Gwangju Institute of Science and Technology advance voice pathology

Researchers from the Gwangju Institute of Science and Technology (GIST) in South Korea have created a technique known as A-TAPT to screen vocal vibrations for diseases.

Voice pathology is the term used to describe aberrant diseases that result in abnormal vibrations in the vocal cords (or vocal folds), such as dysphonia, paralysis, cysts, and even malignancy. Within this framework, vocal pathology detection (VPD) has drawn a lot of interest as a non-invasive method for automatically identifying voice abnormalities. To obtain good VPD performance, machine learning techniques like convolutional neural networks (CNN) and support vector machines (SVM) have been effectively applied as pathological voice detection modules.

However, because the VPD task is a different domain from conversation speech, fine-tuning these models for VPD results in an over fitting issue. Because of this, generalization is hindered by the pretrained model's excessive attention to the training set and poor performance on fresh data.

In order to address this issue, a group of researchers from the Gwangju Institute of Science and Technology (GIST) in South Korea, under the direction of Prof. Hong Kook Kim, have put forth a novel technique known as adversarial task adaptive pretraining (A-TAPT) in conjunction with Wave2Vec 2.0, a self-supervised pretrained model for speech signals, as part of a contrastive learning method. Here, they included adversarial regularization in the process of continuous learning.

Regarding the work's long-term effects, the article's first author, Mr. Park, noted that it may enable early and accurate detection of voice-related diseases, which could result in more successful therapies and enhance the lives of countless people.