Open303 - open source 303 emulation project - collaborators wanted

[autodafe]

this is indeed strange - they sound (and also look) indeed almost identical. this is in contrast with ABL where there seems to be an even faster decay added for accented notes (which looks linear) such that the envelope essentialy becomes a 3-stage decay. mmmhhh.

I also did a quick comparison in soundforge and the seem identical...

As I told you, my TB has a Kenton MIDI retrofit and has an inout to control accent. I guess that the mod disables the Accent knob, but you can control via MIDI CC...
If you need I can further investigate this.

[Music Engineer]

...moreover, there seems to be some strange thing going on when resonance is active (which becomes more pronounced on higher notes) - might this be the transient response of the filter? ...we'll see.

however, i was really surprised by the very low cutoff frequency for the fully open filter - it was somwhere around 500 Hz - can this be?

[Music Engineer]

I also did a quick comparison in soundforge and the seem identical...
As I told you, my TB has a Kenton MIDI retrofit and has an inout to control accent. I guess that the mod disables the Accent knob, but you can control via MIDI CC...
If you need I can further investigate this.

that would be very cool. it indeed looks like the accent was proably somehow bypassed.

[antto]

most DAWs have linear slides

on a slide, there is no retrigger, retrigger is only when there was silence before that

Filter Envelope - Decay Time for accented notes is different, it's constant, it's the same as turning the Decay knob all down (which is still not too fast)
if a normal note is slided to an accented note, when the accented note kicks in, the Decay Time is switched to the short one (for accented notes) and the Envelope falls down faster. but from the same point
to achieve this, yo better use a linear Envelope with decay time that can be modulated while the envelope is runing..
then put a shaper on that

another way is.. i think since the decay curve looks pretty much like a nice exponent, and reminds me of how a 1 pole HP filter output falls down when a step has been fed to it
then your decay time would be the HP filter frequency and calculation of the time is somewhat confusing at first

not yet listened to these samples from autodafe, gonna reply when i do

[antto]

autodafe: not sure what happened to your TB, but it's odd! or you didn't put the right names on the files
the file with R0 D0 C100 is just wrong, i see a looong decay on the filter envelope, either your unit was "hacked" or you wrote the name wrong
A100 <- wtf? have you put accents on the notes?

ok, as far as i see you've got a MIDI modification? damn, sorry
just i don't think this is a good reference then

[Music Engineer]

if a normal note is slided to an accented note, when the accented note kicks in, the Decay Time is switched to the short one (for accented notes)

this is interesting. i did not yet take this into account. will do for the next update.

however - here are my results for matching the amplitude envelope to one of autodafe's samples:

i add two exponential decays (both can be generated by a one-pole leaky integrator) in the following way: the main envelope generator has a time constant (as in the EE definition of 'tau') of 1230 ms, the other (fast) one has tau = 58 ms, and the second one is added to the first with scaling constant of 0.76 (assuming unity as scaling factor for the main envelope). to sum up:

tau1 = 1230 ms
tau2 = 58 ms
c1 = 1
c2 = 0.76

[Music Engineer]

wondering where this second envelope comes from, i did a quick test with the ABL (which shows smilar behavior too) - i checked whether the time constant of the second part depends on the decay parameter - and: yes it does. conclusion: the filter envelope is - to some extent - added to the amp-envelope ...i think

actually, the devil-fish states so as well, but it seems that he's talking only about accented notes there - but it seems that filter envelope is added to the amp-env even on non accented notes. but probably to a lesser extent? more things to check...

[larm]

do you need more samples from a real one? (mine is unmodded)

[fliffo]

why a sub-osc? Wasn't this a 303 emulation?

yes, but as i already stated in the original post:

the plugin should also include the Devil Fish modifications and some useful extensions on its own.

sorry, just wasn't aware of wat the devilfish modifications were about

[Music Engineer]

do you need more samples from a real one? (mine is unmodded)

oh yes, i'm sure i will need more samples during development, so thanks for the offer. i will always post here when i need some samples for analysis and also specify the settings of the knobs then.

mmm... maybe you could also create some samples with zero resonance, cutoff fully open, zero env-mod, zero decay of accented and non-accented notes? just to verify the suspicion that the accent was somehow bypassed in autodafe's samples.

in the meanwhile, i'll try to match the waveforms to autodafe's samples (i have them currently matched to ABL, but these look significantly different from autodafe's).

edit: and it will of course also be of great value to see the variability between different units.

[mistertoast]

I'm trying to make sure I understand what antto is saying about the slide. Let's say I have two notes and the first one is marked as a slide. The frequency is going to change during the playing of that first note to that of the second note.

1) Let's say (for the sake of the explanation) that my bpm is set so the duration of a note is 10,000 samples. How far along has the frequency moved after 2500, 5000, and 7500 samples have gone by? Al I need is a rough idea to help me conceptualize what you are saying?

2) By the end of the first note, has the frequency moved the whole way to the second note's frequency?

[mistertoast]

http://www.oakleysound.com/tb3031-doc.pdf

This documentation for a 303 clone has this interesting detail. Can anyone verify its truth?

