| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.66 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A classificaĆ§Ć£o de patologias relacionadas a voz utilizando conceitos de Deep Learning
vem crescendo consideravelmente nos Ćŗltimos anos. Bons resultados jĆ” foram obtidos
para a classificaĆ§Ć£o em fala sustentada com vogais, mas ainda existem poucos trabalhos
relacionadas a classificaĆ§Ć£o deste problema utilizando fala contĆnua. Por isso, Ć© foco desta
dissertaĆ§Ć£o realizar a implementaĆ§Ć£o dos principais modelos de Deep Learning para a classificaĆ§Ć£o
de patologias da voz em fala contĆnua, utilizando a frase alemĆ£ āGuten Morgen,
wie geht es Ihnen?ā da base de dados Saarbruecken Voice Database. SĆ£o utilizados as patologias
de disfonia, laringite e paralisia das cordas vocais, alƩm da classe dos saudƔveis,
para anƔlises multi classe e binƔria. AlƩm disso, tambƩm Ʃ realizado um estudo prƩvio
para a classificaĆ§Ć£o com vogais nas mesmas patologias. O melhor resultado para as vogais
Ć© de 99% de exatidĆ£o para a implementaĆ§Ć£o de um modelo LSTM com parĆ¢metros Jitter,
Shimmer e AutocorrelaĆ§Ć£o, na classificaĆ§Ć£o binĆ”ria entre laringite e saudĆ”vel. Para as
frases, Ć© realizado um estudo comparativo entre modelos de redes neuronais, convolucionais
e recorrentes para os parĆ¢metros MFCCs e Espectrogramas na escala Mel obtendo
resultados de 76% de medida-F para disfonia x saudƔvel, 68% de medida-F para laringite
x saudĆ”vel, 80% de medida-F para paralisia x saudĆ”vel. Para classificaĆ§Ć£o multi classe Ć©
obtido 59% e 40% de medida-F para 3 classes e 4 classes, respectivamente.
The classification of voice related pathologies using Deep Learning concepts has been increasing considerably in recent years. Good results have already been obtained for classification in sustained speech with vowels, but there are still few studies related to the classification of this problem using continuous speech. Therefore, the focus of this dissertation is to implement the main models of Deep Learning for the classification of voice pathologies in continuous speech, using the German phrase "Guten Morgen, wie geht es Ihnen?"From the Saarbruecken Voice Database. The pathologies of dysphonia, laryngitis and paralysis of the vocal cords, as well as the healthy class, are used for multi-class and binary analyzes. In addition, a previous study for the classification with vowels in the same pathologies is also carried out. The best result for the vowels is 99 % accuracy for the implementation of an LSTM model with parameters Jitter, Shimmer and Autocorrelation, in the binary classification between laryngitis and healthy. For the phrases, a comparative study between neural networks, convolutional and recurrent models with the parameter MFCCs and Spectrograms in the Mel scale, obtaining results of 76% F-measure for dysphonia x healthy, 68% F-measure for laryngitis x healthy, 80% F-measure for healthy x paralysis of the vocal cords. For multi-class classification is obtained 59% and 40% of F-measure for 3 classes and 4 classes, respectively.
The classification of voice related pathologies using Deep Learning concepts has been increasing considerably in recent years. Good results have already been obtained for classification in sustained speech with vowels, but there are still few studies related to the classification of this problem using continuous speech. Therefore, the focus of this dissertation is to implement the main models of Deep Learning for the classification of voice pathologies in continuous speech, using the German phrase "Guten Morgen, wie geht es Ihnen?"From the Saarbruecken Voice Database. The pathologies of dysphonia, laryngitis and paralysis of the vocal cords, as well as the healthy class, are used for multi-class and binary analyzes. In addition, a previous study for the classification with vowels in the same pathologies is also carried out. The best result for the vowels is 99 % accuracy for the implementation of an LSTM model with parameters Jitter, Shimmer and Autocorrelation, in the binary classification between laryngitis and healthy. For the phrases, a comparative study between neural networks, convolutional and recurrent models with the parameter MFCCs and Spectrograms in the Mel scale, obtaining results of 76% F-measure for dysphonia x healthy, 68% F-measure for laryngitis x healthy, 80% F-measure for healthy x paralysis of the vocal cords. For multi-class classification is obtained 59% and 40% of F-measure for 3 classes and 4 classes, respectively.
Description
Dupla diplomaƧaĆ£o com a UTFPR - Universidade TecnolĆ³gica Federal do ParanĆ”
Keywords
Long short-term memory Rede neuronais convolucional Redes neuronais artificias Transfer learning