Music synthesis using artificial intelligence

Rad opisuje problem sinteze glazbe umjetnom inteligencijom. Definira sintezu glazbe kao optimizacijski problem i predlaže algoritam temeljen na umjetnim neuronskim mrežama. Ukratko, kroz primjere, opisana je korištena biblioteka za pisanje programske potpore. Bira se skup podataka nad kojim ´ce mreža uˇciti i opisuje se postupak pretprocesiranja. Predlaže se arhitektura neuronske mreže. Definira se proces uˇcenja i evaluacije modela. Generirane skladbe se analiziraju i uspored¯uju sa skladbama iz skupa podataka za uˇcenje.This thesis describes the problem of music synthesis using artificial intelligence. It defines music synthesis as an optimization problem and offers an algorithm based on artificial neural networks. Briefly, through examples, the library used for writing software is described. The data set through which the network learns is selected and data pre-processing is described. The neural network architecture is proposed and implemented. The process of model learning and evaluating is defined. Generated compositions are analyzed and compared with the compositions of the training data set