BaShoF: Polyphonic Pattern-Matching Dataset

BaShoF is a symbolic dataset of query subjects designed for musical pattern matching, created from the first 12 fugues of J. S. Bach's The Well-Tempered Clavier, Book I, and the first 12 fugues of D. Shostakovich's Op. 87. The dataset includes the 24 original MIDI subjects and a controlled layer of noisy subjects generated from the original ones, with five noise levels, ranging from 10% to 50%, and ten seeds per level. This produced 1200 augmented subjects with added outlier notes and 1200 corrupted subjects with pitch alterations. For each complete MIDI fugue, a CSV file defining a set of overlapping windows is provided together with a window-level ground truth, based on musicological sources, that contains the indices of the windows where subjects occur. In total, the dataset provides 393 positive ground-truth windows distributed across 24 fugues. All the Python scripts used for window generation and noise addition are included. This dataset can be valuable for testing the accuracy and noise tolerance of pattern-matching systems in polyphonic music.