Toward a Sparse Interpretable Audio Codec

Apologies if it isn't clear, but the animated gifs are meant as an illustration of the iterative encoding process, where the encoder decomposes the signal step-by-step, as in matching pursuit. I'll be sure to clarify that point.

I'll add a paragraph on compression rates/ratios, although that isn't necessarily the main focus here; codecs may compress a signal, but they might also transform it into a more useful, easy-to-understand and easy-to-manipulate representation.