Open kevontheweb opened 2 years ago
Hello! Yes, the "neural" mode currently uses a neural network which is sample-rate dependent. The neural networks was trained on audio that was at a 44.1 kHz sample rate, so it should sound closest to the reference circuit when used at that sample rate. 44.1 kHz was chosen since the plugin was also designed to be used on a Teensy microcontroller which can do audio processing at that sample rate.
If you're curious, I've been re-implementing the Centaur model as part of the BYOD plugin with a slightly different approach, including using a different neural network architecture that is sample-rate agnostic. The "neural" mode wasn't present in the last release, but it is in the latest nightly builds if you'd like to try it out!
It seems that some filters are dependent on the sample rate as when I run my project at higher sample rates the plugins becomes much brighter as the sample rate increases. This only seems to be apparent in neural mode to me. I personally like the brighter sounds, but it does have the problem of sounding different when a project is loaded with a different sample rate, and maybe having some built in oversampling would help?