Analyzing emotions in Turkish music with deep learning algorithms

The effects of music on human emotions have long been an intriguing and significant area of research. With the advancement of computer systems, the analysis of emotions in music has become an exciting research field. In particular, the rapid progress in machine learning and deep learning, along with the increasing computational power of modern computers, have accelerated studies in this area. As a result, data and models derived from such research have been widely adopted by many content providers to deliver richer and more personalized experiences to listeners. In this thesis, a study on the analysis of emotions in Turkish music is presented, aiming to train models capable of emotion recognition in music using acoustic features extracted from musical pieces. For this purpose, a dataset consisting of 1,324 music tracks annotated by listeners was constructed. From this dataset, 76 different acoustic features were extracted, and 10,560 spectrogram images were generated for use in Convolutional Neural Network (CNN) models. Various machine learning and deep learning techniques were applied to these datasets to develop models, whose performances were analyzed and compared. The primary objective of this thesis is to investigate and evaluate the applicability of deep learning algorithms in emotion recognition within Turkish music.