Definitely Not A Cop said:
It seems to me like you would treat it just like sampling, or is there something I'm not considering?
Sampling involves directly taking music from somewhere else and putting that recording in a different recording. Sampling gets involved specifically with publishing rights as we as performance rights - two separate items you pay for when paying for music.
Cover songs will pay for publishing rights in order for that song to performed somewhere.
AI can take a voice "trained" to be like someone, put them in a brand new song with different lyrics. There is no publishing since it's a totally different song. There is no performance that is ripped off because the performer didn't perform the song.
Totally new area because you have to be careful about real singers singing 'like' someone else. Also - what happens when you train a generic metal voice and every band that has a generic metal voice says you infringed on their voice.
Also - what happens when it's just easier to train AI on a number of great singers that having real singers sing? AI will take a few of those jobs. As a (crappy) musician, I know there is a formula to music - granted it can become complex and have complicated ways of putting it together. But it seems we could be not far away from ChatGPT for lyrics, singer from this process, music from another. Record labels shut down working with new artists????
