|
Descripción
|
|
|---|---|
| The automatic detection of Parkinson?s Disease (PD) from voice and speech signals has proved to achieve high classification rates specially for patients in medium or advanced state of the disease. Notwithstanding, in the medical practice, specialists use the UPDRS score to recall information on the course of the condition, evaluate treatments and/or assess the patients? quality of life. The prediction of the UPDRS score from voice/speech signals is a more complex task that, unlike detection, has achieved poor results, especially because it requires more complex machine learning models, and most of the databases used in the state of the art include less than 60 patients. This work will explore the use of a transfer learning approach in order to train a DNN model for the prediction of the UPDRS score. In the first step, the network will be trained using voice signals from patients with different organic and functional pathologies; the network architecture will be based on a recent purpose for voice quality assessment according to the GRB scale, that includes the combination of feature learning and feature engineering approaches using a multi-modal strategy. The learning weights of the first layers will be transferred to a second DNN designed to predict the UPDRS score, and the fine tuning of the last layers in the second network will be carried out using two databases of PD patients. The results will give new insights about the potential of transfer learning strategies for the automatic assessment of pathological voices, specially the neurogenic symptoms associated with the hypokinetic dysarthria of PD patients and its correlation with the UPDRS score. | |
|
Internacional
|
Si |
|
Nombre congreso
|
1st Automatic Assessment of Parkinsonian Speech Workshop, 20-21, September, Cambridge, MA, USA, 20-21, September |
|
Tipo de participación
|
960 |
|
Lugar del congreso
|
Cambridge, Masachussetts, EEUU |
|
Revisores
|
Si |
|
ISBN o ISSN
|
|
|
DOI
|
|
|
Fecha inicio congreso
|
20/09/2019 |
|
Fecha fin congreso
|
21/09/2019 |
|
Desde la página
|
4 |
|
Hasta la página
|
4 |
|
Título de las actas
|
1st Automatic Assessment of Parkinsonian Speech Workshop |