We get them from our heads with a microphone in front of our mouth.
If you get anything without permission, you face copyright issues. There is a site called looperman (https://www.looperman.com) where people make and post acapellas based on genre. Everything is royalty free. So you can use those as long as you credit them as the vocal artist.
haha, but ya, dont rip anything from any artist though without permission. If you do, it better be super short. AND your remix or edit should be drastically different than the original.
Getting acapellas is a lot more difficult than it used to be... There was a time when a lot of artists would release them as tracks on their vinyl pressings - no longer the case.
I have seen talk of people matching tracks with vocals against the instrumental versions of the same tracks and then inverting them both in order to extract the vocal - but reports of success and failure were on par.
Your best bet is to either record your own vocal tracks or hunt for classic 12" (1990's) singles that have acapella tracks on them.
Imo, ppl sometimes focus too much on copying instead of bringing original material to the plate.
Sampling is something that hiphop started out with though. Nowadays, sampling for professional use is a *****! Too much copyrights and whatnot. Getting permissions or even getting in contact with artists is sometimes impossible.
Copyleft material and creative commons licensing is something of nowadays that I like a lot.
For non professional use, everything still goes, as long as you don't put it online for the public.
As a final thought, ppl isolate themselves. There's probably tons of ppl that can properly sing right in your neighborhood or social circle.
Spectralayers is surprisingly easy - at least the way I used it to separate the vocal stem:
1. Import the audio track you want to extract the vocals from
2. Insert Spectralayers as an insert
3. A window opens the bottom of the screen
4. On the right it lists the layers that it's detected and asks which you want to hear. The included version only separates vocals, not other layers, so the only 2 selections are 'Vocals' and 'The Rest' (or something). So you just put a check against vocals.
5. Render the resulting track so that only the vocals are preserved.
Oh, and there's a sensitivity control somewhere that you can tweak to help it judge vocals vs rest