the well-groomed rimshot (aka feats of strength)

[xoxos]

i think it was after midnight when i thought i'd have a go at synthesizing the tr-909 rimshot..


i've expressed my gratitude elsewhere to the poster for sharing their well tuned machine. the charming illustration is from the tr-909 service manual, while not scaled exactly, you'll note the contour is a very close match.


this is a rimshot from another machine. (there was a revision to the rim, one of the resistors was swapped out, which i think does not account for the variance alone based on the other samples from the same machine.)


per the service manual, the trigger pulse is routed to three bridged-T oscillators, which are clipped, amplified and highpassed. simple enough.

the process was rather informative, glean what you may from this account.

bridged-T oscillators produce an exponentially decaying sinusoid. i chose to match this with my mass-spring oscillator algorithm. we can do ping.

frequencies took searching for discussions on the internet (in the ballpark), some unfamiliar calculations from the schematics, spectral analysis (fun with short sounds) and measuring waveform peaks. once these were close, it was easier to petition for the phase, amplitude and gain/decay of each.

the second softclipping method i tried fitted the waveform and diode contours. at this point i made an observation: the two lower oscillators both have an attack lag.. you can see initial attack is purely the highest frequency.

at this point i was getting into hours of playing the human fft algorithm, trying to match the contour by changing phase, amplitude, fine tuning frequency of three sinusoids in a softclipper. at one point i had a very nice contour match for everything after the first 450 degrees of the highest frequency... lol

so i separately enveloped the attack of each of the lower oscs and spend awhile there. there seem to be further minor vagaries, most of the differences in the contour at this point could be resolved one way or the other by changing the osc mix.

i have only aging and rudimentary electronics knowledge. after taking the data (description of ideal form) as far as i could i was still left with a rather plasticy ping that sounded like three decaying sinusoids in a softclipper there's a highpass in the circuit but it isn't high enough to affect the contour of this optimal rimshot.

i decided to model the bowed contour 909 samples have - they show a trend towards one side of dc in the attack, and bow out to the other side in the release. this is the step response of the highpass filter


(it'd go straight up if it were only a step response, the soft attack is an amendment)

this is pretty much a kick drum in and of itself, and not part of the voice architecture, but an artifact of the circuit.

adding this to the signal makes it look pretty in a wav editor and gives it some extra punch. but i also routed this contour into the clipper with the oscillators. it's the right thing to do, right? nonlinearity, dc bias? adding what i'm calling the "punch contour" gets rid of the plasticy perfect harmonics series the oscillators alone produce... it looked good, sounded, well, better than it did...

after a while i decided to invert the contour in the nonlinearity and instantly my ears told me this was better. my eyes verified it - the first few cycles on the rim exhibit a dc bias *going into the clipper* and i'd finally been paying enough attention to put that in there the right polarity.

while adding the 'punch contour' into the clipper in the opposite direction as the "make the waveform pretty" summing at the output boosted the correct harmonics, it made the waveform less pretty.

i'm stopping there for today... using a different highpass on the contour-to-clipper to produce a shorter punch and keep the tail bowing in the same way as the original sample would look nicer, but the dc bias doesn't affect the tone as much here, so to me, that would be extra cpu with no timbral justification.


1) the tr-909
2) an emulation with dc0 (no contour)
3) about the same before i inverted the contour.. looks closer, sounds less close
4) the timbral emulation

http://www.xoxos.net/temp/rim/rimdc0.wav #2 (notice chirpy plasticy harmonics on attack)

http://www.xoxos.net/temp/rim/rims.wav (authentic and emulated)

may have a bit more go at it later, driving the clipper and nonlinearity mix is where the "thwack" comes from that the dc0 sample doesn't have. without further insight into the nature of electronics to address some of the "interesting things happening with phase in the attack" that's about as close as i'll take it.

3 sines, not really a feat of strength

[xoxos]

using SPEAR to separate partials.. it wasn't until later that i used adze instead. spear's analysis window ate the attack.

stare at the three partials and see if you can see how they're summed. it's a wilderness.. there's no guarantee that the phase is coherent and ? knows what else.

i noticed there's a thread on video games there, i wonder what i'd be doing. after sorting the three freq/phase/gain dealies, those attack envelopes on the lower two oscs were like the level boss.

[xoxos]

further fine tuning, had to fix the contour (big dc offsets aren't fun to mix) flattened the lowest osc, reduced its decay, changed the dc and drive a bit.

http://www.xoxos.net/temp/rim/rims2.wav
original, yesterday's, current.



you may be reading this thinking who cares, i never want to use the 909 rimshot. if you, eg. don't code, you can now up your synthesis with a wav editor.. highpass a block of samples to make the step response curve, use a bit of lowpass to round off the attack, and layer with your percussion samples of choice.



so.. is today's "closer" an improvement over yesterday's

[ntom]

This was kind of cool - I will admit to skimming a few sections of this read, but either way, I am impressed with how close you got, to me, I don't think I could tell the difference.

SO now, my question is, what are you planning on doing with this? (Sorry if you already answered that, this is the result of me skimming)

[xoxos]

universal eventualities don't need to plan

http://www.xoxos.net/temp/rim/rims_final.wav
(original - 2 days ago - 1 day ago - latest)
tweaked the clipper and stuff, removed a chirpy harmonics series from the attack. it may not be as close as previous, i think the improvements make it a better sound in use, at least one in my mind that works in italodisco tracks. how close it sounds i think depend on the volume (fletcher-munsen).

parameterisation:

http://www.xoxos.net/temp/rim/909rimmod.wav
adds pitch, decay and drive. begins with default. listening to them individually is a better appraisal so that it sounds like itself instead of "the distorted/softer one".

http://www.xoxos.net/temp/rim/909rimtestdc.wav
added bias param, which could almost be called tone (this test goes up to 10.0 - i cut it back to 6.0).