Deep Learning Anomaly Detection methods to passively detect COVID-19 from Audio

Murthy, Shreesha Narasimha

Etd

Deep Learning Anomaly Detection methods to passively detect COVID-19 from Audio

Public

The world has been severely affected by COVID-19, an infectious disease caused by the SARS-Cov-2 coronavirus. COVID-19 incubates in a patient for 7 days before symptoms manifest. During this incubation period, affected individuals, unknowingly, transmit the virus through respiratory droplets released when the individual coughs or sneezes, which has resulted in a record number of daily cases around the world. The identification of the presence of COVID-19 is challenging as its symptoms are similar to influenza symptoms such as cough, cold, runny nose and chills. COVID-19 affects human speech sub-systems involved in respiration, phonation, and articulation. This master thesis proposes a deep anomaly detection framework for passive, speech-based detection of COVID-related anomalies in voice samples of COVID-19 affected individuals. The low percentage of positive cases and extreme imbalance in available COVID audio datasets present a challenge to machine learning classifiers but creates an opportunity to utilize anomaly detection techniques. This thesis investigates COVID detection from audio using various types of deep anomaly detectors and autoencoders. Contrastive loss methods are also explored to force our models to learn the discrepancies between COVID and non-COVID cough data representations. In rigorous evaluation, the variational autoencoder with the elliptic envelope as the anomaly detector analyzing Mel Filterbanks audio representations performed best with an AUC of 65.7, outperforming the state of the art.

Creator