Open respiratory-sound research
Listen closer.
Detect earlier.
A machine-learning toolkit for identifying COPD-related respiratory sounds — designed to make clinical audio more legible, one breath at a time.
Breath carries
information.
COPD changes the way air moves through the lungs. Those changes leave traces in sound — wheezes, crackles, and quiet moments that can be difficult to hear consistently.
SHWASA turns respiratory recordings into structured signals that researchers can inspect, compare, and improve.
Audio waveform
Capture breath cycles as a time-domain signal before any transformation.
Research snapshot Dual-model study
Built to be
questioned.
Two complementary tasks, reported separately: binary COPD screening and four-class sound-event recognition.
held-out accuracy
ICBHI score
sample rate
What does
this breath say?
Choose a sample to see how the prototype translates sound into a readable model output.
Need more control? Open the full Playground →Upload, listen, compare, and inspect telemetry.This is a research demonstration, not a medical device or clinical diagnosis.
Respiratory sound classifier
Model output is illustrative · See methodology