Open303 - open source 303 emulation project - collaborators wanted

[rola]

I think it's completely understandable that you want to continue with your own project. You've already put a lot of effort into this and I think it already sounds great. Maybe even better than the two commercial clones.
If you share what you learn and how you implement things, others can easily include it in the open source project. Since you're still exploring different possibilities of how certain aspects of the sound are formed, contributing to an open source project might even be counterproductive and distracting.
The approach to use an envionment like synthedit is probably the smartest choice at this stage. The whole interface stuff just takes way too much time that can be used to get the sound as close as possible.

[Muon Software Ltd]

ok, i get your point, i'll shut up untill Robin shows up

You misunderstand - I certainly don't think you should shut up, I just think you should structure your communication so it benefits the open-source project. If that is indeed your goal.

the filter "opening/closing" you mean that on the samples i got, it's modulated by the envelopes and because of that, i can't tell what is it's frequency/phase response?
hm.. i hope you didn't mean that

What you are looking at is a slice in time of something that is changing. You would not easily be able to work out the exact frequency and phase responses of the filter for a given cutoff/resonance unless you developed a clever technique that takes into account the modulation across the time slice you are measuring. I can't think of a way in which that could be done.

Either way, LTI systems are easy to model - and the TI in LTI stands for "time-invariant". Don't even get me started on whether or not this system (303 filter) is even linear.

So I still say you'll never get it exact, its always going to be an approximation.

if everyone thinks like you, there will never be one single person to start such a project

Well Robin already started it. I've no intention of starting an open-source project myself, but I *might* contribute to a good one in the future. In my opinion decent analogue emulation should be one of those dead, buried, solved problems.

Robin helped me in the begining, with the filter (i'm not forgetting the other guys there!) and i owe him a lot, i would do whatever i can to help him, even that when i listen to guys like you, it seems i can't?

No again you have misunderstood. I said there was a win/win situation waiting for you - put what you have discovered into the code, and you will learn a lot from doing so. Work with Robin and other developers of his level of talent, and you will learn immeasureably more.

If you don't have time/willingness to do that that, document your findings and your analysis techniques properly so others can follow in your footsteps. Otherwise I don't think you really are contributing to the signal:noise ratio.

you killed my motivation for today, damn, gonna drink my coffee

Well obviously I'm disappointed that I have misunderstood my intentions. If you're working parallel to the task and not really interested in the open source aspect of things that's fine, its your time to spend and no-one elses. Just don't be under the illusion that you are helping anyone except yourself.

[antto]

My 2 cents on your process - you can indeed look at waveforms and samples all you want, but how can you be sure the square shape that you are so keenly trying to emulate is not just the result of phase shifts within the filter and the oscillator? and can you tell what effect this has or doesn't have on the sound? if you're only ever looking at the filtered output you are not really in a position to judge. Either way, don't get me wrong, I respect what you are doing and I look forward to the final open-source effort (though I'm confused by the number of people in this thread the seem to be working on seperate projects and not really much helping Robin it seems).

that was me who was "keenly trying to emulate" the square wave
Robin hasn't yet said anything about it (been offline or something)
well, i suppose you don't think i am looking at a screenshot of the square waveform of a sample, at some note?
in fact, i have an audio sample (from rv0) with the TB-303 playing all the keys on the sequencer (all notes) + transpose, with the filter open, without resonance
and i put my synth next to this sample, configured the same way, with the same tuning, so if there is a difference, it would either be the filter, or the oscillator, you agree?

i can upload a sample of that, so you can see that i got it close
i can also explain why i think the oscillator is doing this and that (based on looking at the output)

as i said earlier, i am sure about 1 thing, there is a high-pass filter filtering the sawtooth, before it goes into the shaper (which turns it into a "square")
i got really close results like that
i could still be wrong, but i need some strong argument, not just "you are wrong."

i will agree with you if you convince me that you are right, or that i am definately wrong
yet you haven't

[Muon Software Ltd]

I know it was you, not Robin - and I know he's been quiet, possibly because he can't get a word in edgeways

I'm not necessarily saying you are wrong, I am saying that there are several reasons why what you are doing can only be considered to be an approximation and not an exact emulation.

