Detalhes bibliográficos
Ano de defesa: |
2008 |
Autor(a) principal: |
Corrêa, Débora Cristina |
Orientador(a): |
Saito, José Hiroki
 |
Banca de defesa: |
Não Informado pela instituição |
Tipo de documento: |
Dissertação
|
Tipo de acesso: |
Acesso aberto |
Idioma: |
por |
Instituição de defesa: |
Universidade Federal de São Carlos
|
Programa de Pós-Graduação: |
Programa de Pós-Graduação em Ciência da Computação - PPGCC
|
Departamento: |
Não Informado pela instituição
|
País: |
BR
|
Palavras-chave em Português: |
|
Área do conhecimento CNPq: |
|
Link de acesso: |
https://repositorio.ufscar.br/handle/20.500.14289/379
|
Resumo: |
Several research studies have been realized in order to achieve a musical composition computational system that could, as much as possible, catch the human mind, skills, and creativity. More recently, artificial neural networks (ANNs), also have been deployed as auxiliary models for musical compositions. For musical computation, connectionist systems, as well as other systems that involve machine learning, are able to learn patterns and features available in the melodies of the training set and to generalize them to compose new melodies. Therefore, the use of neural networks in music learning and composition has attracted researchers and many approaches have been developed. The aim of this study is the proposal of a neural network based system for computer-aided musical composition. This system can be divided into four main processes: training, composition, evaluation and optimization. It is also proposed to complement the training and composition processes with a kind of inspiration, from Nature, using landscapes contours as additional information to the network. The neural networks used in the system are: BPTT (Back-Propagation Through Time) and LSTM (Long-Short Term Memory) networks. The results obtained are compared from both networks and it is observed that the LSTM network performs better. It is also proposed an approach that consists of optimizing the weight initialization process of the LSTM network in addition to an estimative of the ideal configuration of the hidden layer, that contributes to the obtained results. |