• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.100.2022.tde-06032023-141417
Document
Author
Full name
Milton dos Santos
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2022
Supervisor
Committee
Rodrigues Neto, Camilo (President)
Amaral, Amaury de Souza
Crepaldi, Antonio Fernando
Fernández Tuesta, Esteban
Title in Portuguese
Classificação de áudio musical a partir dos coeficientes da Transformada Wavelet utilizando Redes Neurais Convolucionais
Keywords in Portuguese
Coeficientes Wavelet
MIR
Processamento de Sinais
Rede Neural Convolucional
Transformada Wavelet
Abstract in Portuguese
A identificação do estilo musical a que pertence uma música é uma tarefa relativamente simples para um humano, mesmo com pouco treinamento musical. Entretanto, é uma tarefa bastante difícil de ser realizada de forma automatizada. Neste trabalho utilizamos a Transformada Wavelet, que consegue representar uma música em suas componentes de frequência em função do tempo, gerando uma imagem denominada espectrograma. A partir do espectrograma, geramos imagens para treinar uma Rede Neural Convolucional com o objetivo de classificar os sinais de áudio em seus estilos musicais. Apenas os primeiros 15 segundos de cada música são utilizados para gerar o espectrograma, 6.075 músicas no conjunto de treinamento e 2.025 no conjunto de teste, pertencentes a 10 estilos musicais Blues, Clássico, Country, Disco, Hip Hop, Jazz, Metal, Pop, Reggae e Rock. O procedimento é repetido 10 vezes, com o conjunto de treinamento e teste escolhidos aleatoriamente. A média das taxas de acerto ficou entre 70% e 94%, bem acima dos 10% esperados se a classificação fosse por puro acaso.
Title in English
Classification of musical audio from the coefficients of the Wavelet Transform using Convolutional Neural Networks
Keywords in English
Convolutional Neural Network
MIR
Signal Processing
Wavelet Coefficient
Wavelet Transform
Abstract in English
Identifying the musical style to which a song belongs is a relatively simple for a human, even with little musical training. However, it is a task quite difficult to be performed in an easy way. In this work we use the Wavelet Transform, which manages to represent a song in its frequency as a function of time, generating an image called spectrogram. From grass, we generate images of the behavior spectrum a Convolutional Neural Network with the purpose of classifying audio signals into their musical styles. only the first 15 seconds of each song used to generate the spectrogram, 6,075 songs in training set and 2025 in the test set, belonging to 10 musical styles Blues, Classical, Country, Disco, Hip Hop, Jazz, Metal, Pop, Reggae and Rock. The procedure Repetition 10 times, with the training and test set randomly chosen. One average hit rates were between 70% and 94%, well above the 10\% expected if the classification were by pure chance.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
Publishing Date
2023-07-10
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.