In a way, this sounds similar to someone using a DiffSinger or ENUNU voicebank to make music, but without having to write all the lyrics word/syllable for word/syllable on individual notes and change the notes they’re on individually because you pitched them wrong. Also without the backing track too.
Either way, interesting enough that you can even do this kinda stuff. Especially since any time I’ve played around with that kind of AI, it’s always never turned out even remotely halfway decent in any way.
It can definitely work if you’ve trained it right. It’s how I discovered many of my now favorite vocaloid/utau/deepvocal/whatever songs, amongst other groups outside of that vocal synth genre. It definitely isn’t doing me super dirty in the music department.