In this thesis we focus on the discovery of repeated patterns in music, one of the tasks in the research field of music information retrieval. We use a compositional hierarchical model, which was previously successfully applied to other tasks from this field, and adjust it to process symbolic music and find repeated patterns in musical pieces. In order to facilitate testing and interpretation of algorithm's output we develop a visual representation of the model. Results are evaluated on two datasets and compared to results of other researchers. Comparison shows that the proposed model is comparable to or better than other approaches in finding multiple repetitions of one pattern, but lags in identifying at least one occurrence of each structurally salient pattern. We propose several possible solutions for increasing the share of successfully identified salient patterns and suggest some guidelines for further development of the model.
|
