I introduced it to my friend (who is not technical, and has a RTX 3080), his experience was: he selected a 3:32 song, it took 10-15 minutes trying to scan it. In the end it did a "decent job of removing lyrics from the audio and providing accurate ones" "It seemed to be tracking my singing but there is no mic setup or anything like it" "And I think I accidentally wiped the AI model data it took forever to generate with a single click"

In the end he's gone back to karaoke videos on youtube but it seems promising.