mshoxxDB - a Versioned Dataset for Electronic Music

Michael Taenzer (Fraunhofer IDMT / UPF)*

This paper will be presented in person

Abstract:

Deep learning in Music Information Retrieval (MIR) relies on diverse datasets to cover various musical genres and instrumentation, ensuring robust model training and performance. The introduction of the mshoxxDB dataset aims to address the gap in low-resource data for electronic music, featuring 18 full-length songs and offering per-track MIDI files, multi-track files, mixture files, and extensive metadata for various MIR applications. This dataset spans multiple electronic music sub-genres and provides a resource for advancing research in tasks such as Automatic Music Transcription, Multi-Pitch Estimation, and Source Separation, fostering innovation in the field.