How to read it. Every column is a 43 ms slice of the song; every row is a band of frequency on a log scale, the way hearing works. The Separated view is built by running the same analysis on each stem AudioShake separated from the mix, then painting each pixel with whichever stem carries the most energy there. That assignment, pixel by pixel, is the job a source separation model does.

Audio: “Future” by Torches · stems separated with AudioShakeaudioshake.ai