Music Graduate Program
Bu topluluk için Kalıcı Uri
Gözat
Yazar "Akyol, Tuğrul Orkun" ile Music Graduate Program'a göz atma
Sayfa başına sonuç
Sıralama Seçenekleri
-
ÖgeAn Application Of Neural Network-based Music Generation Models In The Context Of Modern And Contemporary Music(Institute of Social Sciences, 2019) Akyol, Tuğrul Orkun ; Ekmektsoglou, Emmanouil ; 614975 ; Music ; MüzikThe purpose of this thesis is to test the performance of the artificial neural network- based music generation models in a different musical context and to determine if they can be used as composition assistance tools by the contemporary composer. A brief explanation of the working principles and recent history of ANNs are given before moving onto the conducted experiments. For the purposes of the thesis, a model called Performance RNN, developed by Project Magenta, is trained three times with different datasets and/or hyperparameters. The model is capable of creating performances with feeling: it is sensitive to time intervals as small as eight milliseconds and it can process dynamic information. The datasets are in MIDI format and they are compiled from audio files with the help of another neural network model called Onsets and Frames, which transcribes raw audio files of piano music into MIDI files. For the first experiment a dataset of Messiaen's complete piano works is used, which spans over six hours. Messiaen is chosen because of his relative consistent musical language and large output of solo piano works.