Is There a Recurrent Neural Network (RNN)?

An output dense neural layer is learned that can translate a random RNN into something that looks like the input pattern. So when you have little grains of this that you feed into a traditional granular synthesizer, rather than having a sample of a small grain, you can generate that grain from an RNN. The leak rate and the specter radius are really interesting. They control kind of the stability and the timbre and pitch range of the net at inference time.