Proceeding of the past Conferences /

New Music Concepts 2020

Detection of Local Boundaries of Music Scores with BLSTM by using Algorithmically Generated Labeled Training Data of GTTM Rules

Part of the New Music Concepts book series (NMC volume 7)
ISBN: 978-88-944350-3-0

Author(s): You-Cheng Xiao, Alvin Wen-Yu Su

Abstract: Preparing a considerable amount of labeled data is always a problem that machine learning related research must deal with. In this paper, we propose a procedure which algorithmically generates a large amount of labeled music scores as training data for local boundary detection problem of music scores. The local boundary rules in the generative theory of tonal music (GTTM) are used as an example, which uses the proposed procedure to generate a dataset and trains bidirectional long short-term memory (BLSTM) networks to detect these rules. The experimental results show that the BLSTM models trained by the algorith- mically generated dataset does outperformed the existing ATTA models. It greatly reduces the effort in designing new rules and the corresponding detection models without time consuming manual labeling.

Keywords: Symbolic Music Processing, Music Boundary Detection, A Genera- tive Theory of Tonal Music, Bidirectional Long Short-Term Memory Net