This one's a weird one. It also uses a Magenta music model called Music VAE.
One of the things this model can do is encode note sequences in a lower dimension space, and then decode
points from that space into a new note sequence. Normally, if you encode a note sequence and then decode that
encoding, you get the original one. Here, I am encoding the drum beat that represents the sheriff (see day 16 for how that works), adding
a little bit of noise, and then decoding the noisy encoding. What you get is...this.