Discovering translational patterns in symbolic representations of music (22/07/10)
Suppose that you wish to get to know a particular piece of music, and that you have a copy of the score of the piece and a recording or a MIDI file. Typically, to become familiar with a piece, one studies/plays through the score and listens, gaining an appreciation of where and how material is reused. The literature on music information retrieval (MIR) contains several algorithmic approaches to this task, referred to as 'intra-opus' pattern discovery. Given a piece of music in a symbolic representation, the aim is to define and evaluate an algorithm that returns patterns occurring within the piece. Some potential applications for such an algorithm are: (1) a pattern discovery tool to aid music students; (2) comparing an algorithm's discoveries with those of a music expert as a means of investigating human perception of music; (3) stylistic composition (the process of writing in the style of another composer or period) assisted by using the patterns/ structure returned by a pattern discovery algorithm. The presentation will address two ways in which my research has improved upon current pattern discovery algorithms, drawing on linear regression techniques and finite-set geometry.