I found this online.
This guy used data from a voice recording of a person speaking to figure out which combination of piano keys (i.e. complex “chords”) would best reproduce each point in the wave form of the speech. Generally these are too many keys, needing to be pressed too rapidly in sequence, for a human pianist to do this. So he used a mechanical piano-playing device to reproduce the speech. It’s just on the edge of comprehensibility. Quite eerie.
