SLPOtherJournal of voice : official journal of the Voice Foundation2024

The Acoustic Dissection of Cough: Diving Into Machine Listening-based COVID-19 Analysis and Detection.

Zhao Ren, Yi Chang, Katrin D Bartl-Pokorny and 2 others

PMID 35835648

WHAT IT FOUND

An algorithm detecting COVID-19 from cough sounds beat chance but remained weak, especially for asymptomatic people.

It is not accurate enough for clinical screening. Self-reported symptoms and unverified test status make the data unreliable for deciding patient care.

Key findings

01The best model distinguishing COVID-19 positive from negative individuals achieved an unweighted average recall of 0.632, which is significantly better than chance but far from diagnostic accuracy.

02Performance was lowest for asymptomatic individuals, where the model incorrectly assigned the highest ratio of positive samples to the negative class.

03Participants self-reported their COVID-19 status without providing test confirmation, and coughs were recorded at unspecified times during infection.

STILL TO COME

How it was doneWhat they foundWhat it means for SLPs

Read the rest of this summary

You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.

Already have one?

What it does not show

COVID-19 status was self-reported without test verification. The time between infection or testing and cough recording was not controlled, meaning disease stage varied. Symptom severity was not recorded, only presence. Crowd-sourced audio quality varied, and participants were advised to use plastic bags, which degrades acoustic fidelity. The dataset was heavily imbalanced, with far more negative than positive samples. It is possible that multiple recordings came from the same person or household, inflating sample independence.

Declared interests

The authors declared no conflicts of interest.

The easy way to misread this

Do not interpret the finding that the algorithm beat chance as evidence that cough sounds are a viable clinical diagnostic marker. The accuracy is too low for screening, and the self-reported, unverified nature of the data means the algorithm may be detecting recording habits or symptom severity rather than the virus itself.

Read it on PubMed →