Anyone have a cheaper Delay module in core?

BLAIR

Hey guys,

I'm working on something which involves a looot of delays and its taking up a lot of CPU. I noticed that the normal delay in core uses 4 samples for interpolation, and I'm wondering if anyone has a more efficient version? I only need linear interpolation, so I think that's just crossfading two samples instead.

Thankss

KoaN

https://community.native-instruments.com/discussion/28005/anyone-have-a-cheaper-delay-module-in-core

There is a 2p interpolation version in Reaktor 5.5 library.

Here it is.

Delay 2p.zip

colB

https://community.native-instruments.com/discussion/28005/anyone-have-a-cheaper-delay-module-in-core

What's it for?

For some things you absolutely need interpolation otherwise it kills the sound, for others you really don't at all. Usually it's a compromise.

e.g. for reverbs, you can use delay lengths that are integer numbers of samples, and get a really great sound (better than interpolated), but then you can't smoothly change or modulate the delay length without very noisy artefacts.

BLAIR

https://community.native-instruments.com/discussion/comment/142519#Comment_142519

Thanks KoaN! This made it more efficient :)

https://community.native-instruments.com/discussion/comment/142535#Comment_142535

Physical modelling patch, so it needs non integer delays for higher notes!

colB

https://community.native-instruments.com/discussion/comment/142579#Comment_142579

Physical modelling patch, so it needs non integer delays for higher notes!

hmm, linear interpolation will increase artefacts, the problem with that in a physical modelling context isn't so much the audio quality, but the consistency of response.

e.g. for something like a karplus style model with the delay in a feedback loop, if the pitch requires a delay setting that has more error due to the linear interpolation, then it's decay characteristic will change compared to a pure setting with less error. It might decay more quickly, or more slowly, and it will also potentially effect the timbre and the amplitude. Of course, this will vary depending on sample rate, so it's not trivial to tune out of the system. Probably cheaper to just use better interpolation in the first place, which will reduce the problem, though not remove it entirely.

It would be worth setting up some A/B tests with linear vs cubic vs laplace or something, making sure to test/compare lots of different pitches and different sample rates, trying to find where the linear interpolation is at it's worst before deciding which to use.

BLAIR

i have already experimented and am fine to deal with the problems of linear interpolation :') I've decreased CPU by like 30% with the abstraction koan showed and I don't mind about the artefacts you're describing.