Obxd synthesizer

[waltercruz]

this is the stuff that you as users could really help out on -- can you pin down precisely which version "sounds best," and prove (e.g. through null tests) which versions "don't work" because they sound different? that'll make it a lot easier for somebody who wants to help, but doesn't totally understand the historical context, to actually be helpful, without inadvertently making things worse.

right now, if some dev looks for "obxd source code," this is what they find: https://github.com/2DaT/Obxd

any idea if this is the right basis or not?

The version from discoDSP is here: https://github.com/reales/OB-Xd

[waltercruz]

and doing a diff between 2DaT git version and reales repository, I can see no change that would alter sounding, it's just skinning and the part of preset browser that change Banks.

[.jon]

Duplicating functionality already available is like the definition of wasting time.

[klinik]

and doing a diff between 2DaT git version and reales repository, I can see no change that would alter sounding, it's just skinning and the part of preset browser that change Banks.

The diference in sound does not come with the DiscoDSP version Vs lastest 2Dat version, but from an earlier 2Dat version to the lastest 2Dat

[fmr]

this is the stuff that you as users could really help out on -- can you pin down precisely which version "sounds best," and prove (e.g. through null tests) which versions "don't work" because they sound different? that'll make it a lot easier for somebody who wants to help, but doesn't totally understand the historical context, to actually be helpful, without inadvertently making things worse.

right now, if some dev looks for "obxd source code," this is what they find: https://github.com/2DaT/Obxd

any idea if this is the right basis or not?

Yes, that seems the correct repository. Not sure if the old code is still there, and 2Dat seems to not be around anymore.

EDIT: The code there says "2DaT Better HQ mode" but is from 2014, so,. it must be from the version 1.1, which I believe is the one people refer when they say it's "better". Version 1.3 was compiled in 2016 <https://obxd.wordpress.com/category/news/> apparently from a different code ("Compiled using dozius fork and Soshi Studio GUI code"), which doesn't seem to be there in the repository.

I still have the old version from 2014. Will try to bounce the same track with the two versions and see how different they are. A little busy right now but I will try to find the time up to the weekend.

[waltercruz]

and doing a diff between 2DaT git version and reales repository, I can see no change that would alter sounding, it's just skinning and the part of preset browser that change Banks.

The diference in sound does not come with the DiscoDSP version Vs lastest 2Dat version, but from an earlier 2Dat version to the lastest 2Dat

Nice!

So, probably the version with the filter that people talk about is the one before this commit on Oct 19, 2014, with the comment "Better 24db filter"

https://github.com/2DaT/Obxd/commit/307 ... 5535c63ad2

[waltercruz]

and doing a diff between 2DaT git version and reales repository, I can see no change that would alter sounding, it's just skinning and the part of preset browser that change Banks.

The diference in sound does not come with the DiscoDSP version Vs lastest 2Dat version, but from an earlier 2Dat version to the lastest 2Dat

Nice!

So, probably the version with the filter that people talk about is the one before this commit on Oct 19, 2014, with the comment "Better 24db filter"

https://github.com/2DaT/Obxd/commit/307 ... 5535c63ad2

or before this one, from Oct 27, 2014 (Better HQ mode).

https://github.com/2DaT/Obxd/commit/5f8 ... diff=split

Sonically, the latest version of 2Dat (Better HQ mode) == discoDSP version.

[waltercruz]

...

or the version before Sep 16, 2014, with commit (Better 12 db filter).

https://github.com/2DaT/Obxd/commit/52a ... 274c2b260b

It's not hard to build it with current Juce Framework on Mac OS X, I build it some months ago (april), but it's needed Xcode installed and I removed Xcode from my computer.

[fmr]

and doing a diff between 2DaT git version and reales repository, I can see no change that would alter sounding, it's just skinning and the part of preset browser that change Banks.

The diference in sound does not come with the DiscoDSP version Vs lastest 2Dat version, but from an earlier 2Dat version to the lastest 2Dat

Nice!

So, probably the version with the filter that people talk about is the one before this commit on Oct 19, 2014, with the comment "Better 24db filter"

https://github.com/2DaT/Obxd/commit/307 ... 5535c63ad2

If that's the case, I'm afraid I don't have it anymore, and the code doesn't seem to be available anymore either.

But I believe that the problem was introduced with the changes made to the code by Soshi, not by 2Dat. The last 2Dat version (version 1.1, I believe) don't have problems, AFAIK

[Tommi_M]

There seems to be something odd in this latest version. When playing, it sometimes goes mute, and cpu jumps to 100 %. Normally while playing it's about 10–30 %. It's not about the amount of voices... I have tried many things and I still dont know what triggers it. Usually I have to restart OBXd to make it work again.
I still have the old version of OBXd, and I never have any problems with that.
I have Ubuntu 16.04 and Renoise.
Any ideas? I'd like to use 2016 version because it looks better, but this behaviour makes it difficult.

[Teksonik]

Update to Ubuntu 17.10 ?

[aciddose]

"Goes mute" is almost definitely a NAN float value. To verify this you'll need to trigger the condition and see if the NAN value shows up in the output or in the plug-in's internal memory and buffers.

If it occurs in the filter buffers for example it means it's unrecoverable unless the plug-in has a proper initialization on reset, which is supposed to be triggered when plugin->dispatch(set status, 0, ...) followed by (set status, 1) is called. Also referred to as "suspend and resume".

NANs occur much more frequently with 32-bit FPU code and standard cmath functions like tan(). Most plug-ins use tan() to compute coefficients and some plug-ins may not account for the fact that out-of-range errors can trigger a NAN result.

The solution is likely USE THE 64-BIT VERSION.

[Tommi_M]

Ok, I have made some more tests. If there is only one voice, no problem. But playing two voices or more could trigger it. If it goes mute once, problem follows to new instruments. Only way to fix it is restart, or play only 1-3 voices.

"Goes mute" is almost definitely a NAN float value. To verify this you'll need to trigger the condition and see if the NAN value shows up in the output or in the plug-in's internal memory and buffers.

I dont know how to do that. I had a terminal open when playing, but there is no errors.

Maybe changing to 64-bit version could help, but I'm not intend to do that very soon.

[aciddose]

Well it's a serious bug in the plug-in itself. If you're unwilling to test the 64-bit version and unwilling to upgrade to a version where the bug is fixed, you're stuck.

[Tommi_M]

Well it's a serious bug in the plug-in itself. If you're unwilling to test the 64-bit version and unwilling to upgrade to a version where the bug is fixed, you're stuck.

I have version 1.4. Isn't that the latest version?