Optimization of HMM with interactive evolutionary computation for composing system

This paper presents the interactive composing system, i-Sonneteer. This system optimized HMM into the model of favorite with interactive evolutionary computation. The user rates and improves a melody with GUI. HMM learns 7 melodies which are improved before by the user, and generates the melody based on the learned data. If HMM learns the melodies that can be selected well, the HMM generates the pleasant melody. In interactive evolutionary computation, an individual is HMM, and a gene is an improved melody. When HMM is not mutated, the HMM generates the melody that has the highest probabilities in oneself. When HMM is mutated, the HMM generates the melody with the randomwalk or the older melodies learning method. This effects that the user´s favorite is improved more clear. I-Sonneteer does not only optimize the HMM but also makes the favorite more clear. If the favorite is defined, the user can compose the pleasant melody. So we guess the definition of the favorite is more important than the optimization of the HMM. We experiment in comparing the randomwalk with the older melodies learning method in the mutation.