Variable delay buffer (vdb) - improving on the crossfading algorithm
Hi there,
I am dealing with delays at the moment and one of the requirements is to be able to change delay times dynamically without pitch shifting or doppler artifacts.
I have read every thread on the subject and used most of the abstractions that implement a vdb (Ejies, Bennies, etc...) Now in my patches I am using the [M4L.vdelay~] object pretty regularly. There are many versions of the basic algorithm and they all manage to suppress pitch shifting artifact *fairly* well, however - no matter the size of the crossfading window - when I change delay time dynamically I can still hear glitches in the resulting audio. This happens especially with tones that have a pretty pure harmonic content such as flutes and bells.
I am wondering if it is possible to improve on the typical crossfading algorithm as implemented in the [M4L.vdelay~] abstraction for example.
This 10-year-old thread makes me think there is hope:
https://cycling74.com/forums/seemless-non-pitch-shifting-variable-delay
There Roman Thilenius said:
"you should use 3 and not only 2 parallel processes for best results."
Roman, could you elaborate on that, please?
Does anybody know about a more polished/effective strategy to dynamically change delay time without introducing pitch shifting artifacts? Any patches to show?
Thanks a million for any suggestion/advice.
- Luigi
the more windows you make, the longer the fade in/outs will be in relation to the playing time.
3 is better than 2 because you dont have to care about sample accurate windowing for that moment where one of the windows is at 0. gain, so it helps to avoid clicks created by minor errors in your code.
but more important is that you should create windows with a variable in and out time - or better yet , of a variable size. then you can adapt the windowing style and size to the audio material while you use the effect.
when windowing stuff you either have artefacts or smearing. for drumloops you would rather have artefacts, for pads and bells you would be better off with "polyphony" moments.
i mostly use instances of wave~ for windowing jobs and offset their phases.
always use fast scopes to visually see what you do.