1. You don't know what the input to the filter is, so you cannot compute its response (phase + frequency).

2. If you take an FFT of any reasonable size at any point in your sample data, you are taking an FFT of some non-static filter cutoff frequency. Again, this leads to difficulties in calculating its frequency response for a given cutoff.

3. There are a range of possible inputs and filter responses that could give you the net results you are looking at.

Consider the effect of phase shifts on waveform shape and the audible frequency response. One of the characteristics of an allpass filter, for example, is that it shifts phase without much affecting frequency response. So a bunch of differently tuned, static allpass filters would make a big difference to how a periodic waveform might look, but little difference to how it sounds. That'd be a satisfactory demonstration of one of the key properties of an LTI system

If you are simulating an oscillator, feeding into a filter, feeding into an amplifier you have to either have a good knowledge of what the input and output of each individual component is doing OR you have to say "I'm happy with a good approximation of the final output".

Kind regards
Dave

[Kritical Audio]

Antto, Don't listen too much what Dave says. I mean.. Download the Tau mk2 demo and have a listen caugh caugh.. Good work on the latest beta! You are getting there! Maybe too much clicks so perhaps higher the release a tiny bit more?

/KA

[rlahalla]

Antto, Don't listen too much what Dave says. I mean.. Download the Tau mk2 demo and have a listen caugh caugh..

Eh? Wow, that wasn't very nice. Dave's a really good programmer.

[Muon Software Ltd]

Download the Tau mk2 demo and have a listen caugh caugh..

If I worried about what the KVR weenies thought of Tau 2 I'd have had a nervous breakdown, so its a good job I couldn't give a stuff

Proof of the pudding is in the sales figures.

[mistertoast]

There are a lot of software attempts, obviously. Robin's doing a great service to give a open source project a try, and I applaud him for it.

annto's doing a great job. We'll see how he scores when he's ready to submit to this: http://www.acidvoice.com/tb_303_vst_plugin_software.htm

Tau Bassline MkII doesn't fare especially well on that site. But it could very well be a bunch of KvR whiners voting there.

[Kritical Audio]

@rlahalla I bet he is in some areas sure, but not in emulating the TB-303. Just my opinion.

[antto]

grr.. the picture got ugly again..

Dave: i'm not doing FFT
some one gave me a ladder filter, with the coefficients, he said that it was *close* to a basic 303 ladder filter, i wasn't sure till i tested it
let me tell you a little bit about it:
i'm comparing my synth, with an audio sample of a TB-303
in the sample, at first, there is no resonance
i tweak the volume of my synth so the level is equal
then, in the sample, TB-303 resonance is increased to maximum (erm, the knob range)
i increase the resonance of my synth as well, guess what? both filters drop the gain equaly, i didn't get the same effect with a moog ladder filter

this was the first clue that the filter could be it..
then, i look at how the waveform changes shape (due to the phase shifts in the filter) when i increase resonance (again, comparing it to the audio sample)
guess what? same *visual* effect, i know what allpass filters do, i know how a basic waveform can change it's shape while sounding the same, don't worry
so, this was another clue that i am going right at that moment

then, i compared how the synth sounds when there is no big modulation of the filter (for example no envmod, static cutoff, long decay)
resonance is increased, my synth resonates in the same way as the audio sample (you know, there is always a little env modulation)
so then, on the sample, the resonance is decreased to minimum, i do too
guess what? both filters drop to the same cutoff frequency (you know, the TB-303 filter has different (wrong) frequency response when with resonance and without)

so, that was another clue
oh, and it sounds good
and better than anything else i got..

and i was mainly testing with the sawtooth waveform (audio samples of sawtooth too)
so i couldn't get much wrong results, it's so simple, a sawtooth, increasing voltage, then a fast drop..

