Sony Music recently released a new recording (made in 2006) of Glenn Gould performing the Goldberg Variations. Curious, not? The recording was made using measurements of the old recordings and then regenerating the performance on a computer-controlled grand piano, a modern pianola.
This technology dates from the early nineties, a time when several piano companies (including Yamaha and Bösendorfer) combined MIDI and modern solenoid technology with the older idea of a pianola. Old paper piano rolls with recordings of Rachmaninoff, Bartok, Stravinsky and others were translated to MIDI and could be reproduced ‘live’ on modern instruments like the Yamaha Disklavier. Until now, the only left challenge was to be able to do this for recordings of which no piano-rolls were available.
Besides the technicalties of all this, for most people the real surprise —or perhaps disillusion— might well be the realization that a piano performance can be reduced to the ‘when’, ‘what’ and ‘how fast’ the piano keys are pressed. Three numbers per note can fully capture a piano performance, and it allows for replicating any performance on a grand piano(-la). The moment a pianist hits the key with a certain velocity, the hammer releases, and any gesture that is made after that can be considered merely dramatic: it will have no effect on the sound. This realization puts all theories about the magic of touché in a different perspective.
Nevertheless, while it is relatively easy to make the translation from audio (say a recording from Glenn Gould from 1955) to the what (which notes), and the when (timing) in a MIDI-like representation, the problem is in the ‘reverse engineering’ of key velocity. What was the speed of Gould’s finger presses on the specific piano he used? The Zenph Studios claim to have solved it for at least a few recordings. Only trust your ears :-)