Extension of knowledge-driven harmonization model for tonal music

The paper presents an extended approach to the automatic harmonization of musical work, based largely on the knowledge of music theory. Automatic harmonization could be seen as a processing of a document containing formal music notation and producing specific meta-data (harmonic functions). A human expert is needed to evaluate the various harmonizations, as their ´correctness´ cannot be determined in algorithmic way. Authors´ approach can be considered as knowledge-based expert system, being in contrast to a data-driven approach, that extract relations from examples. Authors´ approach emphasizes universality - understood as a possibility of direct model modifications - in order to obtain varied harmony characteristics (e.g. a complicated and unusual harmony, or a simple harmony using only a small subset of harmonic functions and few modifiers). Therefore the model is configurable by changing the internal parameters of harmonization mechanisms, as well as importance weights corresponding to each of these mechanisms. The mechanisms are among others: harmonic functions excitement with notes´ pitches, note importance regarding horizontal and vertical placement in measure, succession of harmonic functions. Authors also describe indirect model tuning with a preselected set of examples.