Four-Part Harmonization: Comparison of a Bayesian Network and a Recurrent Neural Network

Tatsuro Yamada, Tetsuro Kitahara, Hiroaki Arie, Tetsuya Ogata

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we compare four-part harmonization produced using two different machine learning models: a Bayesian network (BN) and a recurrent neural network (RNN). Four-part harmonization is widely known as a fundamental problem in harmonization, and various methods, especially based on probabilistic models such as a hidden Markov model, a weighted finite-state transducer, and a BN, have been proposed. Recently, a method using an RNN has also been proposed. In this paper, we conducted an experiment on four-part harmonization using the same data with both a BN and RNN and investigated the differences in the results between the models. The results show that these models have different tendencies. For example, the BN’s harmonies have less dissonance but especially the bass melodies are monotonous, while the RNN’s harmonies have more dissonance but especially bass melodies are smoother.

Original languageEnglish
Title of host publicationMusic Technology with Swing - 13th International Symposium, CMMR 2017, Revised Selected Papers
EditorsMatthew E.P. Davies, Mitsuko Aramaki, Richard Kronland-Martinet, Sølvi Ystad
PublisherSpringer Verlag
Pages213-225
Number of pages13
ISBN (Print)9783030016913
DOIs
Publication statusPublished - 2018 Jan 1
Event13th international Symposium on Computer Music Multidisciplinary Research, CMMR 2017 - Matosinhos, Portugal
Duration: 2017 Sep 252017 Sep 28

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11265 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other13th international Symposium on Computer Music Multidisciplinary Research, CMMR 2017
CountryPortugal
CityMatosinhos
Period17/9/2517/9/28

Keywords

  • Bayesian network
  • Harmonization
  • LSTM-RNN
  • Machine composition

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'Four-Part Harmonization: Comparison of a Bayesian Network and a Recurrent Neural Network'. Together they form a unique fingerprint.

  • Cite this

    Yamada, T., Kitahara, T., Arie, H., & Ogata, T. (2018). Four-Part Harmonization: Comparison of a Bayesian Network and a Recurrent Neural Network. In M. E. P. Davies, M. Aramaki, R. Kronland-Martinet, & S. Ystad (Eds.), Music Technology with Swing - 13th International Symposium, CMMR 2017, Revised Selected Papers (pp. 213-225). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11265 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-01692-0_15