The old standby of those who create: ITS ALL A HACK.
Spectralayers sounded interesting, but I came across something brilliant while looking up alternatives and similar options. This article: http://www.howtogeek.com/61250/how-to-isolate-and-save-vocals-from-music-tracks-using-audacity/ suggests that to isolate vocals from music you need the original vocal-less soundtrack, which you then basically invert and superimpose over the version of the song with vocals. Here, what is not in the instrumental, is by definition the vocals, and as long as the two instrumental tracks are indeed the same track, apparently with some tweaking Audacity can deduce a new audio track (the vocals) from that difference.
In songs, this technique would rarely be useful, as most aren’t released in a vocal-less version (although maybe artists officially distribute a karaoke version, idk) but in the case of movies, any movie that releases its soundtrack (which is common) should be able to have the in-film dialogue isolated/extracted, at least in theory. I wonder what the quality would be though. If you’re mixing music back into the finished audio, any distortion might at least be masked by that.
I’m also guessing scenes where explosions and lasers and whatever are present in addition to the music would probably pollute the ideal vocal result. But again, we’re kind of talking about hacks in the first place, so maybe that is tolerable.