Open303 - open source 303 emulation project - collaborators wanted

[mistertoast]

An interesting question for antto is this: with his way of working, does his pursuit ever end?

Every time he gets a new sample, he does a bit more work to make his output match the samples.

If he starts getting new samples and finds that his output matches it, he might be on the road to completion.

But if every new clip that comes at him necessitates a reworking, the situation is as Dave says it might be.

This happens all the time with modeling. You fit all the info that you have in and then use it to make predictions. It's only when you get new data that you find out if your model has any predictive power or if you just managed to fit the data you have and didn't built a reliable model at all.

I haven't seen clear evidence that antto has been seeing a lot of wave files that cause him to say, "nothing to learn from here." If his model is good (and he's a bit lucky) he'll see fewer and fewer audio clips that surprise him.

[antto]

good question toast!

i've setup my synth in such a way, to be able to "match" an audio sample

most of the times, what i need to change are the HP Filters (probably different samples recorded by with different equipment, different A/D converters
..or the max resonance level (this kinda varies a lot, either due to "age" or modded or the different "revisions")

erm, knob response, i have no clue what are the knob responses of the original 303, this probably depends on these "potentiometers"
in my synth, i've only matched the "end points" uhm.. the full OFF and full ON parts
anything in between is completely NOT accurate at all, since i have no way to know that, i could measure it if someone records me some samples, but it's gonna be a pain for both him and me (a lot of time would be wasted, not sure if it's worth it for now)
so, anything "in-between" i configured by "ear"
this goes for the main knobs (cutoff, reso, envmod, accent, decay..)
so, Cutoff=max + Envmod=min gives 99% the same result as the TB-303 i've used as reference
but Cutoff=middle + Envmod=min.. i can bet it'll differe

anyway
give me a clean sample of a TB-303 playing a pattern (without tempo changes) and i'll give you my approximitation ;]
i don't need to know the pattern, i'll "compute"it ;P~
or give me some other 303-like sample and i'll try to tell you if it looks like TB-303 to me, or not, or fake, or.. what i think
try me ;]

EDIT:
now, few things "surprise" me lately, most of the times it's small details, not something *big*

one thing that surprised me (and it was a sample i already had) was strong resonance over the negative peak of the square wave, when the note hits
this is strange, since the negative peak shouldn't excite the filter there, and it doesn't, that strange thing only shows up when the note begins, and it's cuting the square right on the part after the negative peak

why does it do like that?
i can think of 2 possible reasons:
1. volume envelopes section being before the filter, so, when note begins, the volume envelope opens *instantly* .. oscillator output goes to the filter, and the "click" at the begining excites the filter, no matter where the waveform was at
..but wait, volume envelopes are after the filter section, so this is less possible i guess
2. in the sample, the TB-303 is playing a silly pattern consisting of 2 notes, a very low one, and a very high one
i saw this strange effect on the low one (since it's easier to see it there)
maybe the things comes out naturaly, right before the low note hits, the oscillator is still oscillating with the frequency of the last note, which is the higher one, and this is what might be exciting the filter right before the low note hits

i didn't think this is yet very important, and i wasn't gonna report it till i look around at other samples for the same effect

[Muon Software Ltd]

Here's what I think. Emulating an analogue synth of any degree of complexity is a conceptually hard (and therefore very interesting) problem. There are no "standard" tools to do it, and no standard methodology for doing so. What I'm interested in seeing is if a community of smart, distributed people with a common goal of nothing more than making something cool can get together and crowdsource a solution. Then those tools and methodologies become something useful and powerful and more than the sum of its constituent parts.

For this to happen, there are two things that need to go happen.

1. The source material, analysis data and methodology need to be placed into the public domain so that anyone with something to contribute can rock up, download the data and the tools and get analysing. The results of the analysis can be verified by peer review, and conclusions drawn. If that sounds a bit like the scientific method, well then that's not a coincidence. That's exactly how, over time, observations get refined and improved into some useful knowledge about the world. That is as true of analogue synth emulation as it is of any other kind of research.