13. The secret grit weapon. In the TB303 the slide function is always triggered prior to anynew note. This was not really by design, more of a by-product due to cost saving. However, itmay be responsible for some of the grit associated with the attack portion of the TB sound.Many commercial clones have been criticised for not having the bite that the original had. Myspike circuit may well do the trick. This, like the TB303, slews the pitch CV before every newnote for a small fraction of a second. In the TB303 it was tempo dependent, but in mine it isfixed, since we don't know what speed the sequence will be run.

[antto]

do you need more samples from a real one? (mine is unmodded)

samples are always welcome (since we both don't have a TB-303 in our hands you know..)
tip: sometimes, when you have max-reso and both EnvMod/Cutoff high, you should record at 88200Hz since the filter goes up to about 27-28KHz and this cannot be catched at 44100Hz sampling rate

Robin: how do you mesure the amp envelope!?
if the answer is "visualy" (or dB meter) then i should say that you cannot prevent the filter from modifying the sound

there is a Volume envelope which is probably exponential, but it's blurred away by the filter
not sure if you'll understand what i'm trying to explain
the filter envelope is always modulating the filter, which smoothens the sharp peaks of the waveform

probably to analyze the Volume Envelope, someone has to record a "full-of-slides" pattern, one that has only slides (so the envelopes will be triggered only once, when the pattern starts playing and never again)
with Cutoff = max, Decay = 0
and we'll have to ignore the first part of the audio since there the Filter Envelope will be changing our perspective a bit

otherwise, yes the whole "gimmick" effect (probably the hardest thing to recreate) is done pretty simple in the TB-303
the cutoff envelope on accented notes goes to the filter as normal (tho, with a fixed short time) thru a special circuit
there is a LP filter going on there, and probably a HP too

in my synth, i've actually used another way to achieve the curves, but it only looks 95% correct
PeakFollowers.. linked in a pretty messy circuit

tho, i tried by just cloning the filter envelope and pushing it thru a LP filter, which showed me that this is the way it was done
but also, then you get closer to "emulating" that part of the circuit, and if anything in your synth that has something to do with it is not "emulated" then it will fail
now i'm thinking to try again with it (this approach will also use less CPU)

i'm trying to say that, the gimmick circuit is very tricky if you want to emulate it, since you have to emulate the other parts, which are related to it
(envelopes, exact curves, filter frequency response..)
which is hard, since we all want a "perfectly tuned" filter since it must work the same on all sampling rates
there is still a solution for this, but it gets more tricky..

blah
Robin: i suggest you to leave the DevilFish at least for now, you can always tweak a good emulation to do what DevilFish does
the hard thing is to nail the unmodified TB-303 ;]
just my opinion..

[antto]

do you need more samples from a real one? (mine is unmodded)

oh yes, i'm sure i will need more samples during development, so thanks for the offer. i will always post here when i need some samples for analysis and also specify the settings of the knobs then.

mmm... maybe you could also create some samples with zero resonance, cutoff fully open, zero env-mod, zero decay of accented and non-accented notes? just to verify the suspicion that the accent was somehow bypassed in autodafe's samples.

in the meanwhile, i'll try to match the waveforms to autodafe's samples (i have them currently matched to ABL, but these look significantly different from autodafe's).

edit: and it will of course also be of great value to see the variability between different units.

ABL is the closest (available ) software emulation that i give my thumbs up for
but it looks like Mike has modeled it after a TB-303 which has been processed by a little stronger HP filter (probably a mixer, or the PC's A/D convertor)
i have a clue about how the waveforms have to look when there is no resonance (and i know what the SAW should look like when not filtered at all *)
* and i believe it so much!

erm, the guy said "unmodified" this means there is no "Accented Decay" knob there ;]

[antto]

I'm trying to make sure I understand what antto is saying about the slide. Let's say I have two notes and the first one is marked as a slide. The frequency is going to change during the playing of that first note to that of the second note.

1) Let's say (for the sake of the explanation) that my bpm is set so the duration of a note is 10,000 samples. How far along has the frequency moved after 2500, 5000, and 7500 samples have gone by? Al I need is a rough idea to help me conceptualize what you are saying?

2) By the end of the first note, has the frequency moved the whole way to the second note's frequency?

no, grr.. it would be so simple to draw an image here (tho, i would draw something you can pretty much see on the FFT if you know where to look)

step1 and step2

1: C2 G S
2: C3 G

(G is gate, S is slide)

so, sequencer starts playing step1
the pitch is changed (immediately) to C2
envelopes are triggered (assuming there wasn't any slide before step1)
gate opens up, but since it's a slided note, the gate won't be closed at 0.5 the step time (as non-slided notes will)
Pitch stays C2 during the whole duration of step1

now sequencer starts playing step2
Pitch starts to change to C3
gate is still ON
envelopes are runing (not retriggered, they are continuing from where they were at the end of step1, blah, the envelopes don't give a damn about which step it is, or what is the tempo..)
since step2 doesn't have slide, it will be shut down at 0.5 the step duration (gate OFF)
the Pitch must reach C3 before this

so the whole pitch change from step1 to step2 occures during the first half of step2's time
the curve is very important, it probably is a 1 pole LP filter (since it looks like a step fed to such a filter)
in my synth, i used linear interpolation and a curve (figuring out the right frequency of that LP filter would be another headache)