Detection of Covid via sound of cough by machine-learning with 98.5% accuracy (www.nature.com)

Researchers report that a machine-learning system can identify COVID-19 infection from the sound of a cough with 98.5% accuracy. The work uses audio recordings of coughs as input to an ML pipeline that extracts acoustic features (for example, spectrogram-based representations and MFCCs) and trains a classifier—typically a convolutional or other deep network—to distinguish COVID-positive from COVID-negative coughs. The high reported accuracy suggests cough-derived biomarkers may carry robust signatures of the disease and that automated acoustic screening could serve as a rapid, low-cost triage tool for remote or resource-limited settings. Technically, this approach hinges on careful audio preprocessing, feature engineering or end-to-end spectrogram learning, and rigorous cross-validation to avoid overfitting to recording conditions or demographics. Important implications include potential integration into telehealth apps and population-level screening, but real-world deployment requires larger, geographically and demographically diverse datasets, prospective validation, and controls for confounders (background noise, comorbidities, device variability, and viral variants). Privacy, regulatory approval, and clinical performance (sensitivity/specificity trade-offs) remain key hurdles before such systems can be trusted as diagnostic or public-health tools.