Chorale Harmonization with Weighted Finite-state Transducers
Van der Merwe Ab, Jan Buys · SUNScholar (Stellenbosch University) · 2012
We approach the task of harmonizing chorales through style imitation by probabilistically modelling the harmony of music pieces in the framework of weighted finite-state transducers (WFSTs), which have been used successfully for probabilistic models in speech and language processing. The framework makes it possible to place domain-specific regular constraints on generated sequences, and to integrate models of different levels of complexity. We divide the harmonization generation process into different steps, each performed by inference through transducers. We present a method for four-part harmonization that models vertical and horizontal structure in the generated harmonizations. The weights in our transducers are learned by maximum likelihood estimation from a corpus of chorales. The predictive power of own model, as measured through entropy, is competitive with that of existing approaches.