hjh

@ddw_music cyclone/reson~ seems somewhat similar.. they both seem to have difference equations similar to what is described here as far as I can tell..
https://ccrma.stanford.edu/~jos/filters/Constant_Peak_Gain_Resonator.html

@ddw_music Thanks for sharing that audio! Very interesting timbres.

And also interesting to hear that SC uses different precision for audio versus language-side math. As for Pd, I've seen plenty of discussion on the Pd-list about going full double precision in vanilla, but it sounds like a pretty daunting task, for various reasons.

@seb-harmonik.ar > cyclone/reson~ seems somewhat similar

Tried it -- with Q up around 500, it's pretty close in a side-by-side comparison against Ringz.

It'll take some more trial and error to get the attack/decay variant working -- Ringz's constant skirt gain works out so that Ringz(highQ) - Ringz(lowerQ) suppresses some of the input, while [cyclone/reson~ ... highQ] - [cyclone/reson~ ... lowerQ] doesn't, because the lower Q filter is a lot louder.

Thanks!

hjh

Audio in SC is still 32 bits, right? You can have doubles internally in the object's code, but that's the same in Pd.

Anyway, I have objects in ELSE that correspond to these SC ones. See: [else/resonant~] / [else/resonant2~] / [else/resonbank~] / [else/resonbank2~]