supradough poc

[felixroos]

this is a poc for implementing a small part of superdough without ties to webaudio. it can be used like this:

doughsamples('github:eddyflux/crate')

all(x=>x.supradough())

$: s("crate_bd*[<4 [8 0]> 0] crate_rim")

$: s("<pink white>*8").dec(.07)
  .rarely(ply("2")).delay(.5)
.hpf(sine.range(200,2000).slow(4)).hpq(.2)

$: note("c,eb,g,bb").s("sine").press()
  .add(note(24))
  .fmi(3).fmh(5.1)
  .dec(.5)
  .delay(".6:<.12 .22>:.7")
  .jux(press).rarely(add(note("12")))
  .lpf(400).lpq(.2).lpd(.4).lpenv(3)
  .fmdecay(.2).fmenv(1)


$: chord("<Cm Cm7 Cm9 Cm11 Fm Fm7 Fm9 Fm11>").voicing()
  .s("<sine>").clip(1).rel(.4)
  .gain(.4)
  .fm(1.4).att(1).lpa(.5).lpf(200).lpenv(4)

the supradough method completely bypasses superdough and spawns an instance of the dough-processor worklet DoughVoice a for each event. DoughVoice spawning and despawning is handled by Dough, which runs in the long-lived dough-processor worklet.
the Dough class is self contained, so it could be used to render a strudel pattern in any environment that supports js (e.g. node.js) + it would be easy to prerender a strudel tune from a pattern, without the use of the web audio api. the dough.mjs contains ugens from kabelsalat. we don't really need a graph, as superdough is static.

superdough controls

sounds

• sine
• tri(angle)
• saw(tooth)
• zaw
• square
• supersaw
  • stereo
• pulse
• white
• pink
• brown
• crackle, dust
• samples
•  bank
• gain
• postgain
• orbit
• velocity
• n
• pitch zones

fx

•  distort:distortvol
• penv, pattack, pdecay, psustain, prelease
• fm(i) / fmh
• fmenv, fmattack, fmdecay, fmsustain, fmrelease
• cutoff (lpf) / resonance (lpq)
•  lpenv, lpattack, lpdecay, lpsustain, lprelease
•  hcutoff / hresonance
•  hpenv, hpattack, hpdecay, hpsustain, hprelease
•  bandf, bandq
•  bpenv, bpattack, bpdecay, bpsustain, bprelease
• vib / vibmod
•  phaserrate, phasersweep, phasercenter, phaserdepth
•  room, roomfade, roomlp, roomdim, roomsize, ir

more fx

• delay
• delayfeedback
• delaytime
• density -> dust
•  coarse
•  crush
•  pan
•  shape, shapevol
•  vowel
• ftype (filter model)
• drive (for ladder filter)
• fanchor

other

• fft
•  analyze
• i (whats this?)

[daslyfe]

this looks excellent, I will try it out tomorrow morning

[daslyfe]

Works pretty well! I noticed that with long release times it ends in a popping sound,
$: note("c eb g bb").add(note(12)) .lpf(sine.range(600,7000).slow(2)).lpq(0.2) .s("sawtooth") .att(.3).release(1) .supradough()

volume could also be lowered to match superdough volume. The filter sounds really tasty

[daslyfe]

fixed the release envelope, volume, and clicking issue :)

[daslyfe]

this rocks

[felixroos]

fixed the release envelope, volume, and clicking issue :)

nice! i failed to find the bug with my sleep deprived brain on the way back :D

i still have question marks on how to use this tactic to implement the rest of superdough, mainly:

1. global fx
2. samples
3. ??

i see 2 potential ways:

1. spawn only a single worklet and send haps as postMessages + do voice allocation in dough
2. keep spawning one worklet per hap, implement fx in separate long living worklet + wire them together

imo 1 would be the "purest" approach, making it maximally portable. not sure what's more performant, but i guess we have to try

[felixroos]

ok i've implemented 1 and it seems to be working :) i think i like it. the Dough class could now handle global fx as well

[felixroos]

web audio glue is now super minimal :)

[felixroos]

i've added another poc (dough-export.mjs) to show how to use Dough in a node.js script to render the audio for a pattern

[felixroos]

i've added this to be able to register worklets from the outside without depending on them from superdough

[felixroos]

still a bit of refactoring to do, to make sure stereo samples (and eventually stereo waveforms) are processed accordingly.. all the local effects ugens need to exist once for every channel i guess. envelopes are only needed once. also not sure how to best integrate the existing sample map loading logic

[felixroos]

curious: why only decay and not attack / release ?

