Pioneer composers always try to explore beyond frontiers. Nowadays, they are allowed to listen, read, and play music with machines. Music information retrieval (MIR) technology has brought revolutionary computational music analysis (CMA) in recent years. It provides novel ways to dissect music. On the other hand, algorithmic composition (AC) is able to generate countless pieces with explicit ground truth for MIR experiments. While most data scientists have been seeking the best features which are perfect for discrimination between samples in dissimilar classes, little attention was paid to unclaimed territories in any dimensional feature space. If somewhere there is a tiny ratio of outliers from diverse classes, the unexploited parts might be worth to explore.
To investigate, specific intercultural and cross-level features are implemented to extract from 18009 symbolic and audio samples in 91 sub-datasets which come from various genres in five eras across three continents. Next, the indices of tessitura and mobility are also extracted from all symbolic samples. Besides, 32 jSymbolic features from fewer samples are selected to realize the (non-)concurrence features based on internal correlations between local base features. Final results confirm that vocal styles broadly have larger susceptibilities and narrower register widths than instrumental styles in average. Distributions reveal the unexploited areas in two-dimensional space. Evaluations illustrate (non-)concurrence features’ performance and improvement on the balanced composer classification of Haydn and Mozart’s separated string quartet parts. Manual compositions demonstrate several successes to penetrate the frontiers.
Although modern artificial neural network may automatically learn features to detect unexploited compositional space, those features are not necessary perceivable by human or even not tractable through AC. Contemporary composers have to, however, select controllable features and devise appropriate algorithms and parameters unless they do not want to compose by their own. This research initiates innovative praxis. Its ultimate goal is to evolve toward the mutualism. Composers learn from the dimensional feature space and distributions through MIR and CMA with intent to devise better algorithms and parameters to manipulate in AC. Then AC has the capability to promote better techniques and features for MIR and CMA, which again stimulate composer’s imagination and creativity.
