@Robin
Sitting here trying to remove noise from old dusty music cassettes. Whether I remove noise before or after separation I have the same problem, weak sounds and reverb tails disappear or become choppy.
Is it technologically possible to train the AI separation to perceive noise, so that noise can become its own layer?
Have you tried Unmix Levels and played around with pixelsizes and the threshold value?
I get pretty decent results with thresholds that only put a small portion of the noise in the low level layer.
When it comes to vertical/horizontal pixelsize I actually asked ChatGPT how to use them and I got a very good and valid answer. You can try that if you haven’t.
Thanks for the good answer…
Yes, I’ve tried that, but many details get lost in the noise.
I think that if the AI separator is also trained on different noise types and generates a separate output/Layer (– like each of the instruments) for hiss/noise, it could be revolutionary.
I’ve asked ChatGPT if it’s possible and Chat says it’s completely possible…
Please vote