Audio Texture Synthesis with Scattering Moments
This work addresses audio texture synthesis for applications in sound design and multimedia, presenting an incremental improvement over prior techniques.
The paper tackles audio texture synthesis by using scattering moments to represent stationary processes, achieving good synthesis quality with significantly fewer coefficients than existing methods.
We introduce an audio texture synthesis algorithm based on scattering moments. A scattering transform is computed by iteratively decomposing a signal with complex wavelet filter banks and computing their amplitude envelop. Scattering moments provide general representations of stationary processes computed as expected values of scattering coefficients. They are estimated with low variance estimators from single realizations. Audio signals having prescribed scattering moments are synthesized with a gradient descent algorithms. Audio synthesis examples show that scattering representation provide good synthesis of audio textures with much fewer coefficients than the state of the art.