Αυτόματη ταξινόμηση μελωδίας σε μουσικά είδη με τη χρήση τεχνητών νευρωνικών δικτύων
Βιδάλη, Ροδάνθη Γ.
In this thesis we study the automatic classification of music into genres using artificial neural networks (ANN). Artificial neural networks consist of simple elements operating in parallel. These elements are called neurons and to some extent simulate the biological neural systems. One can train a neural network so that it simulates a particular function by adjusting the connections (weights) between elements. Typically, neural networks are configured, or trained, so that a particular input leads to a specific desired output. In this work we use various descriptors that characterize a musical sequence in order to train a neural network to be able to distinguish the type of music given input we will test different architectures of artificial neural networks (layers, number of neurons per layer, transfer functions, weight updating process). Each separate type of ANN is provisioned with a few hours of music from different genres in order to be trained to recognize these different genres. For the implementation of the work we use the Matlab Neural Network Toolbox. Through experimental numerical results we verify the applicability of artificial neural networks for automatic classification of melody in musical genres and draw different conclusions for each tested architecture.