Which is which (open after you pick)
A = AI, library music (Epidemic tracks). B = the published version (Subash). C = AI, music composed by AI (Eleven Music, a test only: its license does not cover a team channel yet). A and C share the same sound effects.
In A and C your voice was separated from the published mix with a stem splitter, so it can carry slight artifacts. Round 1 was planned blind from the frames and transcript; later rounds added accents where B peaks (the dish cut, the spikes, the glide), an ear model that heard the first pen-stroke sound as "a metal scrape", and a cap that keeps every effect's peak under your voice.
Loudness over time, all three measured the same way (top: published; middle: A; bottom: C). Red = voice, purple = music + sound effects.
