DSP Robotics Support

It started here, but I think it could be considered as a spearate topic.

Exo offers in his developer toolkit several methods of interpolation, that can be used in waveplayer, to get better quality when playback goes for example very slow, like several octaves (and not only semitones or cents) down. There are things like:

Hermite 4pt, 3 order (x-form)
Lagrange 4pt, 3 order (x-form)
Optimal 2x (4pt, 4 order) (z-form)
Optimal 2x (4pt, 2 order) (z-form)
B-spline 4pt, 3 order (x-form)
Watte tri-linear 4pt, 2 order (x-form)
Cubic (not included in Exo's pack?)

And whatever cool or strange these names sound like, I think there could be more focus on sound quality as well...

These interpolation methods differ very much in what they do. Some produce brighter sound, but the sound is dirty (linear, trilinear). Some produce darker, but cleam sound (bspline, optimal), and it's not only a matter of filtering-like difference. And in between there is a problem of mixability, because some of these interpolators seem to produce a sort of "band threshold" effect (probably can be reduced with filters, but filters, you know - they add their stuff as well), which gives the impression of isolation of processed samples.

So my question here. What are better ways of interpolation, that would produce clean and bright sound at the same time?

There is an old thread HERE where I did some experiments with using doubles for interpolation.

Don't expect decent CPU usage though

Thanks, checking on it. I guess it may be problematic anyway, because I'm using 2 mono4 packs right now, but I'm curious where the sound goes compared wit what I've got here.

It's also possible (not tested yet), that mixing sounds coming from 2 different interpolation methods (like trilinear and bspline or opti) may produce desireable output, with possibility to control the level of clarity/dirtiness, and at lower CPU. So I'm very curious what comes out of it.

I was able to convert it briefly to mono/mono4 usage, to hear what it sounds like on channel 1. (there are some node incompatibilities; it works when connected, but if disconnected - can't reconnect again that way).

Well - my first conclusion is - it seems to be rather a matter of yet another method of interpolation to find, than going into doubles (or something did not worked as expected; but it played back the sound, so...).

I'm posting few audio examples to illustrate the problem. Processed by current interpolators here, and source sample with example destination I use (due to good mixability for the type of sounds I deal with).

Hmm... Mixing various interpolations also seems to not bring any specific results.
I doubt that this is done by filters on highs (clarity of higs vs dirt aspect).

It has to be some sort of smart interpolation suited for audio resampling.
BTW, the pitch shift plugin I used for example is (I think) over decade old (or even much older, so it's nothing new.