Towards supervised music structure annotation: a case-based fusion approach

TitleTowards supervised music structure annotation: a case-based fusion approach
Publication TypeMaster Thesis
Year of Publication2014
AuthorsHerrero, G.
AbstractAnalyzing the structure of a musical piece is a well-known task in any music theory or musicological field. However, in recent years, trying to find a way of performing such task in an automated manner has experienced a considerable increase in interest within the music information retrieval (MIR) field. Nonetheless, up to this day, the task of automatically segmenting and analyzing such structures remains an open challenge, with results that are still far from human performance. This thesis presents a novel approach to the task of automatic segmentation and annotation of musical structure by introducing a supervised approach that can take advantage of the information about the music structure of previously annotated pieces. The approach is tested over three different datasets with varying degrees of success. We show how a supervised approach has the potential to outperform state-of-the-art algorithms assuming a large and varied enough dataset is used. The approach is evaluated by computing standard evaluation metrics in order to compare the obtained results with other approaches. Several case studies that are considered relevant are as well presented, along with future implications.
intranet