Finding structure and repetitions in a musical signal is crucial to enable interactive browsing into large databases of music files. Notably, it is useful to produce short summaries of musical pieces, or "audio thumbnails". In this paper, we propose an algorithm to find repeating patterns in an acoustic musical signal. We first segment the signal into a meaningful succession of timbres. This gives a reduced string representation of the music, the texture score, which doesn't encode any pitch information. We then look for patterns in this representation, using two techniques from image processing: Kernel Convolution and Hough Transform. The resulting patterns are relevant to musical structure, which shows that pitch is not the only useful representation for the structural analysis of polyphonic music.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.