The most atomistic unit of digital audio is a sample - a single value that in most cases range from -32,000 - 32,000 and are stored at 44,100 values per second. Custom software was written to sort my entire music library, nearly 23 days long, by sample values using the supercomputing facilities at the Holland Computing Center.
The result is a single A440 square wave that follows the volume of my library from the loudest to the quietest. Since this is a one-to-one process that simply rearranges the data from my library, the resulting waveform is a sample-accurate graph and a highly dense, austere sonic drone.
At approximately 87,385,384,800 values, sorting a list of this size required quite a bit of advance planning. The work was done on Merritt, a supercomputer designed for RAM-intensive jobs (64 nodes, 512GB RAM). Below is the basic process by which the piece was made:
- Since a .wav file is essentially just a set header followed by sample data, the header can be stripped from every file and the results concatenated to each other into a giant binary file.
- The sorting of presorted files is faster, so the binary file was first split into more manageable sections and each section was sorted.
- The presorted sections were sorted into each other over several iterations, like interleaving a decks of cards.
- When the entire data set was sorted and recombined, it was split again into one-hour sections (a single 22-day audio file would be unweildly to play.
- New headers for each file were written as simple binary data, allowing the files to be played on any device supporting the .wav format.
Created in collaboration with the Holland Computing Center - thanks to Adam Caprez and Ashu Guru for their assistance with this project. Programming was mostly done in C using the built-in qsort function and memory-mapping to handle the large files.
EVERY SONG I OWN SORTED NUMERICALLY
YEAR
MEDIUM
DURATION
22 days, 22 hours, 25 minutes, 28 seconds