PAUL-2: An Upgraded Transformer-Based Redesign of the Algorithmic Composer PAUL.

In this paper, we introduce , an algorithmic composer for two-track piano pieces of specifiable difficulty levels, as a ground-up redesign of its predecessor system . While was designed using a long short-term memory neural network, along with a sequence-to-sequence network, is based on the state-of-the-art transformer architecture and makes use of relative attention. A shortcoming of was that it generated unsatisfying accompanying tracks and allowed for only few difficulty levels. overcomes these limitations and theoretically supports an arbitrary number of difficulty classes due to the fact that it utilises an additional encoder for handling difficulty information. We also carried out a medium-scale survey which showed that the output of PAUL-2 was evaluated quite favourably by the participants.