No idea if this would be good for audio but I've always found this paper interesting and not widely distributed: