Evolutionary feature synthesis for content-based audio retrieval