Audio textures are a group of sounds analogous to visual textures, that can be described as having stable structure over long term but can vary locally. In this project we hypothesise that we can produce high fidelity sound synthesis of novel audio textures, whilst enabling meaningful human control. The goals of the research are as follows.

• To leverage the expressive capabilities of deep learning to build on sound modelling techniques for audio texture synthesis.
• To provide meaningful and high-level control of deep generative synthesis models via control parameters such as audio descriptors, data landmarks and texture specific parameters.