Analýza a klasifikace datových souborů GTZAN a SUNO pomocí metod strojového učení v jazyce Python

This bachelor's thesis deals with the comparative analysis of musical data from human-composed tracks and synthetic tracks generated by artificial intelligence. The main objective is to assess the acoustic and stylistic diversity of current generative artificial intelligence (genAI) in comparison to human creation, and to evaluate the ability of machine learning models to distinguish between these two sources. The theoretical part defines musical concepts within the context of Digital Signal Processing and describes the extraction of key acoustic features, such as MFCCs, chroma features, and others. The practical part utilizes the GTZAN and SUNO datasets, upon which structural analysis is performed using principal component analysis (PCA). Subsequently, classification models (logistic regression and random forest) are implemented for music origin detection and genre classification. The results of the thesis identify the differences between the datasets and discuss the outcomes of the analysis as well as the success rate of the machine learning algorithms.