Modeling and Digitizing Reproducing Piano Rolls

Reproducing piano rolls are among the early music storage mediums, preserving fine details of a piano or organ performance on a continuous roll of paper with holes punched onto them. While early acoustic recordings suffer from poor quality sound, reproducing piano rolls benefit from the fidelity of a live piano for playback, and capture all features of a performance in what amounts to an early digital data format. However, due to limited availability of well maintained playback instruments and the condition of fragile paper, rolls have remained elusive and generally inaccessible for study. This paper proposes methods for modeling and digitizing reproducing piano rolls. Starting with an optical scan, we convert the raw image data into the MIDI file format by applying histogram-based image processing and building computational models of the musical expressions encoded on the rolls. Our evaluations show that MIDI emulations from our computational models are accurate on note level and proximate the musical expressions when compared with original playback recordings.