Timbre Interfaces using Adjectives and Adverbs

How can we provide interfaces to synthesis algorithms that will allow us to manipulate timbre directly, using the same timbre-words that are used by human musicians to communicate about timbre? This paper describes ongoing work that uses machine learning methods (principally genetic algorithms and neural networks) to learn (1) to recognise timbral characteristics of sound and (2) to adjust timbral characteristics of existing synthesized sounds