Singing is going to be much harder.There’s so much variation. Plus words just requires it to sound natural, singing requires much more of a performance and we have all kinds of other aspects. Like singing softly, loudly, vibrato, portamento, rhythm, not it mention notes themselves,

This might make it clear. We can do synthesized percussion much better than we can do synthesized tonal instruments like violins, flutes etc. sampling percussion has always been the easiest thing to get realistic and 100% synthesized instruments are no different.

If you want to sample percussion all you really need aside from recording quality is sampling multiple repetitions and a shit-ton of dynamic layers. The best percussion sample libraries today will have like maybe 10-20 dynamic layers and 5+ to 10+ repetition samples sometimes. You don’t even need that many to make it sound convincing. But with instruments like vocals, violins flutes etc that’s not scratching the surface. These are complex on much higher dimensions and you need completely different techniques to capture them, and even then it’s still not quite right or it’s highly limited in it’s use