[daslyfe]

It should be on the release and decay stage, the reason is that seems to be standard on most synthesizers so I played around with different curves and preferred the standard linear rate on attack, it was not scientific

[felixroos]

still a bit of refactoring to do, to make sure stereo samples (and eventually stereo waveforms) are processed accordingly.. all the local effects ugens need to exist once for every channel i guess. envelopes are only needed once.

fx now work channelwise! so stereo samples should now work

[felixroos]

also not sure how to best integrate the existing sample map loading logic

the doughsamples function can now be used to load a sample map:

doughsamples('github:eddyflux/crate')
all(x=>x.supradough())
$: s("crate_bd*[4 0] crate_rim, crate_hh*4")

for now, all samples are immediately loaded into memory! not sure if this is a good idea.. also n is not supported yet.. so far only the first sample for each key is loaded. also, pitch zones are not implemented yet

[felixroos]

new demo pattern:

doughsamples('github:eddyflux/crate')

all(x=>x.supradough())

$: s("crate_bd*[<4 [8 0]> 0] crate_rim")

$: s("<pink white>*8").dec(.07)
  .rarely(ply("2")).delay(.5)
.hpf(sine.range(200,2000).slow(4)).hpq(.2)

$: note("c,eb,g,bb").s("sine").press()
  .add(note(24))
  .fmi(3).fmh(5.1)
  .dec(.5)
  .delay(".6:<.12 .22>:.7")
  .jux(press).rarely(add(note("12")))
  .lpf(400).lpq(.2).lpd(.4).lpenv(3)
  .fmdecay(.2).fmenv(1)


$: chord("<Cm Cm7 Cm9 Cm11 Fm Fm7 Fm9 Fm11>").voicing()
  .s("<sine>").clip(1).rel(.4)
  .gain(.4)
  .fm(1.4).att(1).lpa(.5).lpf(200).lpenv(4)

[daslyfe]

Loading all samples into memory is one of the problem areas in superdirt that causes initialization to be slow, if we can stream from disk somehow or only load samples being used that would probs be better

…

On Mon, Jun 9, 2025 at 11:29 AM Felix Roos ***@***.***> wrote: *felixroos* left a comment (tidalcycles/strudel#1362) <#1362 (comment)> new demo pattern: doughsamples('github:eddyflux/crate') all(x=>x.supradough()) $: s("crate_bd*[4 0] crate_rim") $: s("<pink white>*8").dec(.07) .rarely(ply("2")).delay(.5).hpf(sine.range(200,2000).slow(4)).hpq(.2) $: note("<c2 g2>,c,eb,g,bb").s("sine") .add(note(12)) .fmi(3).fmh(5.02).dec(.2) .delay(".6:<.12 .22>:.7") .jux(press).rarely(add(note("12"))) .lpf(200).lpq(.2).lpd(.2).lpenv(3) .fmdecay(.1).fmenv(1) .postgain(.8) $: chord("<Cm Cm7 Cm9 Cm11 Fm Fm7 Fm9 Fm11>").voicing() .s("<sine>").clip(1).rel(.4) .gain(.4) .fm(1.4).att(1).lpa(.5).lpf(200).lpenv(4) — Reply to this email directly, view it on GitHub <#1362 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALHDMLQZPWJHWV4VGYRAUDL3CVHW7AVCNFSM6AAAAAB6LJN7KGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDSNJVGIYTKOJWGE> . You are receiving this because your review was requested.Message ID: ***@***.***>

[felixroos]

i'd go for a property on the class, rather than creating an array for each sample (avoids memory allocation (i think))

[daslyfe]

Added delayspeed to create pitched and reversed delay effects, implementation inspired by this article

note("d d a# a".fast(2)).s("sawtooth").delay(.8).delaytime(1/2).delayspeed("<2 .5 -1 -2>") .supradough()