Spectral-temporal saliency masks and modulation tensorgrams for generalizable COVID-19 detection

Download PDF
Open Webpage

Yi Zhu, Tiago H. Falk

Published: 01 Jan 2024, Last Modified: 07 Oct 2025Comput. Speech Lang. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Modulation tensorgram (MTR) is a robust representation of COVID pathological speech.•Proposed MTR-based model exceeds benchmarks on within- and cross-dataset evaluations.•Our saliency masks filter disease-irrelevant regions and improves generalizability.•Proposed system is light-weight hence a promising candidate for remote monitoring.