Machine Learning for Songwriting

Ovaj diplomski rad ukratko objašnjava osnove strojnog učenja i neuronskih mreža, te pokriva elemente strojnog učenja potrebnih za ovaj rad. U sklopu rada razvijen je program čija je zadaća naučiti stil pisanja pjesama nekog autora, te na temelju toga generirati novu pjesmu. Pokazani su rezultati traženja optimalnih parametara, rezultati pri treniranju, te primjer konačno generirane pjesme. Također se provodi usporedba s prijašnjim radom. Zaključuje se da program može uspješno naučiti specifičan stil pisanja pjesama, nakon pronalaženja optimalnih parametara.The master's thesis briefly explains the basics of machine learning and neural networks, as well as other elements used in the thesis. As a part of the thesis, a program has been developed that has a task of learning the art style of a specific author, and then generate a song. Results of searching for the optimal parameters have been displayed, as well as training results, and, finally, an example of a generated song. A comparison is also made with the author's earlier work. It is concluded that the program can successfully learn to copy the art style, after it finds the optimal parameters