Thank you. That´s clearer now to understandMeldaProduction wrote: Mon Sep 30, 2019 9:34 amThat's the thing, the machine learning learns the "sound", not meters, so the meters relate to the actual implementation.GNXT wrote: Sun Sep 29, 2019 7:45 pm The problem for me is if I throw in an LA2A emulation, or style of compression and then replace with the MA2A the compression and reduction sounds totally different. (LA2A was jus an example of many that have a ton of compression on default) and I can't get my head around this only a little with the input but still not exactly.
nichtuntun: Not really, in most cases we selected the (according to us) best emulations. As I explained before, this is NOT a simulation in terms of recreating the originals using components, that's nonsense imho. That would be like making Tesla power up steam engine to move the car. Computers are just good at different things. Like machine learning
. So the idea here is to take a generalized model capable of recreating similar response and then let the computer tweak the parameters. It's actually really difficult considering the parameter space, but in this particular case it works great.
Nonetheless, it would be great if all models would be calibrated more to unity gain when changing models. Perhaps in an upcoming update?
Also (very amazingly) great would be an additional function to listen to what the detector of the sidechain sections hears. Like a sidechain-solo feature so to say. I really miss that option.
Nice would also be a FeedForeward and FeeBackward option, especially for the API2500 model.
Cheers
