Analysis of cough sounds for identifying COVID-19 cases automatically

A team of researchers has submitted a system to the Diagnosing COVID-19 using Acoustics (DiCOVA) Challenge at the Interspeech 2021 international congress. The system, developed with funding from the EU's Horizon 2020 research and innovation programme and the AGAUR, Catalan Government, aims to use Artificial Intelligence (AI) techniques to detect the disease.

The research, led by Adrià Mallol and Helena Cuesta, with the participation of Emilia Gómez, Björn Schuller, and others, focuses on whether AI can detect disorders related to COVID-19 through automatic cough analysis. The study, titled "Cough-based COVID-19 Detection with Contextual Attention Convolutional Neural Networks and Gender Information," was presented at the Interspeech 2021 conference and is published in the Proceedings of Interspeech.

The research investigates two different neural network architectures, both with a common structure, for classifying whether features correspond to a patient testing positive for COVID-19 or a healthy patient. The models use the spectrogram, a time-frequency representation of the audio signal, as input.

The authors of the study hypothesize that men's and women's coughs may have different features due to differences in vocal tracts. They found that the patient's gender is an important factor in their research. Experiments suggest that models that incorporate information on the patient's gender obtain better results in their predictions.

The work is a first approach to detecting COVID-19 via automatic cough analysis and offers clues for future research, such as understanding how the cough signal is altered in COVID-19-positive patients. Previous AI systems have proven effective at detecting coughing, sneezing, and respiratory anomalies.

The research comes at a time when the COVID-19 crisis has been testing healthcare systems worldwide. Mass population screening is being conducted to detect positive COVID-19 cases, and access to vaccines has been making the situation more stable.

The study uses a database called Coswara dataset provided by the DiCOVA Challenge, containing 1,040 audio recordings of coughing with associated metadata, including gender and COVID-19 status. The authors of the paper are Yi Xu, Chenyu Yang, and Dong Yu.

The research offers a promising step towards using AI to detect COVID-19, potentially aiding healthcare professionals in diagnosing the disease more efficiently. As the world continues to grapple with the pandemic, research like this could play a crucial role in combating the spread of the virus.