I'm interested in sparse, interpretable and easy-to-manipulate representations of audio.