Interior_Banner_Events

Using Machine Learning and Audio Spectral Features for COVID-19 Testing

Michael Esposito (REU Student, SOLS), Sunil Rao (Graduate ECEE Student), Vivek Narayanaswamy (Graduate ECEE Student), and Andreas Spanias (Professor and Center Director, ECEE, SenSIP Center), 650 E Tyler Mall, Tempe, AZ 85281

As the COVID-19 pandemic continues, it has become necessary to develop rapid testing to identify infected individuals in order to restrict the spread of the virus. In this REU project, we use audio waveform signatures of breathing and coughing to determine whether COVID-19 can be diagnosed. More specifically, we determine audio waveform features and use machine learning algorithms to develop diagnostics for COVID-19 testing from breathing and coughing patterns. The non-invasive rapid and remote testing benefits of this approach relative to existing nose swab, saliva, and blood testing make this method very attractive for deployment on smart phones. Challenges include possible distorted or low-quality audio samples, availability of reliable labeled (ground truth) data, and lack of baseline (healthy) audio recordings for comparison to suspected pathological cases. Currently, we are using the log Mel spectrogram as the singular feature for classification, but we will expand the feature set to include additional statistical and perceptual parameters. We are exploring various machine learning algorithms including convolutional neural networks (CNNs), convolutional recurrent neural networks (CRNNs), and graph convolutional recurrent neural networks (GCRNNs). These algorithms have been implemented in python and comparative results in terms of performance and computational complexity have been obtained. Our preliminary COVID-19 detection results will be presented at the conference.
 
Project funded in part by NSF REU award 1659871 and the SenSIP center and I/UCRC award 1540040.




Additional Abstract Information

Presenter: Michael Esposito

Institution: Arizona State University Main

Type: Poster

Subject: Electrical & Computer Engineering

Status: Approved


Time and Location

Session: Poster 6
Date/Time: Tue 2:00pm-3:00pm
Session Number: 4535