UM Data Science Research Seminar

Schedule

LECTURE: 12:00 - 12:45 Q&A: 12:45 - 13:00

Speakers: Yuyang Yan and Mikołaj Najda

Subject: Speech analysis for the management of chronic respiratory diseases

Abstract: 
Chronic respiratory diseases are one of the leading causes of morbidity worldwide. Multiple efforts are being made to develop tools that better manage or reduce the burden of these conditions. Novel techniques focus on vocal biomarkers since they offer a promising, noninvasive approach to predicting, diagnosing, and monitoring these respiratory conditions.

In this talk, we introduce you to the DACIL project. DACIL aims at developing a digital COPD companion for Improving Lifestyle (DACIL). It investigates speech analysis as a potential tool for managing patients with COPD by providing voice signal exploration through time and frequency representations enhanced by artificial intelligence utilities. 

In our works, we leverage voice in developing predictive AI models. We show how we explore using the Mel Frequency Cepstral Coefficients (MFCC) to detect COPD deterioration. However, MFCC extraction often relies on default parameter settings which have not been thoroughly investigated for COPD or other respiratory conditions in general. Using several voice datasets, we demonstrate how we optimize MFCC parameters and leverage various voice features to build AI models that can predict respiratory conditions or their deterioration.