2. Once the model has been developed, it is structured into code.

What is interesting to me here is the power of the crowd - call it "many hands make light work" if you don't like buzzwords. By working in individual silos, we're looking really at the output of only one person at a time. By not having the analysis data, tools and results in an easily consumable form, we're not getting the peer review aspect either. A 35 page forum thread of your unstructured ramblings does not provide a good starting point for someone who might want to join in to the common goal.

Either way, if after all of this the net result is only improvements to Antto's code, which is closed-source, then the open-source side of this gains nothing in particular because they still have yet to do what Antto has done.

For me this is about seeing if the community can solve a conceptually interesting problem. What we have here instead is a bunch of clearly smart people working on *seperate* goals, which is no better than what went before.

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

Well, useful to who? you? I don't want to be useful to *just* you. It isn't my goal to help you create yet another closed-source 303 emulation. As I've said, what interests me is that this *could* be a completely open, peer-reviewed process that yields a useful set of methodologies and tools for the general case. I'm also interested in how communities can collaborate to reach a common goal. That's just not happening here because you're hogging the bandwidth to serve only your own ends.

All I am saying that you should do is either contribute to the open source effort (and that means writing code for it, or providing proper documentation of your analysis and its methods). You gain by learning how to code C++ better, and the community gains what you know. Now I see that as a win/win situation, and I don't really get why you think this isn't a useful or valid suggestion. Perhaps you don't realise that people like me *employ* people like you.

I need to make a point of saying that I am disappointed very much with the way our industry cultivates new talent. I see open-source projects as a good route into employment for newcomers to this industry as it shows a) how good your analysis and coding skills are and b) how well you work with others, which is critical. Don't forget that people like me are looking to employ people with good analysis and coding skills who work well with others.

Enough from me - either you don't want to contribute to the open-source project or you do. As I said before, its your time and you choose how best to spend it.

[Muon Software Ltd]

This happens all the time with modeling. You fit all the info that you have in and then use it to make predictions. It's only when you get new data that you find out if your model has any predictive power or if you just managed to fit the data you have and didn't built a reliable model at all.

This is an exceptionally good point about top-down versus bottom-up modelling. I don't think that the tools exist to do a good job of top-down modelling (and true bottom-up modelling in this problem domain isn't happening right now). I'd be interested to see if a group of clever, motivated people can come up with an approach that works better than one individual developer working in isolation. Even better if the methodologies are open, available to all and subject to peer review and continuous improvement.

[antto]

taking up the bandwitdth?
there is no one else talking much otherwise, and i feel i got something to say, should i shut up like you?

documentation?! evidence? analyzis?
at the moment, i can't really tell i know of solid facts about a lot of the aspects of the TB-303
i am still learning it, how it works, how it behaves, why..
a lot of times i say something, then after some time i correct it, because it appears wrong now

i could write a book about it probably after a few years, when i get totaly insane, but naah.. can't wait so much time

my approach is try/fail/try_again

Dave: you're a big company, i am a poor avarage man with a hoby.. let me breath!
the guys here will tell me if i should leave, not you
yet, you have nothing to do with this project anyway, you said that yourself, so why do you care so much.. blah

[Muon Software Ltd]

taking up the bandwitdth?
there is no one else talking much otherwise, and i feel i got something to say, should i shut up like you?

Nobody else can contribute because you haven't published your code, or your methods. We can't see what you are working on, nor can use the same tools you are to perform your various analyses.

Open source projects work because people contribute code, review the code, improve the code etc.

my approach is try/fail/try_again

Which is really a private, self-driven process not conducive to shared effort. If you were to frame a problem, talk about your approach, publish your code and then invite comments, review etc you might attract people that might want to help you. Otherwise we're just watching over your shoulder and offering encouragement (or getting bored).

