| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.53 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A classificaĆ§Ć£o automĆ”tica de sons urbanos Ć© importante para o monitoramento ambiental. Este
trabalho apresenta uma nova metodologia para classificar sons urbanos, que se baseia na
descoberta de padrƵes frequentes (motifs) nos sinais sonoros e utiliza-los como atributos para
a classificaĆ§Ć£o. Para extrair os motifs Ć© utilizado um mĆ©todo de descoberta multi-resoluĆ§Ć£o
baseada em SAX. Para a classificaĆ§Ć£o sĆ£o usadas Ć”rvores de decisĆ£o e SVMs. Esta nova
metodologia Ć© comparada com outra bastante utilizada baseada em MFCC. Para a realizaĆ§Ć£o
de experiĆŖncias foi utilizado o dataset UrbanSound disponĆvel publicamente.
Realizadas as experiĆŖncias, foi possĆvel concluir que os atributos motif sĆ£o melhores que os
MFCC a discriminar sons com timbres semelhantes e que os melhores resultados sĆ£o
conseguidos com ambos os tipos de atributos combinados.
Neste trabalho foi tambĆ©m desenvolvida uma aplicaĆ§Ć£o mĆ³vel para Android que permite
utilizar os mĆ©todos de classificaĆ§Ć£o desenvolvidos num contexto de vida real e expandir o
dataset.
The automatic classification of urban sounds is important for environmental monitoring. This work presents a new method to classify urban sounds based on frequent patterns (motifs) in the audio signals and using them as classification attributes. To extract the motifs, a multiresolution discovery based on SAX is used. For the classification itself, decision trees and SVMs are used. This new method is compared with another largely used based on MFCCs. For the experiments, the publicly available UrbanSound dataset was used. After the experiments, it was concluded that motif attributes are better to discriminate sounds with similar timbre and better results are achieved with both attribute types combined. In this work was also developed a mobile application for Android which allows the use of the developed classifications methods in a real life context and to expand the dataset.
The automatic classification of urban sounds is important for environmental monitoring. This work presents a new method to classify urban sounds based on frequent patterns (motifs) in the audio signals and using them as classification attributes. To extract the motifs, a multiresolution discovery based on SAX is used. For the classification itself, decision trees and SVMs are used. This new method is compared with another largely used based on MFCCs. For the experiments, the publicly available UrbanSound dataset was used. After the experiments, it was concluded that motif attributes are better to discriminate sounds with similar timbre and better results are achieved with both attribute types combined. In this work was also developed a mobile application for Android which allows the use of the developed classifications methods in a real life context and to expand the dataset.
Description
Keywords
ClassificaĆ§Ć£o de sons urbanos MineraĆ§Ć£o de dados Motifs MFCC Urban Sound Classification Data mining