We propose a framework for sonic creativity via computational methods and artificial intelligence research. We extend Norman Sieroka’s theory of sound from a 3-fold to a 4-fold hierarchy so that sound now becomes characterised in terms of the acoustic, physiological, speculative and phenomenal layers. We describe how manipulations in the proposed speculative layer can directly act on how one orients the very apprehension of sound. Black box algorithms, in particular deep neural networks as instantiated by Google’s DeepDream, are then discussed as an illustrative example that can be used to manipulate the speculative layer. We caution that DeepDream offers a warning of how easily our tools capture our phenomenal apprehensions, potentially obfuscating what is just beyond our perception in the process.