Dave: you're a big company, i am a poor avarage man with a hoby.. let me breath!

You're certainly a motivated and talented person. Unfortunately that's still not enough to make me want to employ you. Perhaps you have no ambitions in that respect but if you like coding it is natural to want to be better at it. Working in isolation will slow your progress towards becoming a better coder.

the guys here will tell me if i should leave, not you
yet, you have nothing to do with this project anyway, you said that yourself, so why do you care so much.. blah

Well believe me if you carry on as your are, the life of this project will disappear and people will forget about it and move on.

I only care because of the reasons I have already stated very clearly.

1. I want to live to see high quality analogue synth emulation become a solved problem - and if it starts with the 303 then that's great as its a synth I love too.
2. I believe that open-source, collaborative effort may yield interesting tools and methodologies to meet that goal
3. I'm concerned that new talent in this industry is simply not being given the help it needs to grow. Open source projects might be a way for talent to become obvious.

So as you can see, I don't really care much for what *you* are doing, working alone in your room. If I can encourage you to work with *others* towards a common goal, I may yet achieve my own aims.

[antto]

taking up the bandwitdth?
there is no one else talking much otherwise, and i feel i got something to say, should i shut up like you?

Nobody else can contribute because you haven't published your code, or your methods. We can't see what you are working on, nor can use the same tools you are to perform your various analyses.

wtf?! have you read these 30 pages?
you sound to me like you just want to see a "working" code that you can just copy/paste or something
i haven't given up that kind of thing, i have explained in detail how my filters and oscillators work, envelopes, curves, slides, sequencer
what more?!
i'm not convincing anyone to go after me, i am just working on my own, and sharing what i come accross at, while providing enough details so others can judge for themselves
tanh(hpf(sawtooth)*-amp) <- you really need code for that?! oh god!
"i've added a HP filter in the feedback path.. it takes of the resonance for low cutoffs" <- you can't figure how to implement this in a ladder filter?
i'm sure Robin and the other guys CAN
you wan't more examples?

my approach is try/fail/try_again

Which is really a private, self-driven process not conducive to shared effort. If you were to frame a problem, talk about your approach, publish your code...

..my code again?

You're certainly a motivated and talented person. Unfortunately that's still not enough to make me want to employ you. Perhaps you have no ambitions in that respect but if you like coding it is natural to want to be better at it. Working in isolation will slow your progress towards becoming a better coder.

oh, you wanted to employ me? what for? to take my synth and sell it? i would rather starve but not sell this project, not this one
no, i don't want to sell it, it must be freeware, and i would definately colaborate with someone else, who is less "cowardish" like all the guys who helped me so far with various things
..in contrast with you, who just hints "i know everything, but i won't tell you.."
keep it to yourself then, big deal, sooner or later we'll know enough to get a good general 303 model here

Well believe me if you carry on as your are, the life of this project will disappear and people will forget about it and move on.

i don't think so, just because you left your 303 project to gather dust doesn't mean that the guys here aren't motivated enough

So as you can see, I don't really care much for what *you* are doing, working alone in your room. If I can encourage you to work with *others* towards a common goal, I may yet achieve my own aims.

maybe if you were here since the begining of this discussion (you know, the 30 pages) things would have been clearer for you, you would know what we were doing..

[Kritical Audio]

Here's a quick recording of my 303
http://www.sendspace.com/file/qw9iuh

Here's another one:
http://www.sendspace.com/file/w5ox3c

(it is slightly shuffled with Silent Way lfo)

"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"

@Dave, Come on let's be honest here. You are not even a little close with tau mk2 compared to a real one. When rebirth first came out folks reviewed it as "Identical" too. I would love to hear a TB-303 that sounds "identical" to tau mk2!

Cheers,

/KA

[antto]

Here's a quick recording of my 303
http://www.sendspace.com/file/qw9iuh

nice
it's kinda 01:30 and i should be in bed already
but i'll try to approximate that, clock starts ticking..

[rv0]

interesting and sometimes funny twist of the topic here.

@Muon: as you must have read by now, I have been sending antto samples for his project way before this topic ever started. As for his approach, I had my doubts at first, but considering the quality his work and talking to some electrical engineers with dsp knowledge, they kinda praised his approach in a way. I dunno.. The results are here and imo he's not really spinning arround in circles. You said he doesn't contribute his finding: on the contrary, read this topic carefully and he has shared every single bit of information... but the medium is not that great ofcourse..
As for a decent sourceforge or alternative repository where all files and research data are hosted: GREAT. I can't agree more. But that's up for the topicstarter and codebase-starter: Robin. I can't really imagine that antto's synth edit stuff is of any for use for the cpp project. I checked Robin's code briefly and it appeared to be well built from the basis.
I'm no coding genious in cpp either and due to other stuff I don't have time to contribute yet. But a versioning system with branches would really encourage;

Well, you seem to know enough about it, but you do claim you'll never try a realistic modeling of an analogue synth ever again. I don't understand, I have no experience with your software. If you say you measured the internals and stuff, how come there was so much fuz about it (i dont know, thats what i understand from reading the last pages).
What's your opinion on antto's clone compared to yours?


@antto
As I've said before, I really hate the secracy and myths and claims some companies make concerning analogue gear. In hard- and software world that is. Debunking all these myths (I think every clone so far - hard and software - can be debunked some way) on a page along with open source code in whatever format would be a very noble project.
If there's really no other way, I can set up a wiki of some sort on a webhost.. If nobody more involved than me has the option to do that..
If you really have problems with the waveform, I could ask arround you know.. I hate to say but I got the impression some people - not only here + i rarely read who posts what - claim to know things, but don't want to share the information.


Damnit I should sleep too.
Meanwhile, need any new samples?

[antto]

stop the clock, 303_Tweaks_test.ogg 1.5MB

well, kinda like 1 hour
the pattern took me like 10-15 minutes, the rest was automation, i finaly tweaked the HP filters to match the "cheap" A/D convertors that KA seems to have used?
kinda close, eh?

[mistertoast]

stop the clock, 303_Tweaks_test.ogg 1.5MB

well, kinda like 1 hour
the pattern took me like 10-15 minutes, the rest was automation, i finaly tweaked the HP filters to match the "cheap" A/D convertors that KA seems to have used?
kinda close, eh?

Sounds nice, but now what? There are some huge differences in what's happening in some parts.

[antto]

thanks rv0
mistertoast had the idea about collecting all "solid" facts about the synth in a document, which would be great
i think i did contribute there, figuring some facts about the envelope knob ranges, the slide-LPF, and others
would be great

samples? yeah, definately, but tomorow

btw, as i did this last sample, i noticed the 303 square had also this "ugly" edge which i was concerned about, nailed it again? ;]

[mistertoast]

Yes I am still looking around for a good storage solution, as I want to pair "facts" with audio samples.

[antto]

stop the clock, 303_Tweaks_test.ogg 1.5MB

well, kinda like 1 hour
the pattern took me like 10-15 minutes, the rest was automation, i finaly tweaked the HP filters to match the "cheap" A/D convertors that KA seems to have used?
kinda close, eh?

Sounds nice, but now what? There are some huge differences in what's happening in some parts.

yeah, that was kinda fast, what did you mean exactly? i didn't match the knob automation closely, this would take more time (trying to emulate KA's fingers..)
and.. there is still a big miss in my gimmick circuit, don't forget that, the "normal>slide>accented" issue which sounds incorrect in some cases

EDIT: i just figured, my pattern was a little wrong, two notes needed to be one octave lower, dumb
i did this on my speakers, kinda low, since people are sleeping here, so i didn't hear the whole thing good..

also, i didn't automated anything after the part when resonance drops, while on the sample, there is knob activity, resonance is altered, cutoff too, accent too.. i was in a hurry ;]