Cibo v2, by Jeremy Stewart and Shawn Lawson represents an entirely new architecture from the previous Cibo live-coding agent. Cibo v2 performs TidalCycles code in a live-coding setting for musical and sound performance. Cibo v2 is constructed with autoencoder and variational autoencoder architectures as its foundations, with additional neural network modules governing performance progression and variable production. Cibo v2 trains faster and is trained in a number of steps, allowing for a greater degree of flexibility during the development process. The resulting performance agent produces TidalCycles code that is highly reminiscent of the provided training material, while offering a unique, non-human interpretation of TidalCycles performances.
Jump to 1 hour 34 min in video.
Special thanks to Mike Hodnick and Ben Gold for their recorded performances.