New paper on neural audio synthesis. We combine VAEs with GANs for many-to-many transfer of vocal and instrumental timbre.