Lydia Rodriguez-de la Nava is presenting a poster at the 9th International Conference on Digital Libraries for Musicology (DLfM), joined with the annual conference of the International Association of Music Libraries (IAML).
MIDI To Score Automated Drum Transcription
We propose a method for automated drum transcription, which is a special case of music transcription of polyphonic instruments. We implemented this method in qparse, a C++ and python library able to build a whole dataset of music scores from a set of MIDI recorded performances.
Drum notation is an interesting case study for MIDI-to-score polyphonic transcription.
Indeed, although useful for pedagogy and transmission, drum scores are considered as difficult to transcribe manually, due to strong constraints related to playability (with only two drumsticks and two feet), score readability (with large drum kits, drum scores tend to get bloated), high precision requirements, the processing of ornaments (such as “flams”), and errors of MIDI sensors… In return, some transcription subtasks, such as voice separation, are easier in the case of drums than e.g. piano or guitar: In conventional drum score notation, a unique pitch is simply assigned to each element of the drum kit.
