Analogous to our work in representing video as a set of objects and a scripting language, we are exploring methods for recording audio such that the result is a set of localized sound sources and an acoustical model for the space in which they are placed. Techniques involved include source separation and blind deconvolution.