Visual-guided audio source separation: an empirical study