Structural Harmony Method in the Context of Deep Learning on Example of Music by Valentyn Sylvestrov and Philipp Glass

Anna Shvets

doi:10.14236/ewic/eva2019.60

Abstract

The article shows the potential of structural harmony method application in deep learning models training. Harmonic sequences, being represented as the paths in the system of graphs and then schematised, serve as the source of the deep learning models input. The visual character of the dataset generated using schemes allows the application of supervised machine learning technics and is suitable for time series analysis. Therefore, two neural network architectures – Convolutional and Convolutional Long-Short Term Memory were tested on example of music by postmodern composers Valentyn Sylvestrov (represented by vocal cycle “Silent songs”) and Philipp Glass (represented by piano cycle “Metamorphosis”).

Highlights

The development of training methods opened great possibilities in big data analysis, which influences the domains of knowledge, traditionally based on qualitative analysis, introducing quantitative analytical base
The core of the western music development resides in the tonality formation, its extension and dissemination, basing on functional relations and alterations within a given tonality, and the modulation strategies between tonalities: the establishment of the functional relationship began with Palestrina compositions and the decomposition of tonality manifested in the elliptic chains in the late Wagnerian, and more broadly – late romantic music
Structural Harmony Method in the Context of Deep Learning on Example of Music by Valentyn Sylvestrov and Philipp Glass Anna Shvets composer has been shown to the model, which resulted in a very good learning rate: 92% for both test sets

Summary

INTRODUCTION

The development of training methods opened great possibilities in big data analysis, which influences the domains of knowledge, traditionally based on qualitative analysis, introducing quantitative analytical base. The selection of features being analysed, should be performed in a careful manner, as the selection of very few features (such as one musical interval) may be misleading in considering such large amount of data as the history of music styles (Nakamura & Kaneko 2018). The core of the western music development resides in the tonality formation, its extension and dissemination, basing on functional relations and alterations within a given tonality, and the modulation strategies between tonalities: the establishment of the functional relationship began with Palestrina compositions and the decomposition of tonality manifested in the elliptic chains in the late Wagnerian, and more broadly – late romantic music. The analysis of the functional relations within tonality should be considered as the most relevant for the research in tonal music

STRUCTURAL HARMONY METHOD

DATA DESCRIPTION

CNN-LSTM NEURAL NETWORK MODEL CASE

Findings

SUMMARY