A music audio embedding model trained on Discogs data for genre and style classification using 30-second clips.