IRCAM Tutorials: Rave and nn~
RAVE is a signal generator, which is both high-quality (48kHz generation) and blazingly-fast (25x realtime on CPU). Compared to previous (DDSP) approaches, we can generate any type of sounds (polyphonic, noises), and we can also greatly reduce the number of control dimensions. The drawback is that the dimensions themselves are not necessarily understandable. Therefore, we can use it for timbre transfer train another (prior) model to generate audio automatically.
More info about RAVE or nn~