A carregar...
Publicação
Classificador musical com notação ABC : abordagem com ANN
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 814.1 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O objetivo principal deste trabalho foi desenvolver um classificador de géneros musicais utilizando a notação ABC, uma representação textual de música, em vez das abordagens tradicionais envolvendo ficheiros de áudio, que são mais complexos e exigem maior capacidade de processamento computacional. A ideia foi explorar uma alternativa mais eficiente para a categorização de músicas, mantendo, ao mesmo tempo, um bom nível de precisão.
Para isso, foram aplicadas redes neuronais artificiais e testadas diferentes abordagens para lidar com o desequilíbrio entre as classes do conjunto de dados, como o uso de pesos para as classes (class weights) e undersampling. O Nottingham Dataset, composto por músicas folclóricas representadas em notação ABC, foi utilizado para treinar e testar o modelo.
Os resultados mostraram que o modelo conseguiu aprender e distinguir bem as classes, e que o uso de pesos para compensar as classes minoritárias levou, geralmente, a uma melhoria na classificação dessas categorias, embora com uma ligeira perda de performance nas classes mais representadas. No entanto, a pequena dimensão e diversidade do conjunto de dados limitam as conclusões a retirar, não garantindo um bom comportamento do modelo num conjunto de dados mais extenso e complexo.
O principal objetivo do estudo foi alcançado: foi possível desenvolver um classificador eficiente, usando a notação ABC. Contudo, a integração do classificador em sistemas de recomendação de música e a extensão do conjunto de dados para incluir mais géneros musicais são desafios que poderão ser interessantes como estudos futuros.
The main aim of this work was to develop a music genre classifier using ABC notation, a textual representation of music, instead of traditional approaches involving audio files, which are more complex and require greater computer processing capacity. The idea was to explore a more efficient alternative for categorizing music while maintaining a good level of accuracy. To do this, artificial neural networks were applied and different approaches were tested to deal with the imbalance between the classes in the dataset, such as the use of class weights and undersampling. The Nottingham Dataset, consisting of folk songs represented in ABC notation, was used to train and test the model. The results showed that the model was able to learn and distinguish the classes well, and that the use of weights to compensate for minority classes generally led to an improvement in the classification of these categories, although with a slight loss of performance in the most represented classes. However, the small size and diversity of the dataset limits the conclusions to be drawn, and does not guarantee good behaviour of the model in a more extensive and complex dataset. The main objective of the study was achieved: it was possible to develop an efficient classifier using the ABC notation. However, integrating the classifier into music recommendation systems and extending the dataset to include more musical genres are challenges that could be interesting as future studies.
The main aim of this work was to develop a music genre classifier using ABC notation, a textual representation of music, instead of traditional approaches involving audio files, which are more complex and require greater computer processing capacity. The idea was to explore a more efficient alternative for categorizing music while maintaining a good level of accuracy. To do this, artificial neural networks were applied and different approaches were tested to deal with the imbalance between the classes in the dataset, such as the use of class weights and undersampling. The Nottingham Dataset, consisting of folk songs represented in ABC notation, was used to train and test the model. The results showed that the model was able to learn and distinguish the classes well, and that the use of weights to compensate for minority classes generally led to an improvement in the classification of these categories, although with a slight loss of performance in the most represented classes. However, the small size and diversity of the dataset limits the conclusions to be drawn, and does not guarantee good behaviour of the model in a more extensive and complex dataset. The main objective of the study was achieved: it was possible to develop an efficient classifier using the ABC notation. However, integrating the classifier into music recommendation systems and extending the dataset to include more musical genres are challenges that could be interesting as future studies.
Descrição
Mestrado Bolonha em Métodos Quantitativos para a Decisão Económica e Empresarial
Palavras-chave
Classificação Musical Notação ABC Redes Neuronais Machine Learning em Música Musical Classification ABC Notation Neural Networks Machine Learning in Music
Contexto Educativo
Citação
Rosendo, Miguel Moura da Silva (2024). “Classificador musical com notação ABC : abordagem com ANN”. Dissertação de Mestrado. Universidade de Lisboa. Instituto Superior de Economia e Gestão
Editora
Instituto Superior de Economia e Gestão