École Nationale Supérieure de
Techniques Avancées

Abstract

Automatic Music Transcription (AMT) is a problem in Music Information Retrieval (MIR), first explored in 1976. It consists of the conversion of a performance into sheet music using some form of music notation. The particular interest in this task stems from the fact that sheet music is considered to be the aptest form for the preservation of unnotated music in the cultural heritage and a vector of exchange of information within music communities in terms of analysis and interpretation. This has motivated research in this area in an attempt to solve the AMT problem by subdividing it into specific subtasks with regard to each process that needs to be applied. In this work, we consider one of these subtasks, and we focus on optimising the parameters used to transcribe performances in a symbolic format through parsing techniques. The method is tested on a monophonic dataset via Qparse, a tool developed by researchers at INRIA Paris and CNAM.