ClimaCS: A Multi-Genre Dataset of Timed Comments for Music Highlight Detection

Résumé fourni par la source

Recent advances in deep classifier models achieved a remarkable audience experience not only for describing music but also for generating creative outcomes. Motivated by perspectives in automatic music and video synchronization, we consider the challenge of identifying the most salient moments in a music track. To train a model on this task, an appropriate dataset becomes essential. We introduce ClimaCS, an timed comment dataset designed for training models for highlight detection. We describe our methodology to bring together this set, which features over 2,000 tracks from a wide range of music genres and additional metadata. We expose the main characteristics of the content of the dataset and suggest an initial benchmark experiment. Using a simple CNN-based architecture, we demonstrate that highlights can be predicted using exclusively timed comment for training, without expert manual annotation of the highlights.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.