Using machine learning for determining accompanying chords from main melody

Z uporabo strojnega učenja se napovedujejo spremljevalni akordi za spremljavo osnovne melodije. Na zbirki skladb skupine The Beatles je primerjana uspešnost napovedovanja akordov z izbranimi algoritmi strojnega učenja. Vhodni podatki so datoteke MIDI. Zgradi se več različic učnih primerov, ki na različne načine opisujejo skladbo. Zgradi se nova datoteka MIDI, ki omogoča poslušanje izvorne melodije skladbe z napovedanimi akordi sistema. Za izbrano skladbo uspešno napovemo spremljevalne akorde, za splošno uporabo bi opisana metoda potrebovala izboljšave.This work uses machine learning for determining accompanying chords from main melody. The results of machine learning algorithms trained on music of The Beatles are compared. The music iz given in MIDI format. Different versions of training examples are used to build models and the results are compared. A MIDI file containg the melody from original song with the predicted accompanying chords as predicted by the developed system is built. The predicted accompanying chords for the selected song are satisfying. The method used in this work could be used for building an end-user application, although it would need further optimization