Título: LSTM -Based Speech Analysis for Automatic Classification of Dysarthria Severity Levels
Autores: Guilherme B. F. Santos, Pedro L. S. G. Camara, Diego M. P. F. Silva, Andrea M. N. C. Ribeiro, Sergio M. M. Fernandes & Rodrigo de P. Monteiro
Resumo: Dysarthria is a motor speech disorder that affects the articulation and pronunciation of words, typically caused by damages to the neurological system responsible for speech. Classifying the severity levels of dysarthria is a clinically relevant task that can assist healthcare professionals in determining the most appropriate treatment strategies based on the degree of impairment. This study investigates the ability of Long Short-Term Memory (LSTM) models with Mel Frequency Cepstral Coefficients (MFCC) to classify speech samples of individuals with different levels of dysarthria using the UASpeech and Torgo datasets. Results showed good model performance with consistent accuracy across different levels of dysarthria severity, achieving values above 98% regarding the four severity levels considered in this work.
Palavras-chave: Dysarthria; Severity Classification; Automatic Speech Recognition (ASR); Deep Learning; MFCC.
Páginas: 6
Código DOI: 10.21528/CBIC2025-1188154
Artigo em PDF: CBIC_2025_paper1188154.pdf
Arquivo BibTeX:
CBIC_2025_1188154.bib