Rhythm Extraction from Polyphony Symbolic Music

In this paper, we focus on the rhythmic component of symbolic music similarity, proposing several ways to extract a monophonic rhythmic signature from a symbolic polyphonic score. To go beyond the simple extraction of all time intervals between onsets (noteson extraction), we select notes according to their length (short and long extractions) or their intensities (intensity+/− extractions). Once the rhythm is extracted, we use dynamic programming to compare several sequences. We report results of analysis on the size of rhythm patterns that are specific to a unique piece, as well as experiments on similarity queries (ragtime music and Bach chorale variations). These results show that long and intensity extractions are often good choices for rhythm extraction. Our conclusions are that, even from polyphonic symbolic music, rhythm alone can be enough to identify a piece or to perform pertinent music similarity queries, especially when using wise rhythm extractions.