Home » Conference Paper, Publication

MedleyDB: a Multitrack Dataset for Annotation-intensive MIR Research

PDF BIBTEX   16 Januar 2015 1,839 views No Comment
Publication authored by Rachel Bittner and Justin Salamon and Mike Tierney and Matthias Mauch and Chris Cannam and Juan Bello.

We introduce MedleyDB: a dataset of annotated, royalty-free multitrack recordings. The dataset was primarily developed to support research on melody extraction, addressing important shortcomings of existing collections. For each song we provide melody f0 annotations as well as instrument activations for evaluating automatic instrument recognition. The dataset is also useful for research on tasks that require access to the individual tracks of a song such as source separation and automatic mixing. In this paper we provide a detailed description of MedleyDB, including curation, annotation, and musical content. To gain insight into the new challenges presented by the dataset, we run a set of experiments using a state-of-the-art melody extraction algorithm and discuss the results. The dataset is shown to be considerably more challenging than the current test sets used in the MIREX evaluation campaign, thus opening new research avenues in melody extraction research.

Comments are closed.