|
Descripción
|
|
|---|---|
| Parkinson's disease is a neurodegenerative disorder characterized by an impairment of the patients' motor skills. Despite an early diagnosis might improve the quality of life of individuals and reduce costs on health systems, to date, there are no fast, objective and reliable detection tests. Some studies have employed speech signals to assess the presence of Parkinson in patients but none has analyzed in an orderly and thorough manner state-of-the-art approaches typically employed in speaker recognition for the detection of the disorder. With this in mind, this work employs two successful speaker recognition modeling strategies: GMM-UBM and iVectors. By using 10 ms frame lengths, accuracies up to 80% are obtained with iVectors and 78% with GMM-UBM. Results indicate that both techniques perform well when automatically detecting the disease. Moreover, they suggest that important information is found in spectral envelope short-time changes as exhibited by the better performance when using shorter frame lengths. | |
|
Internacional
|
No |
|
Entidad
|
XXXI Simposium Nacional de la Unión Científica Internacional de Radio, URSI 2016 Tipo de participación |
|
Lugar
|
Madrid |
|
Páginas
|
|
|
Referencia/URL
|
http://rfcas.eps.uam.es/ursi2016/index.php |
|
Tipo de publicación
|
Ponencia |