SIARCT-CFP: Improving Precision and the Discovery of Inexact Musical Patterns in Point-Set Representations

The geometric approach to intra-opus pattern discovery (in which notes are represented as points in pitch-time space in order to discover repeated patterns within a piece of music) shows promise particularly for polyphonic music, but has attracted some criticism because: (1) the approach extends to a limited number of inexact repetition types only; (2) typically geometric pattern discovery algorithms have poor precision, returning many false positives. This paper describes and evaluates a solution to the inexactness problem where algorithms for pattern discovery and inexact pattern matching are integrated for the first time. Two complementary solutions are proposed and assessed for the precision problem, one involving categorisation (hence reduction) of output patterns, and the second involving a new algorithm that calculates the difference between consecutive point pairs, rather than all point pairs.