MusicBLAST - Gapped Sequence Alignment for MIR

We propose an algorithm, MusicBLAST, for approximate pattern search/matching on symbolic musical data. MusicBLAST is based on the BLAST algorithm, one of the most commonly used algorithms for similarity search on biological sequence data [1, 2]. MusicBLAST can be used in combination with an arbitrary similarity measure (e.g., melodic, rhythmic or combined) and retrieves multiple occurrences of a given search pattern and its variations. Different from many other pattern matching techniques, it can find incomplete and imperfect occurrences of a given pattern, and produces a significance measure for the accuracy and quality of its results. Like BLAST — and different from many musical pattern matching approaches — MusicBLAST retrieves heuristically optimised bi-directional alignments searching iteratively in forward and backward direction by starting at a dedicated seed note position of a performance.