dont go to keyboard view ptweb does not like it lol
collab actually handles it really well if you want to see

1 voice (sine)
50 / 50 units
494135 / 500000 events

original: Frank C. Stanley's 1910 Indestructible Record performance of "Auld Lang Syne"
https://commons.wikimedia.org/wiki/File:Auld_Lang_Syne.ogg

general idea of how this was generated:
first, chunk input audio into batches, run FFT on each chunk + save the n highest amplitude frequencies for each chunk
then, go through each chunk and assign each unit to one of the frequencies for that chunk, preferring frequencies close to the current note (key)
there's some quantization & thresholding stuff going on too to optimize event count so the quality can be turned up as much as possible
this also utilizes the full precision of key/volume events which isn't normally accessible in collage