Date(s) - 19/10/2018
14 h 00 min - 15 h 00 min
Catégories Pas de Catégories
We introduce a Scattering Inverse Network (SIN) to generate univariate time-series.
The SIN is similar to a deep convolutional autoencoder. However, the encoder is not learned, but computed with a scattering transform defined from prior information on sparse time-frequency properties of time-series. In turn, the generator is trained by solving an inverse problem in an adapted metric. It has a similar causal architecture as a WaveNet and provides a simpler mathematical model related to time-frequency decompositions.
Numerical experiments demonstrate that this SIN generates realistic musical and speech signals. It is able to transform low-level musical attributes such as pitch with a linear transformation in the embedding space of scattering coefficients.