Language Generalization Using Active Learning in the Context of Parkinson’s Disease Classification

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F23%3A00368598" target="_blank" >RIV/68407700:21230/23:00368598 - isvavai.cz</a>

Result on the web

<a href="https://doi.org/10.1007/978-3-031-40498-6_31" target="_blank" >https://doi.org/10.1007/978-3-031-40498-6_31</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-40498-6_31" target="_blank" >10.1007/978-3-031-40498-6_31</a>

Result language

angličtina

Original language name

Language Generalization Using Active Learning in the Context of Parkinson’s Disease Classification

Original language description

Speech traits have enabled the evaluation and monitoring of the neurological state of different disorders, including Parkinson’s Disease (PD) using classical and deep approaches. Considering that speech contains paralinguistic information, the native language of the speaker influences the performance of the trained models when classifying the presence of the disease. Although researchers have performed several studies using corpora from different acoustic and language conditions, there is no baseline for the accuracy of a system to classify PD in cross-language scenarios. This study evaluates the generalization capability of different classical and deep methods to discriminate between PD patients and healthy speakers. The experiments are performed in cross-language scenarios. In particular, an Active Learning (AL) strategy is considered to evaluate the influence of the training data selection to improve the model’s performance under cross-language settings. The results indicate that models based on Wav2Vec 2.0 yielded the best results in detecting the presence of the disease in such non-controlled cross-language scenarios. In addition, the AL selection outperformed the results compared to a random selection of training samples. The considered AL based-approach allows to achieve high accuracies using a careful selection of training data in an adaptively manner. This is particularly important when dealing with non-annotated and limited data, such as the case of pathological speech modeling.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

20601 - Medical engineering

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2023

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Article name in the collection

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

ISBN

9783031404979

ISSN

0302-9743

e-ISSN

—

Number of pages

11

Pages from-to

349-359

Publisher name

Springer

Place of publication

Basel

Event location

Pilsen

Event date

Sep 4, 2023

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—