Hierarchical Classification for Instrument Activity Detection in Orchestral Music Recordings

Instrument activity detection is a fundamental task in music information retrieval, serving as basis for many applications, such recommendation, tagging, or remixing. Most published works on this cover popular and smaller ensembles. In paper, we embrace orchestral opera recordings rarely considered scenario automated instrument detection. Orchestral particularly challenging since it consists of intricate polyphonic polytimbral sound mixtures where multiple instruments are playing simultaneously. can naturally be arranged hierarchical taxonomies, according to families. As the main contribution show that classification approach used detect our scenario, even if only few fine-grained, instrument-level annotations available. We further consider additional loss terms improving consistency predictions. For experiments, collect dataset containing 14 hours with aligned annotations. Finally, perform an analysis behavior proposed regard potential confounding errors.