Home Publications Research Grants Inventions & Patents Awards Additional Research Activities Faculties & Institutions Research Areas

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

Moreno-Acevedo SA, Rios-Urrego CD, Vásquez-Correa JC, Rusz J, Nöth E, Orozco-Arroyave JR (2023)

---

Publication Type: Conference contribution

Publication year: 2023

Journal

Publisher: Springer Science and Business Media Deutschland GmbH

Book Volume: 14102 LNAI

Pages Range: 349-359

Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Event location: Pilsen, CZE

ISBN: 9783031404979

DOI: 10.1007/978-3-031-40498-6_31

Abstract

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.

Authors with CRIS profile

Involved external institutions

How to cite

APA:

Moreno-Acevedo, S.A., Rios-Urrego, C.D., Vásquez-Correa, J.C., Rusz, J., Nöth, E., & Orozco-Arroyave, J.R. (2023). Language Generalization Using Active Learning in the Context of Parkinson’s Disease Classification. In Kamil Ekštein, František Pártl, Miloslav Konopík (Eds.), Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (pp. 349-359). Pilsen, CZE: Springer Science and Business Media Deutschland GmbH.

MLA:

Moreno-Acevedo, S. A., et al. "Language Generalization Using Active Learning in the Context of Parkinson’s Disease Classification." Proceedings of the 26th International Conference on Text, Speech and Dialogue, TSD 2023, Pilsen, CZE Ed. Kamil Ekštein, František Pártl, Miloslav Konopík, Springer Science and Business Media Deutschland GmbH, 2023. 349-359.

BibTeX: Download