Following a Musical Performance from a Partially Specified Score

This paper describes a system that maps a musical performance, recorded as MIDI, onto a partially specified score (lead sheet). Our system breaks the performance into appropriate segments for hypothesized chords by representing it as a directed acyclic graph (DAG) whose edges represent time-intervals of the music. The highest-reward path through the DAG corresponds to the best segmentation of the performance. During segmentation, the name of the best matching chord for each segment label is given to the segment. The sequence of chord names generated by the performance analysis is then aligned against the sequence of chord names in the lead sheet using an alignment technique drawn from gene-sequence analysis. Scores for the alignment are determined by estimating the probability of a random match between chord sequences compared to the probability of the match. Probabilities are calculated through the analysis of chord frequencies on a corpus of tonal music drawn from a standard music theory text.