SAM-Audio (prompt-based separation)

https://github.com/0x0funky/audioghost-ai

More VRAM/performance optimized fork of SAM-Audio by Meta:

https://github.com/facebookresearch/sam-audio
The large model metrics.

"Fish Audio has it on their website but credits are diabolocal"

It’s a prompt-based separator. It can do a decent job when dedicated models fail.
It only works in mono.
Iirc on mid side processed files, it got rid of some noise and knocking when separating SFX and instrumental with a prompt “music” and later with prompt knocking “and selected the part at 0:35 and 0:36 for span prompting (...) You can get better results with more precise span prompting and comp” - nicov_na

“It's not something to throw the entire song at as the single solve all model, but it can absolutely create quality results on parts of the sound that any other model on MVSEP or UVR cannot handle. It's another tool in the toolbox to help create a complete mix (imo).

Just as an example, this crash cymbal had a ton of crowd noise on it at the end of a track. The usual models struggled (top). You can see the break towards the end. The bottom is SAM... or the tail end of SAM anyways. A smooth clean sounding tail off. Resolved the issue vs having the sound just dip out.” - CC Karaoke