Multi-resolution sound texture synthesis using the Dual-Tree Complex Wavelet Transform

This paper presents a novel algorithm for Sound Texture Synthesis. Inspired by the well known Efros and Leung non-parametric 2-D Image Texture Synthesis algorithm, our 1-D interpretation is used to synthesise long, perceptually and statistically similar sound textures from much shorter real-world audio training examples including crowd noise, a baby crying and speech. The process employs the Dual-Tree Complex Wavelet Transform to reduce computational complexity without sacrificing spectral coherency in the synthesised audio. Our approach produces plausible and interesting sound textures that are comparable to the results of other state-of-the-art algorithms.