During the first phase, Amy visited GU in Lindholmen for a first 2 week period. This included meeting other colleagues, and while exploring Gothenburg during a snowy winter period, to design and work on our foundation for the next phases.

To create a solid foundation we engaged with exiting work the domain of generative music and generative AI approaches by catching up on current papers.

As part of this we discussed what parts of existing work would work well for our project and what specific approaches we want to focus on. As part of this we looked into AudioGen, AudioLM as current approaches using large models.

We also looked into something more lightweight, RAVE, to allow for a more custom design using smaller networks as well as an approach that can be trained without the availablity of large compute resources. An alternative to RAVE, could also be using MiGen~ which allows the creation of synthetic instruments based on physical simulation and we can then use AI approaches to innovate and alter properties of the instruments.