RAVE v3 uses Adaptive Instance Normalization to transfer timbre from one audio stream to another

RAVE v3 adds Adaptive Instance Normalization (AdaIN), a technique inspired by StyleGAN, to the reconstruction process. AdaIN normalizes an intermediate feature map using statistics (mean and std) from a ‘source’ signal, then re-scales with statistics from a ‘target’ signal — effectively injecting the target’s timbral envelope. In the nn~ or VST interface, you define source and target styles via the attribute system while feeding the respective audio. The result: the decoder reproduces the latent structure of the input while imposing the spectral envelope of the target — style transfer in a single patch.

Examples

In Max, use the nn~ attribute system to set source and target: play the target audio to learn its style, then feed input audio to hear it re-timbred toward the target.

Assessment

A producer wants a bass guitar to sound like a cello. They have both recordings. Describe the steps in a RAVE v3 Max patch to achieve this timbre transfer.