COVID-19 activity screening by a smart-data-driven multi-band voice analysis.
Gabriel Silva, Patrícia Batista, Pedro Miguel Rodrigues
PMID 36464573WHAT IT FOUND
A voice analysis tool using cough, breathing, and speech sounds correctly distinguished COVID-19 patients from healthy controls in 98.46% of cases.
It also separated different disease stages, though distinguishing recovered patients from healthy controls was less accurate at 88.57%.
Key findings
01The algorithm achieved 98.46% accuracy in distinguishing all stages of COVID-19 patients from healthy controls.
02Accuracy dropped to 88.57% when comparing recovered patients to healthy controls, suggesting voice changes persist after infection.
03The tool struggled to differentiate between mild and moderate COVID-19 cases, achieving 86.67% accuracy due to similar symptom profiles.
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.
What it does not show
The study used a public, crowd-sourced dataset, meaning audio quality and recording environments varied widely among participants. The majority of participants were from India (92.3%), which may limit the generalizability of the findings to other populations. The analysis is retrospective and computational; it does not involve clinical trial data or prospective validation in a healthcare setting. The model's lower accuracy in distinguishing recovered patients from healthy controls suggests voice changes may persist, but the study does not clarify if this is due to long-term effects or residual infection.
Declared interests
The authors declare no conflicts of interest. The study used the public Coswara database.
The easy way to misread this
Do not interpret these accuracy rates as evidence that voice analysis can replace standard diagnostic testing. The high accuracy was achieved in a controlled computational environment using a specific dataset, not in a clinical screening setting with variable patient backgrounds.