Does anyone know of a model that can classify an audio segment as either containing music or not, regardless of whether lyrics are present? (with a paper, not just some random model on HF)