Privacy-Aware Respiratory Symptom Detection

Low-sampling acoustic pipeline to detect coughs and sneezes in the wild with energy-efficient methods

Summary

I built a privacy-preserving, energy-efficient acoustic pipeline to detect coughs and sneezes at low sampling rates (1 kHz), suitable for deployment on resource-constrained devices.

Key contributions

  • Designed feature extraction and decision-tree based classifiers that operate at 1 kHz sampling and preserve privacy by avoiding raw audio uploads.
  • Validated the pipeline in-the-wild and achieved 70% sensitivity for cough detection and 57% sensitivity for sneeze detection.
  • This work received the WellComp Best Paper award at UbiComp/ISWC 2023.

(No public repo available)