CITATION — REFERENCE ENTRY
DeepBach: A steerable model for Bach chorales generation — Proceedings of the 34th International Conference on Machine Learning
- Key
- hadjeres2017-deepbach
- Authors
- Hadjeres, Gaëtan; Pachet, François; Nielsen, Frank
- Issued
- 2017
- Type
- paper-conference
- Container
- Proceedings of the 34th International Conference on Machine Learning
- Volume
- 70
Raw CSL JSON
{
"URL": "https://arxiv.org/abs/1612.01010",
"type": "paper-conference",
"title": "DeepBach: A steerable model for Bach chorales generation",
"author": [
{
"given": "Gaëtan",
"family": "Hadjeres"
},
{
"given": "François",
"family": "Pachet"
},
{
"given": "Frank",
"family": "Nielsen"
}
],
"issued": {
"date-parts": [
[
2017
]
]
},
"volume": "70",
"event-place": "Sydney, Australia",
"container-title": "Proceedings of the 34th International Conference on Machine Learning",
"collection-title": "Proceedings of Machine Learning Research"
}
Claims
-
Bach's 389 chorale harmonizations, in which he harmonized well-known Lutheran hymn melodies for four voices, are a major benchmark for automatic composition. Ebcioğlu's 1988 rule-based system had more than 300 rules, and its author acknowledged that its results did not sound like Bach except for occasional patterns. Hild et al. (1992) used several neural networks for successive subtasks. DeepBach, trained on Bach's chorales, uses pseudo-Gibbs sampling rather than composing sequentially, lets users impose constraints such as fixed notes, rhythms, or cadences, and has a MuseScore plugin; 1,272 people took part in a 'Bach or computer' listening test.
"The corpus of the chorale harmonizations by Johann Sebastian Bach is remarkable by its homogeneity and its size (389 chorales [...]). [...] the composer takes a well-known (at that time) melody from a Lutheran hymn and harmonizes it [...]. It is a rule-based expert system which contains no less than 300 rules [...]. The drawbacks of this method are, as stated by the author, the considerable effort to generate the rule base and the fact that the harmonizations produced "do not sound like Bach, except for occasional Bachian patterns and cadence formulas." [...] A neural-network-based solution was later developed by (Hild et al., 1992). This method relies on several neural networks, each one trained for solving a specific task [...]. DeepBach's strength comes from the use of pseudo-Gibbs sampling [...]. This is in contrast with many automatic music composition approaches which tend to compose music sequentially. Our model is also steerable in the sense that a user can constrain the generation by imposing positional constraints such as notes, rhythms or cadences [...]. We also provide a plugin on top of the MuseScore music editor [...]. For this experiment, 1272 people took this test"
Available in