Is There Any Additional Information in a Neural Network Trained for Pathological Speech Classification?
Rios-Urrego CD, Vásquez-Correa JC, Orozco-Arroyave JR, Nöth E (2021)
Publication Type: Conference contribution
Publication year: 2021
Journal
Publisher: Springer Science and Business Media Deutschland GmbH
Book Volume: 12848 LNAI
Pages Range: 435-447
Conference Proceedings Title: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Event location: Olomouc
ISBN: 9783030835262
DOI: 10.1007/978-3-030-83527-9_37
Abstract
Speech is a biomarker extensively explored by the scientific community for different health-care applications because its reduced cost and non-intrusiveness. Specifically, in Parkinson’s disease, speech signals and deep learning methods have been explored for the automatic assessment and monitoring of patients. Related studies have shown to be very accurate to discriminate pathological vs. healthy speech. In spite of the high accuracies observed to detect the presence of diseases from speech, it is not clear which additional information about the speakers or the environment is implicitly learned by the deep learning systems. This study proposes a methodology to evaluate intermediate representations of a neural network in order to find out which other speaker traits and aspects are learned by the system during the training process. We trained models to detect the presence of Parkinson’s disease from speech. Then, we used intermediate representations of the network to classify additional speaker traits such as gender, age, and the native language. It is important to detect which information is available inside the neural network that can lead to open the black-box and to detect possible algorithmic biases. The results indicate that the network, in addition to adjusting its parameters for disease classification, also acquires knowledge about gender of the speakers in the first layers, and about speech tasks and the native language in the last layers of the network.
Authors with CRIS profile
Involved external institutions
Colombia (CO)How to cite
APA:
Rios-Urrego, C.D., Vásquez-Correa, J.C., Orozco-Arroyave, J.R., & Nöth, E. (2021). Is There Any Additional Information in a Neural Network Trained for Pathological Speech 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. 435-447). Olomouc, CZ: Springer Science and Business Media Deutschland GmbH.
MLA:
Rios-Urrego, C. D., et al. "Is There Any Additional Information in a Neural Network Trained for Pathological Speech Classification?" Proceedings of the 24th International Conference on Text, Speech, and Dialogue, TSD 2021, Olomouc Ed. Kamil Ekštein, František Pártl, Miloslav Konopík, Springer Science and Business Media Deutschland GmbH, 2021. 435-447.
BibTeX: Download