LP-44: Song Describer: a Platform for Collecting Textual Descriptions of Music Recordings

Ilaria Manco, Benno Weck, Philip Tovstogan, Minz Won, Dmitry Bogdanov

Abstract: We present Song Describer, an open-source data annotation platform for crowdsourcing textual descriptions of music recordings. Through this tool, we propose to collect annotations with the goal of creating the first public dataset of audio-caption pairs in the music domain. We believe that such a dataset will be useful in supporting the growing interest in the integration of natural language processing within music information retrieval systems. In this paper, we describe our approach to designing Song Describer, outline the data collection protocol, and illustrate the main steps involved in using the platform.