Kritical Audio: i respect Dave for Tau mk2, even that i have no clue how it sounds or anything
but considering that it was made long ago - i just can't dissrespect that
i am sure that if he rewrites that synth, with the knowlage he has now, and if he is serious about it, he will be rockin, tho, it seems he has other plans

but i am pretty harsh about synths that are recently developed, like for example that DB-303 for the iPhone i mentioned earlier (i was joking)
it's made for iPhone, so it must be new
read what's on theyr website, they try to convince you it's the most accurate emulation, then, hear the samples - wtf?

that kind of advertisement annoys me a lot
if i wanted a nice emulation, i would go for ABL, and it is also available for that iPhone (tho, i don't have one)

Dave: i am not doing an exact emulation, i understand that whatever i do, it will be an approximitation
and i also know that the difference between each TB-303 can be pretty big, and btw, this gives me a little "room"
this is also the reason why there _can't_ exist a "perfect" TB-303 clone, since there is no "perfect" TB-303

[antto]

There are a lot of software attempts, obviously. Robin's doing a great service to give a open source project a try, and I applaud him for it.

annto's doing a great job. We'll see how he scores when he's ready to submit to this: http://www.acidvoice.com/tb_303_vst_plugin_software.htm

Tau Bassline MkII doesn't fare especially well on that site. But it could very well be a bunch of KvR whiners voting there.

you know, i already made a 1:1 test with that pattern, it *sounded* quite close
i guess it'll be a bit better with the changes i did since then (btw, that sample is not really clean, but anyway)

[Muon Software Ltd]

Dave: i am not doing an exact emulation, i understand that whatever i do, it will be an approximitation
and i also know that the difference between each TB-303 can be pretty big, and btw, this gives me a little "room"
this is also the reason why there _can't_ exist a "perfect" TB-303 clone, since there is no "perfect" TB-303

I couldn't agree more that a perfect emulation cannot exist. This is because the tools we are using do not measure what needs to be measured in order to achieve it.

To create an decent, but obviously not perfect emulation, you need to recreate the oscillator, filter, envelopes and amplifier etc. with a high degree of fidelity. A synth is not an LTI system that you can bung a test tone into and get a response. Instead is it a bunch of discrete components that dynamically, together, create a *huge* range of possible nonlinear outputs in relation to their control inputs. Unfortunately, with the tools we have we can only process a series of snapshots of this dynamic system at best. You know that accuracy in the frequency domain comes at the expense of accuracy in the time domain (and vice versa). You know that what you see in the time domain may be the result of inaudible phase differences. But these change over time, even with the same inputs because the synth has a huge range of possible outputs. In fact, the scale of possible outputs is staggering.

Sampling these outputs is effectively a form of data compression, and what you are doing is really just "training" your model so that it is perceptually the same (to your ears/eyes) for the range of inputs and outputs that you tested. If you test a wide enough range of inputs and have a flexible enough model you might be able to come up with something that works. What I am interested to see if this can be distilled or refined into a generic process that produces a "good enough" approximation of a synth.

Unfortunately, whatever you are doing, is personal to you and your project and not really being shared in a form that it can be consumed by others who might want to work to this goal. I really don't have anything further to say.

Tau Bassline MkII doesn't fare especially well on that site. But it could very well be a bunch of KvR whiners voting there.

I said "weenies", not whiners. Tau Bassline MkII fared very well in a group test by Electronic Music magazine and was rated as being "identical" to the 303 under test. As it is human nature to prefer evidence that supports a particular point of view, I obviously prefer that review

[Kriminal]

leave it Dave, its not worth it

[rola]

Unfortunately, whatever you are doing, is personal to you and your project and not really being shared in a form that it can be consumed by others who might want to work to this goal. I really don't have anything further to say.

I think he has shared quite a lot of what he has been doing and anyone can take that information and do whatever they like with it. All you have to do is read this thread. It seems to me you're the one not willing to share any of your 'secrets'.

[antto]

Unfortunately, whatever you are doing, is personal to you and your project and not really being shared in a form that it can be consumed by others who might want to work to this goal. I really don't have anything further to say.

I think he has shared quite a lot of what he has been doing and anyone can take that information and do whatever they like with it. All you have to do is read this thread. It seems to me you're the one not willing to share any of your 'secrets'.

yup, i think everyone here knows the code for a 1 pole HP filter, even better than me

btw, leave it, it seems both Dave and Dave are trying to look "cool" here, without being useful

Dave: why do you think that what i'm doing can't be "consumed" by the others here? (Robin, Toast..)
or was you refering to yourself and Kriminal?