Identifying polyphonic patterns from audio recordings using music segmentation techniques

This paper presents a method for discovering patterns of note collections that repeatedly occur in a piece of music. We assume occurrences of these patterns must appear at least twice across a musical work and that they may contain slight differences in harmony, timbre, or rhythm. We describe an algorithm that makes use of techniques from the music information retrieval task of music segmentation, which exploits repetitive features in order to automatically identify polyphonic musical patterns from audio recordings. The novel algorithm is assessed using the recently published JKU Patterns Development Dataset, and we show how it obtains state-of-the-art results employing the standard evaluation metrics.