Modeling Baroque Two-Part Counterpoint with Neural Machine Translation
This is an incremental approach for music generation researchers, applying existing NMT methods to a new domain with limited success in capturing complex musical constraints.
The paper tackled generating Baroque two-part counterpoint by framing it as a neural machine translation problem, training an attention-based model on a custom dataset, and found it produced some idiomatic features like imitation but failed to learn strict stylistic rules such as avoiding parallel fifths.
We propose a system for contrapuntal music generation based on a Neural Machine Translation (NMT) paradigm. We consider Baroque counterpoint and are interested in modeling the interaction between any two given parts as a mapping between a given source material and an appropriate target material. Like in translation, the former imposes some constraints on the latter, but doesn't define it completely. We collate and edit a bespoke dataset of Baroque pieces, use it to train an attention-based neural network model, and evaluate the generated output via BLEU score and musicological analysis. We show that our model is able to respond with some idiomatic trademarks, such as imitation and appropriate rhythmic offset, although it falls short of having learned stylistically correct contrapuntal motion (e.g., avoidance of parallel fifths) or stricter imitative rules, such as canon.