Bazille's phase knob noise

[Yorrrrrr]

In Bazille, dial a simple sine wave with the PD knob and, while playing a note, turn the phase knob. I hear some zipper noise. The faster I turn, the stronger the effect. Like it's not smooth. Also happens in Hive, but not quite as strong. Doesn't happen at all in Sylenth1, BTW.

Is that normal or intended behavior?

[Howard]

Bazille: Normal, not intended
OTOH in Hive (Constant -> Osc1:Phase in the matrix) I can't hear any zipper noise (Macbook / Ju-X host).
Could you record a snippet of that (Hive)?

[Yorrrrrr]

Bazille: Normal, not intended

Oh. So, is it something fixable? or just be done and deal with it?

OTOH in Hive (Constant -> Osc1:Phase in the matrix) I can't hear any zipper noise (Macbook / Ju-X host).

It's extremely subtle, but it's there, and more noticeable with the use of a spectrogram.

[Howard]

Oh. So, is it something fixable? or just be done and deal with it?

Very likely fixable (and might be fixed in future), but for now "deal with it" is the best advice.

[Yorrrrrr]

Thank you Howard. I hope it gets fixed. A super high quality mode or higher oversampling to lower aliasing would be great too (as an option, of course). Everything that makes u-he plugins cleaner and higher quality, taking advantage of the great CPUs in 2017, is good.

[Yorrrrrr]

Just curious..does noise in phase knob in Bazille (and in Hive constant > Osc1 phase) lead to 'noisy' phase modulation synthesis? I hope Urs can have a word on this matter. In other well known VST synths, noise is not present when modulating phase.

[Urs]

It's afaik just the interpolation of the parameter tied to the knob. Modulation of any kind isn't affected.

Oscillator Phase is typically a parameter which isn't ever automated. As such it's worth neglecting for zipper noise to squeeze another percent of CPU out of these things. If we have 100+ parameters and we smooth each and every parameter for every sample, we get a higher performance impact than any Diva filter. So we need to pick where it matters.

[david.beholder]

If we have 100+ parameters and we smooth each and every parameter for every sample, we get a higher performance impact than any Diva filter.

Can you pick only ones that are modified?

[Urs]

If we have 100+ parameters and we smooth each and every parameter for every sample, we get a higher performance impact than any Diva filter.

Can you pick only ones that are modified?

Then we have to limit the number and still use slower code to generically interpolate parameters. I'm not sure if that's any better.

[david.beholder]

Then we have to limit the number and still use slower code to generically interpolate parameters. I'm not sure if that's any better.

Definitely better by letting user decide what knobs are be interpolated - the one they are using
Yes, you should inform user that if more than X knobs are being turned simultaneously (or automated) they might have issues like itt and set this X to some meaningful value (say 8+1 by number of ableton slots)
Idea of dynamic allocation of interpolators looks like silver bullet here.

[Urs]

dynamic allocation is the nemesis of CPU optimization

[Urs]

point being... most parameters that make sense to interpolate are already interpolated for free due to the modulation assignments. Some parameters are not interpolated because the math behind them make rarely calculated intermediates desirable, i.e. it's not just interpolating the parameter itself, there may be a whole lot of additional maths involved. I usually wrap those parameters in drop downs, but I honestly can't remember what I did back then... the code is humongous, as it's all in one giant loop.

[david.beholder]

dynamic allocation is the nemesis of CPU optimization

real talk! devs itt

the code is humongous, as it's all in one giant loop.

Unrelated: Is it like voice per thread - and whole voice is just one big loop?

[DuX]

It's a feature, not a bug.

Some people love zipper noise.

[Urs]

the code is humongous, as it's all in one giant loop.

Unrelated: Is it like voice per thread - and whole voice is just one big loop?

Yep, almost. Stuff that hasn't got cable inputs isn't in that loop (envelopes, LFO, Maps). But all the heavy duty stuff is: Oscs, Multiples, Filters, Modifiers, Sequencer. It's pretty much thoroughly written in SSE intrinsics, so it's huge and very hard to read.