This sounds like a dirty mellotron.

Welcome to Soundiron’s Phonautograms library!

This vocal sample-based “instrument” was created by cutting, editing and programming a selection of the earliest known sound recordings, as restored by the non-profit group First Sounds. This original method of recording, called ‘Phonautography’ was invented in the early 1850s, by French inventor Édouard-Léon Scott de Martinville. The sounds were captured by projecting the voice and other sounds into a cylindrical horn attached to a stylus, which in transfered the vibration into lines over the surface of oil lamp soot-blackened sheets of paper.

They had been preserved by the archives of the French Academy of Sciences (Académie des sciences de l’Institut de France). One of First Sounds’ founders, David Giovannoni, tells us that “the Academy is the institution whose stone walls have preserved these magnificent documents through social upheavals, political revolutions, world wars, and other disturbances since Scott’s deposit of Gamme de la Voix in 1861 — just five years short of the bicentennial of the Academy’s founding in 1666.”

Essentially, these recordings were purely optical and no device existed which could translate the recorded acoustic information back into sound, until the First Sounds organization acquired the preserved papers in France and brought them to scientists at Lawrence Berkeley National Laboratory in Berkeley, who devised a method of scanning and deciphering the images.