Work with separation non-voice sounds

[Exxi80]

Hello,
We need to separate one sound(not voice) from others.
As I know, the RTX Voice source code had the option to train the user's own model but it's not accessible now.
Is it possible it your software to train my own sound model?
For example, I play piano In a public place and I want to separate only piano sounds and cut all the sounds.

Or maybe you can help to create this software?

Thank you.